But there’s a second door now, and almost nobody in India is using it, and the strange part is that it pays you to audition.

You solve a real problem in a real company’s codebase. Your pull request gets merged. You get paid within a few days. And the company now has proof of your work that no proposal could ever give them.

Here’s what I’m breaking down:

• The channel where you get paid to interview, and why the payment is the least valuable part

• What a merged pull request does that a five paragraph proposal never will

• Four platforms, what each actually pays, and the one I’d start on this week

• Why “I don’t have open source contributions” is not the blocker you think it is

• The payment rail detail that quietly makes this better for Indian developers than most freelance sites

• Realistic money: the actual bounty range in 2026, and the outlier that skews everyone’s expectations

• The platform that held people’s money and died, and what it should teach you about where you park your effort

• How to read an issue and know within ten minutes whether you can win it

• What you say to the company after the merge, which is the entire point

“Isn’t this just Upwork with extra steps?”

The opposite, and this is the bit worth slowing down on.

On Upwork you write about your ability. You describe past work, you attach a portfolio, you price yourself against a Pakistani dev charging half and an American charging four times, and a stranger who has never seen your code decides from a text box. The proposal is the product.

On a bounty platform there is no proposal. There’s an issue, in a real repository, with money attached to it. You read the code, you fix the thing, you open a pull request. It merges or it doesn’t.

Your merged code is the proposal.

That’s a completely different game for anyone who codes better than they sell. Which, in my experience with Indian developers, is most of you. I have watched genuinely excellent engineers lose Upwork jobs to worse engineers with better English, and there was nothing I could tell them that fixed it.

Go open Algora’s bounty board right now, before you read the rest.

Not to claim anything. Just to see that companies with real funding are sitting there with cash attached to issues, waiting.

“$50 to $500 per issue. That’s not a living.”

Correct. It isn’t, and anyone selling it to you as one is lying.

The visible range on Algora in 2026 runs roughly $50 to $2,500 per bounty. Opire sits similar, mostly $50 to $500, with rare outliers into the thousands. One Godot Engine bounty reportedly hit around $13,000, and that number gets passed around constantly, which tells you how unusual it is.

So if you’re calculating rupees per hour, close the tab. The math doesn’t work and it was never supposed to.

Here’s what the money is actually for. Algora’s own positioning now leans on companies using bounties and contract work to trial developers before hiring them. Their site carries testimonials from founders who met a contributor through a bounty, contracted them for a few months, then brought them on full-time. One is from Firecrawl. One is from Trigger.

Read that again from your side of the table.

A funded startup has a mechanism where they pay strangers small amounts to demonstrate they can work in their codebase, and they use it to decide who to hire. You are being invited to interview, with money, asynchronously, with no call, no timezone problem, and no one asking why you left your last job.

The $200 isn’t income. It’s a receipt for work a funded company watched you do.

“I don’t have open source contributions”

Nobody does, right up until their first merge.

The thing people get wrong is treating this like reputation building, where you need a year of commits before you’re allowed to earn. You don’t need standing. You need one issue you can genuinely close.

The mistake I’d expect you to make, because it’s the one I’d make, is reading the issue and not the code. The issue tells you what someone wants. The codebase tells you whether the fix is twenty lines or a week of untangling someone’s conventions. Read the code first, every time. Most of what separates a merged PR from a rejected one is whether you wrote it the way the project already writes things.

Start with a project whose stack you actually use daily. Not the most interesting bounty. The most boring one, in the framework you know best.

Where to actually go

Four worth your time, with honest differences.

Algora is where I’d start. GitHub native, companies fund issues in USD, you get paid on merge through Stripe Connect. The bounty board is public and there’s an open endpoint that lists live bounties, which means you can automate your own alerts instead of refreshing a page. Deepest pool of funded companies of the four.

Opire lets anyone put a bounty on any GitHub issue, not just maintainers. The developer receives the full reward, with the fee paid on top by whoever created it. Wider surface, thinner concentration of serious money.

Polar.sh is really a monetization platform for maintainers, with issue funding as one feature among several. Relevant if you’re building your own thing and want it funded, less so if you’re hunting.

One caution that isn’t optional. BountySource, an earlier version of this idea, ended up holding contributors’ funds, and the way that ended still shapes how these platforms are designed — Opire’s whole payment structure exists partly as a reaction to it. Newer platforms in this category also come and go. Do not let money sit unwithdrawn on any of them because you’re waiting to hit a nicer round number.

“Why would they pick me over an American developer?”

Because the pull request doesn’t have a nationality on it, and because the payment rails finally work.

Algora settles through Stripe to roughly 150 countries, within a few business days of merge. That sounds boring. It isn’t. The reason this channel barely existed for Indian developers before is that getting small international payments in was more painful than the payment was worth.

There’s also a structural advantage nobody mentions: this is asynchronous. No 11pm calls with a client in California. You work when you work, the PR sits there, it gets reviewed. For anyone doing this alongside a job, that difference is the whole thing.

The comparison I keep coming back to

Upwork: write a proposal, spend connects, wait, get compared on price, maybe get a reply. Repeat forty times. Zero paid until someone picks you.

Bounties: read an issue, write code, get merged, get paid, and the company now has a working sample of you.

Both work. I did the first one for years and it built everything I have. But Upwork charges you for the chance to be evaluated, and bounties pay you while it’s happening.

If you’re currently in a job and looking for the first crack in the wall, I know which one I’d hand you.

Two things I’m keeping out of a free article, because they’re the parts that decide whether this works or wastes three weekends:

How to triage a bounty before you write a line. There’s a checklist for reading an issue’s comment thread, the maintainer’s response speed, and whether someone else has silently been working on it for a week. Most people lose their first three bounties to issues that were never winnable.

And what you send after the merge. That message is worth more than every bounty you’ll claim, and almost everyone skips it because they think the transaction ended when the money landed. It didn’t. That’s the moment a company that just watched you work is most likely to say yes to something bigger.

No call, no audit

I’m not going to review your GitHub or tell you which bounty to pick. Nothing useful comes from looking at one issue.

What exists instead is the room. It’s called The High Agency Club, it’s free right now, and it’s mostly people trying to build income outside a job without quitting first.

join community

If you’re a founder reading this, come look anyway — the people learning to close bounties in other companies’ codebases are exactly the people you want in yours.

4 years of doing this the hard way, and the door I’d walk through today is not the one I walked through.

P.S. The question I don’t have a clean answer to yet: what happens when your PR is good and the maintainer just never responds? It’s happened to me and I handled it badly. Tell me your version in the comments and I’ll build part two around it.

Before that, I did two jobs for free. Not cheap — free. They were hard deep-learning problems, the kind most people can’t solve, and I solved them and charged nothing. Why? Because I didn’t know my value yet. When you can’t see your own worth, the market sets it to zero for you.

That $15 project became more than $250,000 earned on Upwork. Clients across the world. Products built off the back of it. A VC fund reached out to me from Abu Dhabi. ( built an AI research lab burnt $1M last year)

The distance between $15 and $250K was never talent. It was the willingness to stop hiding.

I’m telling you this because of someone I met on a call this week — and I think she might be you.

The woman sitting on a launchpad

Her name is Sana. She’s a business lawyer — contracts, due diligence, startup documents. On Upwork since 2023. Seven completed jobs. Every single one five stars. Real clients in the US, UAE, Germany.

Total earned: $400.

Not because she can’t do the work. Because she’s frozen. Her words: “I’m confused about the strategy. If I get an invitation, I apply. Maybe once a week I find a job that fits.”

Once a week. Sitting on a five-star, seven-review profile — one of the strongest signals a freelancer can have — applying when she remembers to.

Here’s what I told her, and it’s the reason I’m writing this:

That $400 profile isn’t a small number. It’s a loaded gun she won’t fire.

The thing nobody tells you about who’s on the other side

Most people think freelancing means hunting. Cold emails, cold calls, chasing strangers, convincing them they have a problem.

On Upwork it’s the opposite — and this is the shift that changes everything.

The clients who post are high-intent. They’ve already decided they need a lawyer, an engineer, a QA lead. They already know what they want. You’re not walking up to a stranger to convince them. They’ve made up their mind and they’re holding the budget, waiting for the right person to raise a hand.

Sana felt this the second I said it. Her pain with clients from LinkedIn and referrals: three months of follow-up, one client lecturing her before even paying about how she “might run away,” endless back-and-forth. On Upwork that friction is gone. Client verified, money in escrow, dispute system if anything breaks. They give the work, you do it, you get paid.

There are startups all over the world — Silicon Valley especially — that need exactly what you do and don’t know how to hire for it. I run a company. We need legal documents, and it never even occurred to me I could hire a consultant on Upwork until recently. That’s not a gap in you. It’s a market waiting on the other side of a profile you haven’t finished building.

Why the first twenty days feel impossible

Let me be honest about the hard part, because pretending it’s easy helps no one.

When you’re new, you’ll send proposals into silence. On that same call, one member had sent 17 proposals — 2 viewed. Another sent 7, none viewed yet, and he’s making a custom video for every single one. This is normal. It is not you failing.

Here’s the mechanic underneath it: Upwork is an algorithm, and it’s cold to you until you have proof. Once you have three or four clean reviews, it flips — the platform starts pushing your profile, sending invites, putting you in front of clients looking for exactly what you did last time.

The first twenty days are a filter, not a wall. Most people quit inside it. The ones who don’t come out the other side, where it stops being a grind and starts being a system that feeds them.

And here’s the part that should stop you cold: two people on that call got Upwork’s Rising Talent badge without earning a single dollar. That badge means the platform already sees something — you’re effectively top 10% of profiles before you’ve made a rupee. Most people who open an account never get it.

So when you have that, and you stall — that’s the real tragedy. I’ve watched people hit $10,000 and freeze, and the moment they go quiet, the algorithm pushes them back down. Their account was tuned to grow from exactly where they stood, and they walked away from the launchpad.

That’s Sana’s actual risk. Not failure. Stalling on a profile that’s ready to fly.

So I recorded exactly how to not be Sana

I could keep writing, but this is the kind of thing you need to see — the actual profile, the actual filters, the actual moves, on screen.

So I recorded the full breakdown and put it inside our community. In the video I walk through:

• How to turn a Rising Talent badge into actual invites (most people never do this)

• The exact way I read a client’s profile before spending a single Connect — so you stop firing into silence

• How to write the first line of a proposal that a busy client actually stops on

• And the real reason your proposals aren’t getting viewed yet (it’s fixable this week)

It’s free. It’s inside the community, next to the room where people post their profiles and get them pulled apart — the confused ones, the stuck ones, the ones one small fix from their first client.

→ Watch the full breakdown here: skool community

If you’re Sana — reviews on the board, frozen anyway — this is the fifteen minutes that unsticks you. If you’re at zero, it’s the map for the first twenty days.

The clients are already out there. They’ve already decided to pay. The only question is whether your profile is built to be found when they go looking — and whether you’ll stop setting your own price to zero the way I did with those two free jobs.

That’s the whole game. The video shows you how to play it.

→ live call 2 aug

P.S. Next week I’m breaking down the exact profile setup that turns a Rising Talent badge into invites — the thing Sana’s missing. It’ll go in the community first. If there’s a specific thing about your own profile that’s got you stuck, reply and tell me — I’ll answer the sharpest ones in that breakdown.

You came here from the reel, so I’ll skip the warm-up.

That balance came from a government route most Indian founders never touch. No pitch deck. No accelerator. No investor introduction. No sales call. It’s a form on the AWS console and a certificate you probably already have sitting in a folder somewhere.

Five thousand dollars in AWS credits. For DPIIT-recognised startups. That’s roughly ₹4 lakh of cloud runway, and if you’re running GPU workloads or Bedrock inference, you already know what that buys.

Most people I talk to fall into one of two camps. Either they’ve never heard of the DPIIT route at all, or they heard about it, assumed they wouldn’t qualify, and never checked. The second group is bigger, and they’re wrong more often than they’re right.

Here’s the whole thing.

What you’ll find below

• The five eligibility rules, and the one that disqualifies most people who apply

• Where your Organization ID actually lives on the DPIIT certificate

• What happens if you already claimed AWS credits before

• Why the partner perks are worth more than the credits themselves

• The form that connects you to a vendor if you’d rather not do this alone

• What I’d spend it on if I were starting again today

“Is this actually a real government program?”

Yes. It’s on the Startup India portal, which is run by DPIIT under the Ministry of Commerce and Industry.

The specific thing is called AWS Activate Portfolio. There are two ways into Activate generally. The self-serve tier gives you $1,000 and asks almost nothing of you. The Portfolio tier gives more, and normally it requires you to be backed by an approved accelerator or VC.

DPIIT recognition is what gets you into Portfolio without the accelerator. That’s the whole trick. Your government startup certificate does the job a YC acceptance letter would otherwise do.

One honest caveat before you get excited. AWS states plainly that they can accept or reject any application at their own discretion, and the credits are subject to their promotional credit terms. This is not an entitlement. It’s an application with good odds if you tick the boxes.

“Do I actually qualify?”

Five rules. All five, not four.

One. You’re recognised by Startup India DPIIT and you have your Organization ID.

Two. You have not previously received AWS Activate credits of equal or greater value from any Activate provider. Read that again if you’ve dabbled with AWS before, because this is the rule that kills most applications.

Three. Self-funded, or funded up to pre-Series B.

Four. You have a fully functioning company website.

Five. Founded within the past 10 years.

Number four sounds trivial and it isn’t. A Notion page with a waitlist form is not a functioning company website. A dead link in your registration documents is worse than no link. Fix that before you apply, not after you get rejected.

“Where do I find my Organization ID?”

It’s printed on your DPIIT recognition certificate.

[IMAGE 2 HERE — screenshot of a DPIIT certificate with the Organization ID circled or highlighted. Blur your own company details. This is the single most useful image in the whole article, because it’s the thing people hunt for and can’t find.]

If you can’t find your certificate, you can verify and download it again from the Startup India portal under the Recognition menu. If you’ve never registered for DPIIT recognition at all, that’s step zero, and it’s a separate application before any of this becomes relevant to you.

“I already got AWS credits once. Am I finished?”

Depends entirely on how much.

The rule is equal or greater value. So if you took the $1,000 self-serve Founders credit at some point, that’s less than $5,000, and you’re not automatically out. If you’ve already had $5,000 or more from an Activate provider, you are.

I’ve seen people assume any prior AWS relationship disqualifies them and quietly give up. Check your actual number before you decide that about yourself.

“What else comes with it besides the credits?”

This is the part the reel didn’t have room for, and honestly it might matter more than the cash.

Alongside the credits, you get access to AWS partner offers valued up to $800,000. That figure is a ceiling across the whole catalogue, not a cheque, so don’t quote it as money you received. But the individual offers inside it are real, and some of them cover tools you’re already paying for out of pocket.

You also get premium AWS training content and access to curated accelerator programs.

The credits run out. The partner offers you actually use tend to stick around in your stack much longer.

“Can someone just do this for me?”

Fair question, and this is where the form comes in.

I’ve set up a short form. Fill it in, and it helps you get connected with a vendor who handles this route. It’s free. There’s no charge to you, no commission taken out of your credits, and I’m not making money on your submission.

$5000 eligibility check

What the form needs: your company name, whether you’re DPIIT recognised already, your Organization ID if you have it, your website, and roughly what you’re planning to run on AWS. That last one matters because it changes what the vendor recommends.

If you’d rather do it entirely yourself, that’s completely fine. The AWS console application and the official application guide are both linked from the Startup India AWS page. You don’t need me or anyone else in the middle. The form exists for founders who’d rather have someone who’s done it before walk them through it.

“What would you spend $5,000 on?”

Since you’re reading a founder’s newsletter, not AWS documentation, here’s the honest answer.

Bedrock inference, if you’re building anything with AI in it. You get Claude, Llama, and Mistral through it, which means your model costs run on AWS money instead of yours during the exact months when you have no revenue and every rupee is coming out of your own pocket.

I trained my first machine learning models on my university’s GPUs, because I could not afford compute. That’s not a humble-brag, it’s just what happened. It’s also the reason I chase every single credit program I hear about, and why I keep posting them instead of hoarding them.

[IMAGE 3 HERE — optional. Your AWS console showing actual usage or spend against the credit. Proof that you used it, not just claimed it. Skip this if you don’t have a clean screenshot.]

Set a calendar reminder for your expiry date the same day your credits land. AWS has been strict about expiration and there are no extensions. Credits you never spent are worth exactly zero, which is the same as never having applied.

Where the rest of this lives

I run a free community called Crack Freelancing. It started as an Upwork thing and it’s turned into where I put every credit program, every tool discount, and every teardown of what actually worked for us at PGAGI.

Join community

If you found this useful, that’s where the next one goes first. Link’s in my Instagram community.

No paid tier, no upsell. I’ll ask you for nothing except that you actually apply.

P.S. — The question I keep getting and haven’t answered yet: what to do when your DPIIT application itself gets rejected, which happens more than people admit. If that’s where you’re stuck, tell me in the comments. That’s part two.

Everything above is from the official Startup India AWS page and AWS Activate’s own terms as of publication. Programs change, figures change, and AWS can reject applications at their discretion. Verify against the official pages before you rely on any of it.

That’s my usage screen this week. The weekly ceiling is 50% higher than it was back in April. I didn’t upgrade my plan. I didn’t apply for anything. No email came.

It just showed up. And if you’re paying for Claude Code right now, it’s almost certainly sitting in your account too, unused, because nobody told you either.

I ran the numbers on what that extra headroom is worth in API terms. For how I use it, it lands around ~$2,600 over the promo window. I’ll show you exactly how I got there, because a number that big deserves the working, not a screenshot and a “trust me.”

Here’s the whole thing, before it ends on August 19.

What you’ll find below:

• What Anthropic actually changed in May, and why almost no one noticed

• Whether this is real or another “credits” clickbait, and the honest answer

• Who gets it, who doesn’t, and the one line in the terms everyone misreads

• How I calculated the API-equivalent value step by step, so you can run it on your own usage

• The second promo hiding right next to this one, worth even more if you touch Cowork

• The exact command to check what’s already in your account

• What I actually built with the extra headroom that I couldn’t ship before

• Why “no cash value” in the terms doesn’t mean what you think

Straight into the questions I keep getting.

“Is this actually real, or just more credits hype?”

Real. But not the way the hype accounts are framing it.

Here’s the thing you have to get straight first, because I nearly got it wrong myself. This is not a $2,600 credit. Anthropic did not put money in your account. No balance, no coupon, no “apply here” button. Anyone telling you to comment a keyword to claim your free $2,600 in Claude credits is selling you something that doesn’t exist.

What’s real is a usage-limit increase. From May 13 through August 19, 2026, your weekly usage limit in Claude Code is 50% higher. That’s the whole promotion. It applies everywhere you use Code: the CLI, the IDE extensions, desktop, web.

The dollar figure still matters, though, because usage has a market price. If you bought that same throughput as API tokens, it would cost you real money. So the promo carries value even though it isn’t a credit. Those are two different sentences and I’m keeping them apart on purpose, because that gap is the whole difference between honest and clickbait.

“Okay, so do I even qualify?”

Probably, if you already pay.

It’s on for Pro, Max, and Team plans, plus legacy seat-based Enterprise. Free plans miss out. Consumption-based Enterprise seats miss out. And here’s the part people skip right past: there’s nothing to do. If you’re on an eligible plan, it applied itself. You’ve been sitting on it for weeks.

One command shows you. I’ll get to it, but see the value math first, because it changes how you’ll want to spend the next few weeks.

“Fine. How much is it really worth?”

This is where I have to be careful, and where most people just invent a number.

I won’t. I’ll show you the shape of the calculation and let you run your own, because your number won’t be mine.

The method is simple. Take your typical weekly Claude Code consumption. Work out what that same volume of input and output tokens would cost at current API rates for whichever model you actually run. Then apply the extra 50% headroom across the roughly 14 weeks of the promo window. That’s your figure.

Mine came out around ~ $2,600, and honestly that’s only because I lean on Code so hard for client builds at PGAGI. A lighter user might see a few hundred dollars of equivalent value. Somebody living inside Code all day could clear more than me.

The exact rate card I used, the per-week breakdown, my model mix, and the sheet I plug my own usage into, I keep all of that inside the Crack Freelancing community. Free to join. I’m not pasting a 40-row cost table into a Substack post. It lives somewhere I can keep it current when the rates move, and they move.

“Wait, there’s a second one?”

Yeah. And if you use Cowork it might matter more than the first.

Different promo, different product. From June 5 through August 5, your 5-hour usage limit in Claude Cowork is doubled. Not the weekly cap. The 5-hour window. Automatic again, Pro/Max/Team again, Free excluded again. Most people running one don’t clock that the other is live too. Figuring out how to split your work across Code and Cowork while both windows overlap is the real optimization, and almost nobody’s doing it.

“How do I check what I’ve got right now?”

Open Claude Code, run usage in the CLI. Your updated weekly limit is right there. Thirty seconds and you know.

If it reads higher than you remember from spring, that’s the promo. It’s been quietly working whether you noticed or not.

“What does ‘no cash value’ in the terms mean, then?”

It means you can’t withdraw it, transfer it, or stack it with another offer. It does not mean the extra usage is worthless. Different claims. The terms are being precise, not stingy. You can’t cash it out. You can absolutely use it.

Which is the only thing that actually matters here.

The promo isn’t the story. It ends August 19 no matter what. What you build with the extra room between now and then, that’s the story. I used mine to finish a client tool I’d been rationing tokens on for weeks. It’s still running now that the promo’s winding down. I borrowed the usage for a month and kept what it made.

If you want the value sheet, the API rate card, the Cowork stacking method, and the running log of what I’m building with the extra room, it’s all inside Crack Freelancing. Free community, link’s in the channel. I won’t do 1:1 “is my usage worth it” audits, though. That’s literally what the sheet is for.

Go run usage. See what’s been sitting there this whole time.

P.S. — the limit questions are going to pile up, I can already tell. Which model to run to stretch the headroom furthest. Whether the weekly cap or the 5-hour cap is your real bottleneck. How to read the reset timing. Drop your Claude Code limit question in the comments. I’m answering the best ones in part 2.

Not on ads.

On clippers.

More than 1,500 people cutting clips. Nearly 70,000 clips/month pushed across the internet.

[IMAGE: Screenshot of Devin Nash / Clavicular clipping campaign post showing $650K/month, 1,500+ clippers, 70,000 clips/month]

And that is just one campaign.

Daniel Bitton posted that Content Rewards paid out $50,000+ to clippers in one week.

[IMAGE: Screenshot of Daniel Bitton tweet showing $50K+ paid to clippers in one week]

Musa Mustafa reportedly made $22,000 in one month from TikTok alone without showing his face.

Some platforms now have entire armies of people clipping podcasts, streamers, founders, and brands.

See, the crazy part is not that people are making money from clips.

The crazy part is that most of them are still doing it manually.

They find a long video.
They cut the clip.
They add captions.
They export.
They upload to TikTok.
Then Instagram.
Then YouTube Shorts.
Then X.
Then LinkedIn.
Then they do it again tomorrow.

That is not a business.

That is a content treadmill.

The people who will actually win this game will not be the people who edit faster.

It will be the people who build systems.

That is why I’m writing this.

I promised this article on Instagram after people commented “clip” on the reel. So this is the full breakdown of what I call:

The Clipping Agent Stack

The basic idea is simple.

You take long-form content from one creator, founder, streamer, podcast, or niche.

You turn it into short clips.

You distribute those clips across every platform.

Then you monetize through creator payouts, brand deals, affiliate offers, client retainers, or your own distribution.

Now add agents.

Suddenly, the work changes.

Instead of manually editing 5 clips a day, you send one message:

“Take the latest long-form video, find the best 5 moments, caption them, format them for short-form, and schedule them across TikTok, Reels, Shorts, X, and LinkedIn.”

And the system runs.

[IMAGE: Screenshot or diagram showing one Telegram message triggering clipping, captioning, scheduling, and posting]

That is the game now.

Not clipping.

Distribution.

Why this is exploding right now

Every creator has the same problem.

They make one long video, podcast, stream, webinar, interview, or YouTube upload.

Then 95% of the value dies inside that one post.

But a good clip can travel.

One 40-minute video can become:

• 10 TikToks

• 10 Reels

• 10 Shorts

• 5 X posts

• 5 LinkedIn posts

• 5 Reddit-style discussion posts

• 3 newsletter angles

• 1 carousel

• 1 founder story

That is not repurposing.

That is distribution leverage.

And honestly, this is why clipping is becoming one of the easiest online businesses to understand.

You do not need to invent a new product.

You do not need to become famous first.

You do not even need to show your face.

You take attention that already exists and move it into places where people are already scrolling.

That is why streamers use clippers.

That is why founders need clippers.

That is why agencies can sell this.

And that is why I think clipping is going to become one of the biggest ways creators, freelancers, and small agencies make money online over the next 12 to 18 months.

What a clipping page actually does

A clipping page is very simple.

You pick one creator or one niche.

Then you cut their long-form content into short clips and post those clips everywhere.

That’s it.

A streamer says something wild on a livestream.
You clip it.

A founder explains something useful in a podcast.
You clip it.

A finance creator drops a strong opinion.
You clip it.

A podcast guest says one line that could trigger comments.
You clip it.

The clips get views.

Views become money.

Sometimes the creator pays directly.

Sometimes platforms pay through clipping campaigns.

Sometimes brands pay for UGC-style distribution.

Sometimes you use the clips to drive traffic into an offer.

And sometimes, if you are smart, you use the clipping page as proof to sell distribution as a service.

That is where the money starts becoming interesting.

The trap most people fall into

Most people think clipping is an editing business.

It is not.

Editing is the labor layer.

Distribution is the business layer.

If you are only editing clips one by one, you will hit a ceiling very fast.

Because to make real money, you need volume.

Let’s say you want to post:

• 5 clips/day

• across 4 platforms

• for 3 accounts

That is 60 uploads/day.

And some kids are doing even more than that.

Nobody serious is doing this by hand for long.

They either burn out, hire people, or build systems.

The clipping economy rewards volume, consistency, and pattern recognition.

The agent stack handles the volume.

Your taste handles the pattern recognition.

That is the part people miss.

The agent does not replace taste.

It replaces the boring clicking.

You still pick the niche.
You still decide the angle.
You still understand what will trigger attention.
You still know which clips have tension.

But once you make that decision, the system should do the work.

Why I care about this as a founder

I am not looking at clipping only as a side hustle.

I am looking at it as a distribution layer.

At PGAGI and Toingg , we are building our own version of this internally.

[IMAGE: Screenshot of PGAGI / Toingg content dashboard or internal distribution workflow]

We have already crossed 3M+ views earlier without even building a full clipping engine around it.
We have also had content distribution moments that touched 14M+ impressions around 3 months back.
Now the goal is to build toward a 100M+ distribution system across Instagram, Reddit, LinkedIn, X, and other channels.

[IMAGE: Screenshot of Instagram reel analytics showing 3M+ views or high-performing distribution proof]

[IMAGE: Screenshot showing 70M+ impressions in total or distribution proof]

And see, this is where founders should pay attention.

AI changed how fast you can build.

But social media is changing how fast you can distribute.

The next unfair advantage is not just “I can build with AI.”

Everyone can build with AI now.

The unfair advantage is:

Can you distribute your ideas, product, proof, and story every single day without needing a 20-person content team?

That is what this stack is really about.

For creators, it is a money-making system.

For freelancers, it is a service.

For agencies, it is a retainer offer.

For founders, it is distribution infrastructure.

Same system.

Different monetization.

The high-level system map

Here is the simple version.

You need four layers.

1. Source layer

This is where the long-form content comes from.

It can be:

• YouTube podcasts

• livestreams

• founder interviews

• webinars

• recorded sales calls

• Instagram Lives

• X Spaces

• internal founder videos

• long talking-head videos

• client content

For my own companies, this can be founder content from PGAGI and Toingg.

[IMAGE: Screenshot of long-form content library or content source folder]

2. Clipping layer

This is where the AI finds the best moments and turns long videos into short-form clips.

This is where tools like Vugola, Opus-style clippers, or other AI clipping tools come in.

The job here is not just cutting.

The job is finding moments with:

• tension

• curiosity

• strong opinion

• money angle

• status shift

• controversy

• practical value

• emotional spike

This is where most bad clipping pages fail.

They clip random moments.

Good clipping pages clip moments that make people stop scrolling.

3. Agent layer

This is the brain.

This is where something like Hermes Agent, OpenClaw-style workflows, or another agent harness coordinates everything.

Instead of you opening five tools manually, the agent does it.

It can:

• take a command from Telegram

• fetch the video

• send it to the clipping tool

• wait for clips

• download clips

• send them to the scheduler

• prepare captions

• post across platforms

• report back when done

[IMAGE: Screenshot of Hermes ]

This is the difference between “I use tools” and “I run a system.”

4. Distribution layer

This is where the clips actually go out.

Postiz, Zernio, or similar tools can help schedule and publish across different platforms.

The goal is not to post once.

The goal is to build a machine that keeps pushing content across every surface where attention exists.

Instagram.
TikTok.
YouTube Shorts.
X.
LinkedIn.
Reddit.
Threads.
Maybe even newsletters.

[IMAGE: Screenshot of Postiz / Zernio dashboard with connected platforms]

That is the whole map.

Source → clipping → agent → distribution.

Simple.

But not easy if you are doing it manually.

The money model

There are three ways to make money with this.

1. Clip for creators

This is the most obvious.

You find creators who already have long-form content but are not distributing it properly.

You offer to clip their content.

At first, you can do it for free or on a performance deal.

Then you move into monthly retainers.

For example:

• $500/month for basic clipping

• $1,000/month for clipping + posting

• $2,000/month for multi-platform distribution

• $3,000/month if you manage the full system and reporting

This is where freelancers can start.

Not by saying, “I edit videos.”

But by saying:

“You are sitting on 40 hours of content. I’ll turn that into 100 short-form assets/month and distribute it across 4 platforms.”

That sounds different.

2. Join clipping campaigns

Some creators and platforms already run payout campaigns.

You clip the content, post it, and get paid based on views or performance.

This is why people are talking about Content Rewards, Whop-style clipping communities, Clipping.net, and creator clipper programs.

[IMAGE: Screenshot of clipping campaign payout page or campaign marketplace]

This model is more volatile.

You can make money fast, but you are dependent on the campaign.

Still, it is one of the easiest ways to learn the game.

3. Build distribution for yourself

This is the founder version.

You stop thinking of content as “posting.”

You start thinking of it as a distribution asset.

One founder video becomes 30 clips.

One podcast becomes 50 assets.

One case study becomes 10 reels, 5 LinkedIn posts, 3 Reddit posts, and 1 Substack article.

That is what we are building internally.

Because if you can build a system that distributes your own company, you are not dependent on ads, agencies, or random viral luck.

You have a repeatable engine.

What most people should do first

Do not start with 10 creators.

Do not start with 5 niches.

Do not start by building some complicated SaaS.

Start with one niche.

One creator type.

One posting system.

One offer.

For example:

“I help AI founders turn their long-form content into 100 short-form assets/month.”

Or:

“I help finance creators clip podcasts into Reels, Shorts, and TikToks.”

Or:

“I help streamers build clipping pages that push 5 clips/day across TikTok and Instagram.”

That is enough.

Then build the system behind it.

The tool stack is not the business.

The offer is the business.

The stack just gives you leverage.

The free map

Here is the whole thing before we go deeper.

If you want to make money with clipping, you need:

1. A niche where attention already exists

2. A creator or content source with long-form material

3. A clipping tool that can create short-form clips fast

4. An agent layer that reduces manual work

5. A scheduler/API layer that posts everywhere

6. A simple monetization path

7. A 30-day test before you overbuild

That is the map.

Inside it, I’ll break down the actual stack, setup, commands, workflow, and the 30-day execution plan.

Before you go

If you are reading this and thinking:

“I don’t just want to read about this. I want to actually build something that makes money.”

That is exactly why I’m running a 90-day challenge inside my Skool community.

The goal is simple:

Help you make your first $10,000 online from freelancing.

People DM me asking to review their proposals and profiles one by one. Honestly, I can’t there’s a delivery team and two other companies between me and that inbox.

Every week, I write one post for subscribers with the stuff that doesn’t belong in a public article.

• How I closed a single $60,000 client: the actual transcript, what I said on the call, and the moment the deal almost died

• How a venture fund from UAE was after to build a AI research lab, without me applying for anything. ultimately got funded, and burned a million dollars last year. I’ll tell you the whole story, including the part that hurt..

• The filter I use to only take clients with maximum lifetime value, so one right client pays like five average ones

• And honestly, a lot more. The demo formats per job type, the full call structure, the word-for-word scope clauses. Plus the thing nobody packages: how my thinking changed at each level. There was a time a $5,000 project made my month. Today it doesn’t make sense for me to take it. That shift took me three years. I write so you can make it in less

My goal with this page is simple. I got from zero to $250K+ on Upwork, and I want the people reading this to do the same. That’s it. That’s the whole agenda.

Subscribe now

P.S. Tell me in the comments where you’re actually stuck right now connects, first review, pricing, scope, anything. The most-asked one becomes next week’s post.

You already know how software breaks.

You know where engineers rush.

You know where auth gets messy.

You know the backend is sometimes trusting the frontend way more than it should.

You know one user should not be able to see another user’s invoice, but you also know some tired developer somewhere might have shipped exactly that at 2:00 AM.

And here’s the funny part.

Companies are literally paying people to find these mistakes.

Legally.

It’s called bug bounty hunting.

A company tells the world:

“Here are the systems you are allowed to test. If you find a real security issue and report it properly, we’ll pay you.”

That’s it.

You find a flaw.
You report it.
They verify it.
They pay you.

And no, this is not small money.

HackerOne alone has crossed $300M+ in all-time rewards paid to ethical hackers.

Google’s own Bug Hunters page shows $81M+ in total rewards since 2010.

Now I’ll be very honest with you.

This is not day-one money.

If someone tells you, “learn bug bounty and make $5,000 this weekend,” please run in the opposite direction.

Your first month may be boring. Your first few reports may get rejected. You may spend hours on a program and find nothing. You may find something and realize someone else already reported it.

That is normal.

But see, if you already write code, you are not starting from zero.

You already understand how systems are built.

Bug bounty teaches you how to see where they fail.

Here is the beginner map I would send to a developer who wants to explore this seriously.

A heads up before we start: some things I’ll give you fully here. A few layers do not fit in one public post, like exact lab order, exact first-report template, how to pick low-competition programs, and how to avoid wasting 30 days on the wrong target. I’ll tell you where that deeper layer lives when we get there.

What’s in this one:

Why bug bounty is legal only when you stay inside scope

Why developers have an unfair advantage over random “hacking tutorial” people

The first mistake beginners make when they chase Google on day one

The bug type I would learn first if I was starting today

The platforms to learn on before you touch real companies

What a good bug report actually looks like

Why AI is making bug bounty noisier, and why serious people can still win

The 30-day starting direction I would follow

The honest money truth, because this is not quick cash

1. Bug bounty is not “go hack random websites”

This is the first thing to understand.

Bug bounty is legal because the company gives permission.

And that permission has boundaries.

Those boundaries are called scope.

Scope tells you:

Which domains you can test

Which apps you can test

Which bug types are accepted

Which actions are banned

What kind of proof they need

What kind of testing can get you removed from the program

Most beginners hear “companies pay hackers” and immediately start thinking like a movie character.

Wrong.

If you test outside scope, you are not doing bug bounty anymore.

You are just creating problems for yourself.

Before touching any system, read the rules.

Read what is in scope.
Read what is out of scope.
Read the banned testing methods.
Read the reward table.
Read the duplicate policy.
Read the disclosure policy.

A good bug bounty hunter reads scope like a contract.

Because honestly, that is what it is.

2. Developers have a real unfair advantage here

Most people think bug bounty is about wearing a hoodie and running some crazy tool.

It’s not.

A lot of bug bounty is just understanding how software behaves when users do things the product team did not expect.

And developers are already trained for this.

You know how login works.

You know how APIs are usually structured.

You know where user IDs show up.

You know what happens when role checks are added late.

You know frontend restrictions mean nothing if the backend does not enforce them.

You know how dashboards are built.

You know how payment flows can get messy.

You know how file uploads can go wrong.

You know how “temporary fixes” become permanent architecture.

That knowledge is valuable.

A normal user sees a dashboard.

A developer sees routes, roles, IDs, tokens, requests, permissions, and assumptions.

Bug bounty is often about finding the assumption that breaks.

3. Don’t chase Google on day one

This is where most beginners waste time.

They go to a bug bounty platform.

They see Google, Apple, Meta, Shopify, Coinbase, some huge company.

Then they think:

“If I find one bug here, I’m sorted.”

And then they disappear for three weeks and find nothing.

Big programs are not bad.

But they are crowded.

Very crowded.

There are researchers with years of experience looking at the same targets. Some are full-time. Some have their own tooling. Some understand one bug class so deeply that you are not even playing the same game yet.

Your first goal is not to beat the best hunters in the world.

Your first goal is to build pattern recognition.

That means you learn first.

Then you practice in labs.

Then you read real reports.

Then you hunt small.

Most people skip the middle and directly try to win prize money.

That is like joining Upwork with no reviews, applying only to $50,000 jobs, then saying the platform does not work.

The platform may work.

Your approach does not.

4. If I was starting today, I would learn access control first

If you’re a developer, the first bug class I would seriously study is broken access control.

Especially IDOR.

IDOR means Insecure Direct Object Reference.

Simple example.

You log in and open your invoice.

The URL or API request has something like:

/invoice/123

Now imagine you change 123 to 124.

And suddenly you can see someone else’s invoice.

That is not just a small bug.

That means the system is checking whether you are logged in, but not checking whether that invoice belongs to you.

This is where many real bugs live.

Authentication asks:

“Are you logged in?”

Authorization asks:

“Are you allowed to access this exact thing?”

That difference is everything.

Developers understand this quickly because we have all seen systems where permission checks were added later, rushed, duplicated, or forgotten in one endpoint.

After access control, I would learn:

XSS

API authorization issues

Business logic bugs

Sensitive data exposure

Broken authentication

File upload issues

Basic SQL injection concepts

But don’t try to learn everything in one week.

That is another beginner trap.

Pick one category.

Go deep enough that you can actually recognize it in the wild.

5. Learn on labs before touching real programs

If you are new, don’t start with real companies.

Start with safe labs.

The first resources I’d open:

PortSwigger Web Security Academy

Hacker101

Bugcrowd University

Hack The Box Academy

PortSwigger is very strong because it gives you actual labs. You are not just watching a person explain XSS on YouTube and feeling productive.

You are doing it.

Hacker101 is also good because it was built for web security learning and bug bounty basics.

Your first 2-3 weeks should be boring on purpose.

HTTP requests.
Responses.
Cookies.
Sessions.
Headers.
Auth.
Authorization.
Burp Suite basics.
Reading API calls.
Understanding impact.

This is not sexy.

But neither is getting 15 rejected reports because you didn’t understand what the company actually rewards.

6. A bug is not valuable because it looks weird

This is important.

Companies don’t pay you because you found something strange.

They pay you because you found risk.

There is a difference.

Weak finding:

“The app behaves weirdly when I change this parameter.”

Strong finding:

“Any authenticated user can access another user’s private billing data by changing the invoice ID in the API request.”

The second one has impact.

That is what gets paid.

So every time you find something, ask:

Can I access data I should not access?

Can I perform an action I should not perform?

Can I bypass payment?

Can I take over an account?

Can I expose private information?

Can I escalate my privileges?

Can I affect another user?

If the answer is no, it may still be interesting.

But it may not be payable.

This is where developers can do well.

Because you can explain impact in the language engineering teams understand.

7. Your report matters almost as much as the bug

Finding the bug is half the game.

Explaining it is the other half.

A bad report says:

“IDOR found. Please fix.”

A good report says:

Title: User can access another user’s invoice by changing invoice ID

Asset tested: app.example.com

Impact: Any logged-in user can view private invoice data belonging to other users.

Steps to reproduce:

Create two test accounts.

Log in as User A.

Open User A’s invoice.

Capture the request.

Replace the invoice ID with User B’s invoice ID.

Send the request.

The server returns User B’s invoice data to User A.

Expected result:

User A should receive an authorization error.

Actual result:

User A receives invoice data belonging to User B.

Suggested fix:

Check object ownership on the backend before returning invoice data.

See the difference?

The first report makes the security team work.

The second report makes their job easy.

And if you make their job easy, you increase your chances of being taken seriously.

There is a deeper structure behind good reports: title format, impact framing, screenshot order, video proof, how to avoid sounding dramatic, and how to write suggested fixes without sounding like you’re teaching their team their job.

That deserves its own post.

8. AI is making the space noisier

Now, this part is interesting.

AI has made it easier for people to generate reports.

Which also means many programs are getting flooded with garbage.

People paste something into AI, generate a confident-looking report, and submit it without understanding impact.

Security teams are not stupid.

They can tell.

But honestly, this creates an opportunity for serious people.

Because when everyone is spamming low-quality reports, a clean, reproducible, well-written report stands out more.

Use AI to learn faster.

Use it to understand a concept.

Use it to clean your explanation.

Use it to think through impact.

But don’t use it to fake understanding.

Clients can smell fake work.

Security teams can smell fake reports.

Same game.

Different market.

9. The first 30 days should not be about money

If I was starting as a developer today, this is how I would spend the first 30 days.

Week 1: learn the web security basics

Understand requests, responses, cookies, sessions, authentication, authorization, headers, and how APIs behave.

If you don’t understand the web, bug bounty becomes gambling.

Week 2: practice labs

Do access control, IDOR, XSS, file upload, and basic injection labs.

Do not watch ten videos and call it learning.

Break things in a safe environment.

Week 3: read public reports

Go through disclosed reports.

See what got accepted.

See how the researcher explained impact.

See what screenshots they used.

See what made the report valid.

This is underrated.

You learn faster from accepted reports than from random motivational content.

Week 4: pick one beginner-friendly program

Not ten.

One.

Read the scope twice.

Understand the product.

Create test accounts if allowed.

Map the features.

Look at user roles.

Look at where one user interacts with another user’s data.

Ask yourself:

“If I built this fast, where would I forget the permission check?”

That question alone can open your eyes.

10. The money truth

Yes, bug bounty can pay.

Some bugs pay $100.

Some pay $500.

Some pay $1,000 to $5,000.

Some critical bugs in top programs can pay much more.

But if you are starting, your first goal is not money.

Your first goal is one valid report.

Then another.

Then one paid report.

Then one repeatable pattern.

The money comes after pattern recognition.

This is exactly like freelancing.

Your first proposal may not convert.

Your first client may be small.

Your first review may take time.

But once the pattern clicks, the game changes.

Bug bounty is similar.

At first, you are not earning.

You are training your eyes.

And once your eyes change, you start seeing what other people miss.

Most developers are paid to make software work.

Bug bounty hunters are paid to find where software fails.

Same foundation.

Different angle.

Before you go

People DM me asking to review their upwork proposals and profiles one by one. Honestly, I can’t there’s a delivery team and two other companies between me and that inbox.

Every week, I write one post for subscribers with the stuff that doesn’t belong in a public article.

• How I closed a single $60,000 client: the actual transcript, what I said on the call, and the moment the deal almost died

• How a venture fund from UAE was after to build a AI research lab, without me applying for anything. ultimately got funded, and burned a million dollars last year. I’ll tell you the whole story, including the part that hurt..

• The filter I use to only take clients with maximum lifetime value, so one right client pays like five average ones

• And honestly, a lot more. The demo formats per job type, the full call structure, the word-for-word scope clauses. Plus the thing nobody packages: how my thinking changed at each level. There was a time a $5,000 project made my month. Today it doesn’t make sense for me to take it. That shift took me three years. I write so you can make it in less

My goal with this page is simple. I got from zero to $250K+ on Upwork, and I want the people reading this to do the same. That’s it. That’s the whole agenda.

Subscribe now

P.S. Tell me in the comments where you’re actually stuck right now connects, first review, pricing, scope, anything. The most-asked one becomes next week’s post.

“If I just write a good proposal, the work will come.”

It does not work like that. It almost never works like that.

I have now done $200K+ on the platform. I dont do anything now. I have a 40-person team running delivery. I get inbound invites daily ( I genuinely can't remember the last time I applied to a job) and most months are $12,000 to $15,000 in pure Upwork revenue. (For context: my profile is here.)

Almost none of that came from “better proposals.”

It came from understanding what Upwork actually rewards, which is not what most freelancers are taught.

Here is the playbook I would send to a younger version of me. It is the playbook I now teach in my Saturday sessions.

A heads up before we start: some of these I'll give you in full, right here. A few of them have a layer that doesn't fit in a post exact scripts, exact clauses, exact demo formats. I'll tell you where that layer lives when we get there.

What’s in this one:

• Why your first 100 connects are venture capital and the 30-second check before you spend a single one

• The $10 job I tell everyone to take (most gurus will hate this)

• The one move that 10x’d my proposal acceptance rate and why almost nobody copies it

• The first line that gets proposals read, and the one that gets them scrolled past

• Five tells that a client will burn your weekend visible in the first call

• Why I show up like a consultant, not a freelancer (and the 60-second trust move)

• The scope rule that has saved me more money than any pricing trick

• My exact rate ladder to $89/hour and the mistake that keeps people stuck at $15

• The moment the platform flips and starts working for you

1. Treat your first 100 connects like venture capital

You get 100 connects. Most freelancers spray them across every job that looks vaguely relevant. That is how you end the first month with zero replies and a bruised ego.

Before you spend a single connect, check the client:

• How many proposals are already on the job?

• How many jobs has this client posted historically?

• What is their hire rate? If they have posted 12 jobs and hired on 2, they are a window shopper. Skip.

• Is their payment verified?

• Is the job description specific, or copy-pasted?

• Are they looking for a specialist or "a unicorn at intern price"?
  

2. Your first review matters more than your next 50 proposals

I am going to say something most freelance gurus will not:

Take the $10 job.

I’m serious. Without a single review, you are invisible on this platform. With one 5-star review, you exist.

When you land that first job, don’t just deliver. Overdeliver to a degree that feels almost embarrassing. Send extra. Reply faster than they expect. Make the client feel like they accidentally hired a senior consultant for $10.

You’re not optimizing for money on this job. You’re optimizing for the review that unlocks every job after it.

My first review, for the record:

3. Build the thing before you propose to build the thing

This is the biggest unfair advantage available to freelancers right now and almost nobody uses it.

In the AI era, writing a proposal takes 3 minutes. Every client is drowning in well-written proposals that all sound the same. Most of them are written by AI, and clients can tell.

What is rare is proof.

Before you submit a proposal, build a tiny version of what the client is asking for. A 5-minute Loom of a working demo. A one-page mockup. A small prototype. Whatever takes you 30-45 minutes to make.

Attach it. Open with:

“I built a small version of what you need. Take 60 seconds to look before reading the rest of this proposal.”

This single move has 10x’d my acceptance rate. Not because the demo is brilliant. Because no other freelancer is doing it.

Now - what exactly do you build? That depends on the job type, and this is where most people get it wrong. Build the wrong format and you waste 3 hours per proposal. Build the right one and it takes 30-45 minutes, max. I have a specific format for each job category, plus the Loom structure that goes with it, and honestly it’s too much for one section of one post. It’s a full Saturday session in my community, and I’m writing the condensed version as a subscriber post soon.

One rule I’ll give you now because it’s non-negotiable: do not use AI to fake the demo. Clients spot it instantly, and you’ve just proven the opposite of what you wanted to prove.

4. The first line of your proposal is the whole proposal

Clients are not reading your full proposal. They are skimming the first line of 30 of them.

They do not care about:

• Your degree

• Your past companies

• Your years of experience

• Your “passion for the field”

They care about exactly one thing: “Can this person solve my specific problem?”

Your first line has to answer that question, or the rest of the proposal will never get read.

What works:

“I built a small version of what you need. Here it is.”

What does not work:

“Hi, I am a passionate full-stack developer with 5 years of experience...”

The first version triggers curiosity. The second triggers a scroll.

lets stop your scroll too quick question:

5. Bad clients leak in the first meeting

Even after the filter, some bad clients slip through. They show themselves on the first call. Watch for:
They want a $5,000 scope for $500. The requirements have changed twice in 20 minutes. They can't articulate what "done" looks like. They want a discount and unlimited revisions. They keep referencing what "the last freelancer" did wrong you're about to be the next one they say that about.

A good client respects scope, respects your time, and explains the problem clearly because they have actually thought about it.

If something feels off in the first call, it will be ten times worse in week three. Pass.

6. You are a consultant, not a freelancer

This is positioning, not tactics, and it changed how clients treated me before I said a word.

When you join a client call: good camera, good mic, clean background. You’re not auditioning. You’re showing up like someone the client would be lucky to work with.

I’ll give you the single highest-leverage move from inside the call: repeat the requirement back in your own words. “So what I’m hearing is...” That one sentence builds more trust in 60 seconds than 10 minutes of describing your skills. Most freelancers spend the call talking about themselves, and the client is silently rating that as “this person did not listen.”

There’s a full call structure behind this — the questions I ask, the order, the note format, how I close. That’s one of the things I walk through live on Saturdays, because it only really lands when you see it run on a real call.

You’re not selling skills. You’re selling the feeling that the client made the right choice.

7. Never start a fixed-price job without ruthless scope

This rule has saved me more money, time, and sanity than anything else on this list.

Most freelancer horror stories — “the client kept asking for more, for free, for 3 months” — are not the client’s fault. They’re the freelancer’s fault for not writing the scope down.

Before any fixed-price project starts, the contract states exactly what gets delivered, what does NOT get delivered, how many revisions are included, and what counts as scope creep and how it gets priced. That last one is the clause that does the heavy lifting, and the exact wording matters — phrase it wrong and it sounds hostile, phrase it right and good clients actually thank you for it.

The word-for-word clauses I use are going in an upcoming subscriber post, because they deserve their own teardown. For now, the principle: if a client pushes back on clarity, that’s your first red flag. Good clients welcome it. It protects them too.

8. Raise your rate in a ladder, not a leap

Here is roughly the path I climbed:

$8 → $15 → $25 → $40 → $69 → $89/hour

Each jump came after 4-5 strong deliveries at the previous rate. Not after one job. Not after one review. After a stable cluster of evidence that I could deliver at that level.

Why ladder and not leap? Because every rate jump on Upwork comes with a small drop in conversion. You need a base of proof to absorb that drop. If you leap from $15 to $89 with two reviews, you will sit at zero replies for months and start to doubt yourself.

Today I charge high and clients still come, but not because I am a better writer of proposals. Because the proof stack underneath the rate is undeniable.

9. Then the platform flips

This is the part nobody tells you when you start.

For the first 12-18 months, you are pushing. Every job is an outbound effort. Every proposal is a small bet.

At some point, the curve flips. Reviews compound. Job success score stabilizes. The algorithm starts surfacing you. Invites start coming in unprompted. I now get daily invites. Most months are $12,000 to $15,000 in Upwork revenue alone. I have a 40-person team running delivery so I can focus on other businesses.

None of it came from a magic proposal template. It came from one stack, repeated:

1 small job → 1 review → 1 small demo → 1 strong proposal → 1 happy client → 1 referral → repeat.

The compounding looks like nothing for a year. Then it looks like everything.

Most freelancers fail on Upwork because they treat it like a job board. It’s not. It’s a reputation marketplace. Every connect, every proposal, every delivery is a deposit into a reputation account that pays compounding interest for years.

The freelancer charging $89/hour and the freelancer charging $8/hour are usually not separated by skill. They’re separated by 18 months of deposits.

Start making them.

Before you go

People DM me asking to review their proposals and profiles one by one. Honestly, I can’t there’s a delivery team and two other companies between me and that inbox.

Every week, I write one post for subscribers with the stuff that doesn’t belong in a public article.

• How I closed a single $60,000 client: the actual transcript, what I said on the call, and the moment the deal almost died

• How a venture fund from UAE was after to build a AI research lab, without me applying for anything. ultimately got funded, and burned a million dollars last year. I’ll tell you the whole story, including the part that hurt..

• The filter I use to only take clients with maximum lifetime value, so one right client pays like five average ones

• And honestly, a lot more. The demo formats per job type, the full call structure, the word-for-word scope clauses. Plus the thing nobody packages: how my thinking changed at each level. There was a time a $5,000 project made my month. Today it doesn’t make sense for me to take it. That shift took me three years. I write so you can make it in less

My goal with this page is simple. I got from zero to $250K+ on Upwork, and I want the people reading this to do the same. That’s it. That’s the whole agenda.

Subscribe now

P.S. Tell me in the comments where you're actually stuck right now connects, first review, pricing, scope, anything. The most-asked one becomes next week's post.

The client gets on the monthly call. They have not checked the leads. They have not listened to a single sales call. They have not opened the CRM in two weeks.

But they say it anyway:

“These leads are not converting. Your leads are low quality.”

You know it is not your leads. The CTR is healthy. The CPL is in line with the benchmark you sold them three months ago. The form fills look fine.

But you do not have the data to fight back. So you do what every Meta agency does. You tweak the audience. You refresh the creative. You add a qualifying question to the Instant Form. You promise to “improve lead quality” next month.

And the client churns anyway.

Here is what I want to convince you of in the next 8 minutes:

The lead quality fight is the wrong fight. And the agency that figures this out first is going to eat the ones that do not.

The “lead quality” lie

I have spent the last 3 years building Toingg, an AI-first CRM with voice and WhatsApp infrastructure for SMBs. We have looked at the post-click data for hundreds of businesses that get most of their inbound from Meta ads.

The pattern is almost identical in every account:

The leads are fine. Often better than fine.

What is broken is everything that happens after Meta hands the lead to the business.

• A WhatsApp lead comes in at 9:42 PM. Nobody replies until 11 AM the next day.

• A landing page lead form fills. The salesperson calls 3 days later from a personal phone with no script.

• A click-to-WhatsApp lead asks a real question. The team sends a generic catalog PDF and nothing else.

• A hot lead asks for a quote. The follow-up never happens.

• The CRM gets updated three days later, missing half the conversation.

The lead was qualified. The campaign worked. The pipeline died in the post-click layer because nobody owned it.

Your client looks at the conversion rate and blames the source. You optimize the source. The leak stays exactly where it is.

This is the structural problem with the Meta agency model right now:

You are accountable for an outcome (revenue) but you only have control over an input (leads).

That gap is where your retention dies.

5 leaks killing Meta agency clients (and your retention)

Leak 1: The speed-to-lead gap

Meta leads have a brutal half-life. A lead replied to in under 5 minutes is worth significantly more than the same lead replied to 2 hours later. (The exact multiplier varies by industry and source. I will not pretend I have one clean number you should quote your clients.)

But most SMBs you serve are not built to reply in 5 minutes. They are a small team. They sell. They travel. They forget.

So your beautifully optimized $4 CPL becomes a $40 effective CPL because most of those leads went cold before a human said hello.

You cannot fix this with better creative. This is an operational problem, not a media problem.

Leak 2: The Instant Form trap

Instant Forms get you volume. They also get you the lowest-intent lead Meta has to offer, because the user never left the feed and never confirmed real interest.

You already knew this. You sold it to your client as “we will qualify them later.”

The problem: later usually means a salesperson reading a row in a Google Sheet two days after the form fill. Of course it converts at 2%.

Click-to-WhatsApp and click-to-call leads are dramatically more valuable, but only if the system on the other end of the click handles the conversation in real time. Without that system, you are paying a premium CPM for clicks that go to voicemail.

Leak 3: The WhatsApp black hole

Click-to-WhatsApp is one of the most undervalued ad units on Meta right now, and also the most leaked.

The lead clicks. They land in WhatsApp. They send “Hi, interested.” Then one of three things happens:

1. Nobody replies for hours.

2. Someone replies with a generic “How can I help?” and the lead ghosts.

3. The conversation drifts across 4 days of unanswered questions and dies.

The conversation is the conversion event. If the conversation has no operator, you do not have a campaign. You have a donation to Meta.

Leak 4: The follow-up vacuum

Most leads do not convert on the first conversation. They convert on the third, fifth, or seventh touch.

In almost every SMB account I have looked at, the second touch does not happen. The first conversation ends with “let me think about it” and that is the end of the relationship.

This is not the client being lazy. They are running a business, doing 11 things at once, and follow-up is the thing that always loses to whatever feels urgent.

Your conversion rate is being held hostage by the human attention span of a 4-person sales team.

Leak 5: The agency ceiling

This is the meta-leak. The one that ends the engagement.

You sold leads. The leads came. The client did not convert them. The client blamed you. You lost the account.

The next agency does the exact same thing. Same outcome. Same churn.

The category is in a bad equilibrium. Every agency is selling the same input. Almost nobody is selling the outcome.

What you can sell that is not “more leads”

The agencies I see breaking through this ceiling are not winning on better creative or cheaper CPL. They are winning by selling the layer underneath the ad.

Call it whatever you want. Conversion ops. Lead recovery. Post-click services. The pitch to your client becomes:

“We are not just going to send you leads anymore. We are going to make sure those leads actually become conversations, conversations become qualified opportunities, and opportunities reach a salesperson with full context. Then we report on revenue, not CPL.”

This is a different conversation. It is not “we need more budget.” It is “we are taking responsibility for what happens after the click.”

A few things change when you sell this:

Your contract value goes up. A media retainer is typically $2K-$8K/month. A conversion ops retainer on top can add $2K-$5K. Same client, often double the LTV. (These ranges are based on what I have observed in the market, not a benchmark. Verify against your own pricing data before quoting them to anyone.)

Your churn drops. You are no longer fightable on lead quality alone. You own the outcome side too. The “your leads suck” call gets replaced with “here is exactly where the leak is, and here is what we did about it last month.”

Your reporting changes. Instead of:

“We got you 312 leads at $4.20 CPL this month.”

You report:

“We got you 312 leads. AI handled 287 first conversations within 60 seconds. 54 qualified for sales. 18 had calls transferred to your team with full context. 4 closed for $X. Here is where the other 308 are in the pipeline.”

That second report is undismissable.

This is the upsell. Right now almost nobody is selling it.

What this looks like in practice

Two examples from our customer base, framed how an agency would think about them.

Maputo Rides. We helped recover 34 dormant drivers. At roughly $600 LTV each, that is about $20,400 in recovered lifetime value from people already inside the system. If you were the agency on this account, you just turned a dead list into a recovery campaign you can charge for separately.

TL Pathak Group. We helped generate 600 hot leads. At roughly $4,000+ per qualified lead, that is around $2.4M+ in opportunity value. If you were the agency, that is a number you can put on a renewal deck.

The point is not the absolute numbers. The point is the type of number you get to report. You stop reporting “leads delivered” and start reporting “revenue defended.”

The new stack for Meta-first agencies

If you want to add conversion ops to your offering, the operational stack looks roughly like this:

1. Inbound capture across every Meta surface. Lead forms, click-to-WhatsApp, click-to-call, Messenger, landing page forms, Instagram DMs. All of it lands in one inbox.

2. Sub-60-second first response. AI handles the first conversation on every channel, in the client’s voice, with the right qualifying questions for their industry.

3. One shared memory across channels. Lead starts on WhatsApp, jumps to a phone call, comes back on Instagram. The system remembers everything.

4. Smart human handoff. When a lead goes hot, the call transfers to a human salesperson with the three things they need to know already in front of them. No “sir, please explain again.”

5. Automatic follow-up discipline. Warm leads, cold leads, “call me later” leads all get scheduled follow-up without anyone remembering.

6. AI-first CRM. Updated by the conversations themselves, not by humans typing notes. So your client reporting is always real.

7. Revenue reporting. Not CPL. Not leads. Conversations, qualified opportunities, transfers to sales, closed revenue. Tied back to the campaign that sourced them.

That is what Toingg is built to be. But the more important point is the positioning shift. Whether you build it, partner for it, or white-label it, this is the layer your agency needs to own.

A 5-minute audit you can run on any client today

Pick your highest-spend Meta client. Answer these honestly:

1. What is the median response time to a lead form, WhatsApp, or Messenger message? If you do not know, that is leak #1.

2. What percent of leads got a second touch within 7 days? If under 50%, that is leak #2.

3. How many WhatsApp leads got no reply at all last month? Pull the number. It will scare you.

4. What does the CRM say about leads from 60 days ago? If most are stuck at “new” or “uncontacted,” your client is buying leads and burying them.

5. What is the client’s actual close rate on Meta leads? If it is under 5% and they blame the leads, the leaks above are why.

You will find 2 to 4 of these broken in almost every account.

Each one is a service you could be selling.

Who this is for

This is for you if:

• You run a Meta-first agency with at least 5 clients spending $3K+/month on ads

• Your retention is the metric that keeps you up at night

• You are tired of the “your leads are crap” monthly call

• You want to expand contract value without doubling your media team

• You are open to partnering on the post-click layer instead of building it from zero

It is not for you if:

• You are still figuring out your first 3 clients

• You think a CRM integration plugin is the same as a conversion system

• You do not have basic reporting on what happens after the lead

What to do next

I am opening a small set of agency partnerships through Toingg this quarter. The structure is flexible: implementation partner, revenue share, white-label, or referral commission. We bring the AI-first CRM, voice and WhatsApp infrastructure, and human-in-loop workflow. You bring the client relationship and own the strategy layer.

Want the Meta Agency Lead Leakage Audit? (A longer version of the 5-minute audit above, designed so you can run it as a paid or free service for your clients.) Comment AGENCY below.

Want to talk partnership? Book a call here
or DM me PARTNER.

We are on a mission to save $1B in leaked leads by 2027. We are at $7.78M so far. Most of it has come from SMBs who were already paying Meta agencies.

The agencies just did not own the layer where the recovery happened.

The next wave of winners in your category will.

I am betting on that.

Voice AI. WhatsApp and Instagram bots. CRM sync. Memory. Follow-ups. Analytics. Dashboards. Integrations.

The product worked. The AI replied. The calls happened. The dashboard updated. The CRM filled in.

And then I learned the most expensive lesson of my founder life:

Software working is not the same as the customer getting an outcome.

That one sentence rewired how we build Toingg. And I think it is about to rewire the next generation of B2B software.

Because the future is not just SaaS. The future is Outcome-as-a-Service.

Not all software will die. But the kind of software that gets bought, logged into twice, and quietly dies inside a company? That category is in serious trouble.

The lesson nobody tells you when you build AI for SMBs

In the early days, people came to us because they wanted to “try AI.”

That sounds exciting as a founder. It is also dangerous. Because “trying AI” is not the same as solving a real business problem.

One of our UK customers in career consulting had 3 team members specifically assigned to implement our system. Smart people. But when it came to writing prompts, configuring workflows, designing qualification logic, and operating the system day to day, they struggled.

That is not a criticism of them. That is the reality of SMBs.

They do not have AI operators sitting inside the company. No forward-deployed engineers. No internal automation team. They do not wake up thinking about prompt tuning, CRM hygiene, or model behavior.

They wake up thinking:

Where are my leads? Who followed up? Which customer is serious? Why is revenue not moving?

We generated around $55K+ in pipeline for that customer. But the bigger lesson was not the pipeline. It was this:

A customer can get value from software and still fail to turn that value into a repeatable outcome.

That is where most AI tools break. Not at the demo. Not at the dashboard. Not at the model. They break in the messy middle of the business, where humans forget, CRMs go stale, salespeople miss follow-ups, and nobody knows what happened after the first conversation.

The messy middle is where software usually stops.

It is also where outcomes begin.

5 traps quietly killing AI software for SMBs and agencies

Trap 1: The Dashboard Trap

A dashboard looks beautiful. Charts. Filters. Lead stages. Conversation history. Conversion numbers.

A dashboard does not convert a lead. It shows you what already happened. If nobody acts on it, it becomes a graveyard with better design.

SMBs do not need one more place to look. They need a system that moves the lead forward.

Trap 2: The CRM Graveyard

Most SMBs say they have a CRM. Look closer and it is messy, outdated, or simply not used by the actual sales team.

This is especially true in ground-level industries. Machinery. Real estate. Mortgage. Education. Local services. Recruitment. Fleet.

The owner wants a CRM. The sales team is working on WhatsApp, phone calls, Excel sheets, missed calls, and memory.

One of our heavy-machinery customers had accumulated 2,000+ leads that were never properly stored in one usable CRM. Not because the team was lazy. Because their daily workflow was not CRM-first. They were selling machines, sharing catalogs, checking inventory, handling delivery questions.

Traditional CRM expects humans to update the system. In the real world, humans are busy closing, forgetting, traveling, and firefighting.

So we built Toingg differently. Not as a traditional CRM. As an AI-first CRM where the system updates the lead based on the latest WhatsApp message, phone call, Instagram DM, human takeover, follow-up, or call transfer.

The CRM should not depend on humans typing notes. It should learn from the communication itself.

Trap 3: The Configuration Gap

Most AI tools assume the customer knows how to configure AI. That assumption is wrong.

A founder understands the business. A salesperson understands the customer. Neither necessarily knows how to write the right prompt, structure qualification, define hot vs warm vs cold, or build fallback rules.

A machine buyer does not want to answer a rigid form. They may not even know exactly which machine they need. If AI immediately fires off:

“What is your budget?” “What is your timeline?” “What is your location?”

…before understanding the actual requirement, it feels stupid.

Good qualification is not a checklist. It is a conversation. Understand first. Clarify. Follow up. Qualify intent. Move the lead to the right next step.

The value is not in having AI. The value is in making AI behave correctly inside a specific workflow.

Trap 4: The “AI Will Do Everything” Fantasy

A lot of people think AI should handle everything end to end. That is not how real businesses work.

Real example: AI called a lead about a specific machine. The machine was not in inventory. AI could understand the request. But the human team knew something the AI did not: that machine would be back in stock in 3 months.

If that insight dies in a human conversation, the lead is lost.

So the system should capture it. The follow-up should be scheduled automatically. In 3 months, AI calls again. If the lead is still interested, the call transfers to sales with full context.

That is the real workflow. Not “AI replaces everyone.” More like: AI captures, remembers, follows up, and brings the human back at exactly the right moment.

Trap 5: The Agency Ceiling

This one is for agencies.

SEO. Meta ads. Lead generation. Landing pages. CRO. Most agencies are still selling the top of the funnel: traffic, leads, forms, ad campaigns, WhatsApp clicks.

Your client does not actually want leads. They want revenue.

You can generate a great lead, and the client can still lose it. They reply late. They forget follow-up. They do not qualify. They do not call back. They do not update the CRM. Then they blame the campaign.

That is the ceiling. The agency did its job. The client still says “these leads are not converting.”

The next big agency upsell is not more traffic. It is the conversion layer after the lead arrives.

What changed inside Toingg

We stopped thinking like a pure software company and started thinking like an outcome company.

The question shifted from:

“How do we give customers AI tools?”

to:

“How do we make sure the lead actually moves forward?”

That changed everything.

Toingg now connects with every place leads already come from: Meta ads, website WhatsApp, landing pages, inbound calls, Instagram DMs, chatbots, CRM, outbound (when the business knows its ICP).

Then it creates one shared memory across the entire journey:

• Lead talks on WhatsApp? Remembered.

• Lead gets an AI phone call? Remembered.

• Call transfers to a human? Context preserved.

• Human adds a detail? System updated.

• Lead needs follow-up in 3 months? Scheduled.

• Lead becomes hot? Quick AI call triggered.

• Customer needs human help? Takeover allowed.

• Salesperson picks up? They already know what the customer wants.

No repeated questions. No lost context. No “Sir, please explain again.”

That detail matters more than people think. When a customer has already explained their need to AI, repeating it to a human feels insulting.

In Toingg, when a call transfers, the salesperson sees the three things that matter immediately. For a machinery buyer: what machine, where to deliver, when.

The human enters with context, not confusion.

The proof that changed our conviction

For Maputo Rides, we helped recover 34 dormant drivers. At roughly $600 LTV each, that is around $20,400 in recovered lifetime value from drivers already sitting inside the system.

For TL Pathak Group, we helped generate 600 hot leads. At roughly $4,000+ per hot lead, that is around $2.4M+ in potential hot-lead opportunity value.

Neither was magic demand from zero. The demand was already there. The leads were already coming. The problem was leakage.

Once you see this, you cannot unsee it.

Most businesses are not failing because they have no leads. They are failing because their lead handling is broken. They spend money to generate demand, then lose that demand after the click, the call, or the WhatsApp message.

Fixing that leak is more valuable than adding another dashboard.

Why this is bigger than Toingg

Look at where the AI market is going.

The biggest AI companies are not just selling models anymore. They are moving toward deployment. Implementation. Forward-deployed teams. Consulting partnerships. Workflow ownership.

Why? Because everyone is realizing the same thing:

AI value does not come from access to intelligence. AI value comes from intelligence deployed inside a real workflow.

Enterprises can afford forward-deployed engineers, consultants, and transformation teams.

SMBs cannot. They do not need a 6-month AI transformation project. They need a simpler promise:

“Here are my leads. Help me convert them.”

That is the gap we are building for. Outcome-as-a-Service for SMBs.

And agencies can become a powerful part of this, because agencies already sit close to the client’s growth problem. You already run their ads, build their landing pages, improve their SEO, generate their leads.

You can go one layer deeper and help them convert what arrives.

That is a much stronger position. Not “we got you 500 leads.” Instead:

“We helped you recover the leads you were already losing.”

The new playbook for agencies

If you run an agency, here is the shift to think about.

Stop selling only traffic. Start selling the full revenue path. Not in theory. In operations.

The new agency stack should answer:

• Where did the lead come from?

• Was the response fast enough?

• Was the conversation qualified?

• Was the lead hot, warm, or cold?

• Did AI call them?

• Did the call transfer to sales?

• Did the salesperson get context?

• Was follow-up scheduled?

• Did the lead get stuck?

• Did anyone recover it?

• Did the CRM update automatically?

• Did the business actually get an outcome?

That is the difference between a lead generation agency and an outcome partner.

The next wave of agencies will not compete on who generates the cheapest lead. They will compete on who converts the highest percentage of existing demand.

The 7-day lead recovery checklist

Audit your client or your own business in a week.

1. Map every lead arrival point. Meta ads, SEO, landing pages, WhatsApp, inbound calls, Instagram, chatbots, marketplaces, referrals. Most businesses do not have one clean view. That is the first leak.

2. Measure response time. 30 seconds? 5 minutes? 1 hour? Next day? Intent peaks at the moment of reach-out. Every delay costs you.

3. Audit qualification quality. Robotic questions, or a real conversation? Bad qualification feels like a form. Good qualification feels useful to the customer.

4. Test memory across channels. Lead starts on WhatsApp, then a phone call, then Instagram. Does your system remember everything, or does every channel start from scratch?

5. Define human intervention points. When should AI stop and ask for help? When should it transfer? When should sales get notified? No human layer means AI eventually gets stuck.

6. Check follow-up discipline. Does every warm lead get followed up? Does every “call me later” actually become a scheduled follow-up? Most leads die after the first conversation, not during it.

7. Find the stuck points. Before qualification? During transfer? After human handoff? Inside CRM? After quote? If you cannot see where leads die, you cannot fix the system.

Who this is for

We are opening this to agencies and businesses that already have demand.

This is for:

• SEO agencies with clients getting inbound traffic

• Meta ads agencies generating WhatsApp or landing page leads

• Lead-gen agencies tired of being blamed for poor follow-up

• CRO agencies that want to own the post-click layer

• SMBs already getting leads from ads, SEO, calls, or WhatsApp

This is not for everyone. Please skip if you are still idea-stage, if you do not know your ICP, or if you expect AI to magically create demand from zero.

If you or your client already gets 10+ leads/day, and you know leads are leaking because of slow response, poor follow-up, messy CRM, or weak qualification, this is exactly the problem we solve.

We work with agencies as implementation partners. You bring the client relationship. We bring the conversion system, AI-first CRM, voice and WhatsApp infrastructure, human-in-loop workflow, and dashboard.

Once outcomes show up, we structure the partnership (commission, revenue share, implementation, white-label) in a way that works for both sides.

We are on a mission to save $1B in leaked leads by 2027. We are at $7.78M so far.

The bottom line

Software is not dead. Software that stops at the dashboard is.

AI tools that automate but do not own the outcome will struggle.

SMBs do not want another login. Agencies do not need another reporting tool. Businesses want revenue from the demand they already created.

That is why I believe Outcome-as-a-Service will eat software. Not because software is useless, but because outcomes are what customers wanted all along.

We learned it the hard way. Two years of watching customers struggle with configuration. CRMs becoming graveyards. Sales teams forgetting follow-ups. AI getting stuck without human ground truth.

The missing layer was never intelligence.

It was ownership.

One question before you go: What is the single biggest leak in your current lead flow? One sentence is enough.

I read every comment. Best ones get a free 30-min audit call.

Leave a comment

Want to plug the leak?

Want the free Lead Leakage Audit Checklist? Comment LEAKAGE below.

Want to try Toingg yourself? Get Toingg (self-served) →

Already getting 10+ leads/day and want us to audit where they leak? DM me AUDIT, or book a call with a Lead Recovery Expert →.

We will show you where the money is getting lost before you spend more money generating leads.

They have a workflow problem.

Too many smart people are still being paid to do things that do not create strategy:

pulling reports, cleaning CSVs, rewriting ad copy, resizing creatives, checking spend, comparing CTRs, updating slides, and manually passing learnings from one campaign to the next.

That is expensive.

And worse, it is slow.

The companies that win in the next 24 months will not be the ones with the biggest teams.

They will be the ones that turn their teams into high-speed experimentation machines.

That is the real lesson behind Anthropic’s published example: its growth marketing function was run by a non-technical team of one using Claude Code to automate repetitive work across paid search, paid social, app stores, email, and SEO. Anthropic says this workflow used CSV-based analysis, specialized sub-agents for headlines and descriptions, a Figma plugin for creative batching, Meta Ads API access through an MCP server, and a lightweight memory system for hypotheses and results. Anthropic also says this cut ad-copy creation from 2 hours to 15 minutes and increased creative output by 10x.

That should make every CMO stop for a second.

Because this is not really a story about AI.

It is a story about removing drag.

And once you see that, the next question becomes dangerous:

Any how the question this article will answer but try yours

The shift most companies are missing

Most businesses still think AI in marketing means one of two things:

1. “Write me 10 ad headlines.”

2. “Make content faster.”

That is too small.

The real opportunity is not faster content.

It is faster learning.

Bain has long argued that experimentation is a major growth lever because structured test-and-learn systems improve marketing performance over time. McKinsey’s research also shows that the biggest AI value comes when companies redesign workflows, not when they just bolt AI onto old habits.

That means your goal is not:

“Where can we use AI?”

Your goal is:

“Where are humans doing repeatable work between insight and action?”

That gap is where your money leaks.

What large marketing teams should automate first

Do not start with brand strategy.

Do not start with storytelling.

Do not start with your best creative director’s brain.

Start with the work your team already hates.

Anthropic’s example is useful because it started with highly repetitive, API-friendly workflows: ad analysis, constrained copy generation, creative batching, campaign analytics, and experiment logging. Anthropic explicitly recommends identifying repetitive tasks connected to tools with APIs, breaking workflows into specialized sub-agents, and planning prompts step by step instead of forcing a one-shot system.

That is the blueprint.

So here is the practical version.

The AI Marketing Operating System

A step-by-step framework for big teams

Step 1: Build one source of truth

Before you automate anything, centralize the inputs.

That means one clean place where your AI can see:

• campaign name

• channel

• audience

• spend

• impressions

• CTR

• CVR

• CPL/CAC

• creative ID

• landing page

• hook angle

• offer

• result

For most companies, this begins as a CSV export or a simple warehouse table.

Why this matters:

Because most teams do not lack ideas.

They lack a clean way to ask:

• What is underperforming?

• What changed this week?

• Which hook works for which audience?

• Which creative style is dying?

• Which landing page is burning spend?

If your data is scattered across Meta, Google, HubSpot, Sheets, Figma, Slack, and dashboards, your team will stay slow no matter how many people you hire.

Rule: no AI layer before a clean data layer.

Step 2: Create an analyst agent before you create a writer agent

This is where most teams get it backward.

They ask AI to generate content before asking AI to diagnose the problem.

Bad move.

Your first agent should be an analyst.

Its job:

• ingest campaign exports

• flag wasted spend

• identify underperformers

• cluster winners by angle

• detect fatigue patterns

• summarize what changed

• produce a ranked list of actions

Meta itself documents creative fatigue as a real issue when audiences see the same creative too often, which is exactly why faster variation generation matters.

So the sequence should be:

analyze first → generate second

Not the other way around.

Because good marketing is not “make more stuff.”

Good marketing is “make the next right test.”

Step 3: Split one big workflow into specialist agents

This is where quality jumps.

Anthropic’s team split creative generation into specialized sub-agents:

• one for headlines

• one for descriptions

That matters more than it sounds.

Why?

Because constraints create better output.

When you give one agent a narrow job, it performs better.

A headline agent can obsess over:

• curiosity

• punch

• brevity

• rhythm

• emotional trigger

A description agent can focus on:

• clarity

• benefit

• objection handling

• CTA

• character limits

Anthropic specifically says this separation improved output quality and made debugging easier.

Your team should copy this logic across the stack.

Example specialist agents:

• Hook Agent → generates thumb-stopping first lines

• Angle Agent → creates different value propositions

• Proof Agent → injects case studies, stats, testimonials

• Creative Brief Agent → turns winning hooks into designer-ready briefs

• Landing Page Agent → rewrites hero, CTA, and sections to match audience intent

• Email Agent → adapts the same angle for lifecycle campaigns

• Reporting Agent → sends weekly learnings to the team

This is how you stop AI from becoming “just another copy tool.”

It becomes a system.

Step 4: Connect copy to creative production

Most teams lose speed here.

They generate copy fast.

Then they hit design bottlenecks.

That is why Anthropic’s Figma plugin example is important. Their workflow reportedly generated up to 100 ad variations by swapping headlines and descriptions across frames, turning hours of manual editing into roughly half a second per batch.

The deeper lesson:

Copy without creative automation is still a bottleneck.

So your next layer should connect AI output to design templates.

That can be done with:

• Figma plugins

• Canva workflows

• template libraries

• dynamic asset generators

• creative automation tools

Your design team should not be manually rebuilding the same ad 40 times in 3 aspect ratios.

They should be doing higher-order work:

• brand systems

• concept direction

• emotional differentiation

• visual hierarchy

• ad language that competitors cannot copy

Machines should do the swapping.

Humans should do the taste.

Step 5: Bring live channel data into the same workspace

This is where things get serious.

Anthropic described an MCP-based setup that connected Claude to Meta Ads API so campaign performance could be queried directly inside the same environment.

For big teams, this matters because it removes the most invisible tax in marketing:

context switching

One tab for spend.

Another for creatives.

Another for analytics.

Another for CRM.

Another for notes.

Another for approvals.

That kills thinking.

Your goal is one command layer where a marketer can ask:

• What wasted spend in the last 7 days?

• Which audience improved CVR?

• Which creative family is fatiguing?

• Which landing page has strong CTR but weak conversion?

• Which experiments should we kill, scale, or clone?

When your team can ask questions in plain language and get answers from live systems, meetings get shorter and decisions get better.

That is when AI stops being “interesting.”

It becomes operational leverage.

Step 6: Add memory, or your team will keep repeating itself

This is the part most companies skip.

And it is why many AI workflows look impressive once, then become useless.

Anthropic says their team logged hypotheses and experiment results so the system could pull prior learnings into new iterations, creating a self-improving testing framework.

That is huge.

Because marketing teams forget fast.

People change.

Agencies rotate.

Notion gets messy.

Slides disappear.

A winning angle from 4 months ago gets rediscovered like it is new.

This is waste.

So every experiment should log:

• hypothesis

• asset used

• audience

• channel

• spend

• expected outcome

• result

• lesson

• next action

Now your AI can do something powerful:

It can generate new ideas using institutional memory, not just internet knowledge.

That is when the system compounds.

Step 7: Keep humans on the work that actually needs humans

This part is critical.

AI should not replace judgment.

It should replace repetition.

McKinsey’s more recent work shows that companies getting more value from AI tend to pair it with clearer operating models, validation processes, and human oversight.

So here is the split:

AI should own:

• data cleanup

• performance summaries

• first-pass diagnosis

• draft generation

• creative variants

• repetitive resizing/swapping

• experiment logging

• pattern detection

Humans should own:

• positioning

• brand taste

• narrative

• ethical decisions

• audience psychology

• offer strategy

• final approval

• budget risk

• big bets

If you automate the wrong layer, your brand gets generic.

If you automate the right layer, your team becomes dangerous.

The psychology layer most teams still ignore

Now let’s get to the part that matters for actual performance.

Because tools alone do not win.

Human attention wins.

And human attention is driven by tension.

That means your AI system should not just generate “more copy.”

It should generate copy that understands why people click.

Here are the 5 psychology principles your workflow should encode:

1. Curiosity beats completeness

People do not click because you explained everything.

They click because they feel a gap.

Bad headline:

“New AI workflow for paid ads”

Better headline:

“The 20-minute bottleneck killing your ad team”

The second one creates tension.

2. Specificity feels true

Vague sounds fake.

Specific sounds earned.

Bad:

“Improve performance with AI”

Better:

“Cut ad refresh time from 2 hours to 15 minutes”

Specific numbers anchor belief. Anthropic itself used this exact kind of before/after framing in its published workflow examples.

3. People care about themselves, not your tool

Nobody wants “AI-powered automation.”

They want:

• lower CAC

• less wasted spend

• faster testing

• fewer meetings

• more pipeline

• better output from the same headcount

Your messaging should always translate feature → selfish outcome.

4. Open loops keep attention alive

Do not reveal everything at once.

Sequence matters.

Example:

• First: show the wasted spend problem

• Then: reveal the analyst agent

• Then: reveal the creative batching system

• Then: reveal the memory loop

• Then: show how this changes org design

Each answer should open the next question.

That is how good content keeps people reading.

5. Steps create completion

People trust structure.

That is why numbered steps work.

A messy idea feels risky.

A step-by-step system feels implementable.

So every internal rollout, training deck, and AI playbook should be structured as:

1. what it does

2. why it matters

3. how to use it

4. what to measure

5. what happens next

Simple wins.

What this means for large marketing teams

This does not mean you should fire your team.

It means you should redesign the team.

The future marketing org will likely look more like this:

• fewer people doing manual production

• more people doing systems thinking

• more experimentation velocity

• smaller execution loops

• more reusable playbooks

• stronger creative direction

• tighter feedback cycles between performance and brand

Bain’s work on experimentation and McKinsey’s work on AI operating models point in the same direction: value comes when organizations systematize learning and redesign workflows, not when they just add more tools.

So the real question is not:

“Can AI replace marketers?”

The real question is:

“Can your current org structure survive if another company learns 10x faster than you?”

That is the threat.

Not AI.

Speed.

A simple 30-day implementation plan

Here is the no-fluff version.

Week 1: Audit the repetitive work

List every repeated workflow in paid, lifecycle, SEO, social, analytics, and creative ops.

Circle the ones that:

• happen every week

• follow rules

• use APIs or exports

• consume senior time

• require low judgment

Start there.

Week 2: Build the analyst layer

Create one agent that can read exports and answer:

• what underperformed

• what changed

• where money leaked

• what deserves testing next

Do not build five agents yet.

Build one good one.

Week 3: Add specialist generation agents

Create focused agents for:

• headlines

• descriptions

• CTA variants

• landing page hooks

• email subject lines

Keep each job narrow.

Week 4: Connect to creative and memory

Pipe outputs into Figma or Canva templates.

Log every experiment.

Create a simple “what we learned” memory file.

Now you have the start of a compounding system.

Not just a content tool.

A learning engine.

Final thought

The companies that win will not use AI to make more noise.

They will use AI to reduce delay between:

signal → insight → test → learning → next test

That is the whole game.

Anthropic’s example matters because it shows what happens when one person stops behaving like a marketer and starts behaving like a system designer.

And once that happens, the old promise of many agencies starts to sound fragile:

“Pay us every month and we will manually make creatives for you.”

That model will not disappear overnight.

But it will be forced to move upmarket fast.

Because the busywork layer is already being eaten.

The teams that survive will sell judgment.

The teams that do not will sell labor in a market where labor is getting cheaper by the month.

And that is the open loop every large marketing team should be thinking about right now.

I thought it was because the market wasn’t ready.

Turns out I wasn’t ready.

The part I didn’t want to admit

For six months, I chased features.

“What if we add Slack integration?”“What if we build a Chrome extension?”“What if we make the UI prettier?”

Every demo, I listed capabilities.

Every pitch, I emphasized potential.

And every prospect said the same thing:

“Interesting. Let me think about it.”

Translation: “I don’t believe you yet.”

Then I stopped chasing

Instead of building for everyone, we tuned for one ground truth:

Meetings that actually happen.

We called it “Communication OS.”

Not because it sounded impressive.

Because it did one thing obsessively well:

• Call in 30 seconds

• Remember every conversation

• Follow up 5 times without fatigue

• Book the meeting or escalate

One customer. One flow. One proof point.

We tuned it until it was undeniable.

Then came the compound effect

This week: closing 4 customers.

Not because we suddenly got better at sales.

Because the product finally proved itself instead of promising itself.

One customer told another.

One case study led to three intros.

One grilled CTO became our best reference.

But that grilling…

Let me tell you about that.

Day before yesterday: the moment I got exposed

An enterprise CTO gave me 45 minutes.

(For a CTO running a 200-person eng team, 45 minutes is everything.)

I thought he wanted a demo.

He wanted due diligence.

In front of my whole team, he asked:

“What’s your cost per lead?”I gave a range. He wanted specifics.

“Show me unit economics.”I talked about ROI. He wanted a spreadsheet.

“What’s your churn data?”I said it was early. He said that’s not an answer.

“How do I prove this to my CFO in 90 days?”I fumbled. Badly.

The silence that followed was the loudest thing I’ve heard all year.

What I realized in that moment

I had stories.

I had case studies.

I had features.

But I didn’t have the receipt ready when he asked for it.

And he could see it in my face.

The question that broke me:

“If I buy this and it works, how do I show my board that it was worth it?”

I gave him some vague answer about “pipeline impact” and “time saved.”

He nodded politely.

But we both knew I didn’t have the answer.

Here’s what I did the second that call ended

I told my team:

“We’re building an internal dashboard. Customer-wise metrics. Unit economics. Usage data. Cost per lead. ROI timeline. Live. I want it by tomorrow.”

Not because I was angry.

Because I realized something:

That CTO just wrote my entire landing page for me.

The gift hidden in the grilling

Every question he asked = a question every prospect has.

Every time I fumbled = a place where we’re losing deals.

Every pause = a gap in our positioning.

He didn’t just grill me.

He gave me the exact blueprint for closing every future enterprise deal.

What I’m changing now

1. Landing page revamp:Every question he asked gets answered in the first fold.No “let me get back to you.”No “we can discuss that later.”The CFO’s question? Already answered before the demo.

2. Internal dashboard:Customer unit economics, live.Cost per lead, ROI per campaign, containment rates.If a prospect asks for data, I screen-share the dashboard—right there.

3. Sales deck restructure:Lead with numbers, not narratives.Prove ROI in slide 3, not slide 23.Show the receipt before they ask for it.

4. Zero fumbling rule:If I don’t have the data to answer a question confidently, we don’t book the enterprise call yet.Readiness > speed.

The twist

That CTO?

He’s ready to close.

Big enterprise deal.

Not because I impressed him.

Because I stopped pretending and started preparing.

He didn’t want a visionary.He wanted someone who could hand him a spreadsheet his CFO would approve.

And the grilling forced me to become that person.

The compounding truth

It took 6 months to close the first customer.

We’re closing 4 this week.

Not because we’re suddenly better at sales.

Because we stopped selling and started proving.

Here’s what I learned:

Compounding doesn’t happen when you chase every customer.

It happens when you tune for one case so deeply that the rest become inevitable.

We didn’t scale by adding features.

We scaled by removing every reason to say no.

The scar I’ll carry forward

I never want to fumble in front of a CTO again.

Not because it hurt my ego.

Because it wasted his 45 minutes.

He gave me his time.He gave me his questions.He gave me the blueprint.

The least I can do is show up prepared next time.

For founders reading this who are where I was 6 months ago:

If you’re chasing features, stop.

Tune for one undeniable outcome instead.

If you’re losing deals in the final stretch, listen to the questions prospects are asking.

They’re telling you exactly what’s missing.

If you’re getting grilled and fumbling, don’t defend yourself.

Build the dashboard they’re asking for. Then come back.

The truth no one tells you:

Sales isn’t about being impressive.

Sales is about being undeniable.

You don’t close deals with potential.You close them with proof.

And sometimes, the proof only gets built after someone forces you to admit you don’t have it yet.

Field Notes @ 2:07 a.m. (Day 187)Feeling: humbled and ready at the same timeFlaw: I didn’t have the numbers when the CTO askedFix: built the dashboard, rewrote the landing page, prepared the receiptProof: CTO is closing, 4 more deals this week, compounding is realNext: never fumble again—readiness over speed

The grilling wasn’t punishment.It was the gift I didn’t know I needed.

Now I’m ready.

Walk with me.

Proof over pitch. Always.

Subscribe now

Why I’m Writing This at 2:07 a.m.

I’ve been staring at this blank page for twenty-three minutes.

Not because I don’t have things to say. Because I have too many scars to choose from.

But here’s the one that matters:

I waited seventeen minutes to make a call and lost a $30K deal.

That’s when I stopped pretending I could be everywhere.

What this space is (and isn’t)

This isn’t a business newsletter.

It’s Field Notes from someone who bleeds in public and then shows receipts.

You’ll find:

• The 2 a.m. moments when imposter syndrome is loudest

• One mistake I made this week (named plainly)

• One thing I fixed (usually small, always honest)

• One number that moved (no fluff, just proof)

• An invitation to walk with me

What you won’t find:

• 47-point frameworks

• “Crushing it” energy

• Advice I haven’t bled for first

Why now

Because the gap between what I feel and what I post was making me sick.

I built a Voice AI company. ₹150K ARR. Bootstrapped. Growing.

On LinkedIn, I look like I have answers.

At 2:07 a.m., I’m still asking: “What if this all falls apart tomorrow?”

This space is where those two versions meet.

The one who shows up. And the one who’s scared it won’t matter.

What to expect

Every week, you’ll get:

📬 Field Notes (Wednesday nights)
→ One feeling, one flaw, one fix, one proof, one invitation
→ 3-5 minute read, short lines, lots of white space

📊 The Receipt (occasional)
→ Behind the ₹55K pipeline from ₹50 in credits
→ How 128 silent drivers came back
→ The actual templates, flows, and KPIs I use

💬 Kitchen Table Talks (when something breaks me open)
→ The messy middle
→ The conversations I’m scared to have
→ The truths that only come out at midnight

Free subscribers get:

• Weekly Field Notes

• All the vulnerability, all the proof

Paid subscribers get:

• The templates and flow diagrams

• KPI sheets and dashboards

• First access to tools I build

• Monthly 1:1 voice note exchange (I actually reply)

The kind of community I’m building

This is for you if:

• You’ve stared at a lead list at midnight and felt tired

• You’ve lost deals because you were just... late

• You want proof over pitch

• You’re doing everything right except being everywhere at once

• You appreciate honesty more than hype

This isn’t for you if:

• You want rah-rah motivation

• You need someone to tell you you’re already perfect

• You’re allergic to numbers and receipts

I’m looking for the people who bleed in public and still show up tomorrow.

The ones who admit mistakes. The ones who share receipts. The ones who walk even when they’re scared.

What I promise you

1. I will always name the scar before offering the fix
   No “I scaled to 7 figures” without the “I cried in my car three times”

2. I will show receipts, not screenshots
   Real numbers. Real context. Real lessons.

3. I will write when it’s raw
   Most of these notes happen between 2-3 a.m. because that’s when the truth comes out.

4. I will invite, never pitch
   If something I built can help you, I’ll share it. But you decide. Always.

5. I will keep the seat warm
   Even when this grows, even when I’m tired, I’ll remember you’re a person, not a metric.

The first scar I’ll share

Last week, I lost a ₹2.8L contract because I was “too honest” in the pitch.

The client said: “We love your energy, but you shared too many doubts. We need someone more confident.”

I spent two days wondering if I should start lying.

Then a founder texted me: “I signed with you because you told me what wouldn’t work. Everyone else just promised the moon.”

So here’s my bet:

Honesty compounds slower than hype.
But it compounds longer.

And I’m building for the long game.

Walk with me

If you’re here, you already know something’s broken in the way we do business.

Too much noise. Too many features that don’t move numbers. Too many people pretending they have it all figured out.

I don’t have it all figured out.

But I have:

• ₹55K pipeline from ₹50 in credits

• 128 drivers who came back from silence

• ₹150K ARR built on consistency, not cleverness

• A lot of 2 a.m. notes about what actually works

If that sounds like the kind of thing you want to read—
If you want to see the messy middle, not just the highlight reel—
If you believe proof matters more than pitch—

I kept a seat warm.

Subscribe now