In some of the companies I talk to, the CEO burns more tokens than the entire IT department.

Seriously. Claude open all day. Reports, emails, analyses - automated. And IT? IT stands still and says: “that won’t work here.”

That picture is the whole story of AI adoption in 2026 in one sentence. Individual adoption happened. Organizational transformation didn’t.

Two numbers. JetBrains surveyed developers in January 2026: 90% use AI regularly. DX, which measures engineering work across 400+ companies, sees 93% tool adoption. And throughput, measured in pull requests? Median gain: under 8%.

Everyone is using AI. Companies ship almost exactly as much as before. I call this the 93/8 Gap.

For almost a year I’ve been running a training on how to close that gap. The material comes from one place: Open Mercato. We’re building an ERP-class system there where not a single line of code was written by hand. 100+ contributors, 1,200+ GitHub stars, and hand-written code is treated as a bug.

But this text isn’t about code. It’s about what follows from it if you run an agency or a software house. Because you have two problems at once: you have to switch your own company over, and you have to be able to walk a client through the same thing. So for each concept you get two things: what to do at home, and what to sell to the client.

1. The Law of the Belts

The first time I heard this story it was a proper mindfuck.

At the start of industrialization, a factory was powered by one giant steam engine. From it, line shafts and belts ran to every floor, to every machine. Then electricity arrived. And what did factories do? They ripped out the steam engine and put a giant electric motor in the exact same spot. The belts stayed.

It took about 40 years before a small motor at every machine became the standard. Only then did productivity jump. The economist Paul David wrote this up in 1990 to explain why computers weren’t showing up in the productivity statistics. Now the story is playing for the third time.

Companies swap the steam engine for an electric one, one to one. Copilot inside the old process. A chatbot on top of the old workflow. The same two-week sprint, just with AI.

2. Customer Zero

Before you demand anything from your people, show your own numbers.

I took one marketing process from 8 hours a week down to an hour and a half. Same work, exactly the same. Claude, Perplexity, a few flows wired together. And when I tell the team about it, I’m not quoting someone else’s report. I went through it myself.

The inverted adoption curve - CEO uses more than IT - isn’t just a symptom. It’s an asset. A board that automated its own work has credibility.

3. The Alibi Test and the One-Person Rule

When is change a priority, and when is it an experiment?

Three questions.

1. Is AI named in the company strategy?

2. Is there one sponsor on the board who will hold an umbrella over the team?

3. Does someone have the agency to act without waiting for a committee?

And now the twist. Say the sponsor is there, the budget is there, the umbrella is there. It’s still not enough. I’ve seen companies with full board backing sit still for six months because they had nobody to put in charge. And companies with zero budget that moved in a month because that person was already sitting there.

The bottleneck isn’t budget or technology - it’s the person who combines business sense, an understanding of what AI can actually do today, and the agency to act.

4. The Three Waves

The order in which AI enters companies is predictable. Almost nobody draws the map.

1. Code. It went first because the feedback loop is short, the output can be checked, and the person using the agent is the same person who can judge it.

2. Document-heavy back office. Invoices, claims, applications, contracts, onboarding. High volume, clear input and output structure, a countable cost per case.

3. Customer-facing. High risk, hard to verify, ROI impossible to compute.

Half of failed rollouts start on the wrong wave. The company goes for a customer chatbot and six months later has “proof” that AI doesn’t work.

5. The Pirate Ship and the 10% Rule

The board wants it, the wave is chosen. How do you roll it out? I see two patterns at clients.

The pirate ship

A 400-person IT department is a tanker - it turns over kilometers. The CEO says: “I can’t turn the tanker, so I’ll launch a small boat alongside it.” A team 10-20% the size of IT, all volunteers, reporting straight to the sponsor. The old IT stays the source of truth. The pirates build on top.

I know this pattern from another era. Fifteen-odd years ago at Divante we were rolling out e-commerce for the largest retailers in Poland. Nobody in a big company wanted to build the online channel. Sales saw cannibalization, IT saw one more system to maintain, logistics saw parcels instead of pallets. Smart boards did one thing: they created a small, independent e-commerce unit on the side. That unit delivered 5% of sales, then 10%. And only at 10% did it get absorbed - because by then nobody was asking whether online made sense.

Transform IT from the inside

The CEO explains to the department why it makes sense. The people doing the work know it better than their bosses. So they can argue, convincingly, why it won’t work. A million arguments, each one reasonable on its own.

The pirate ship has things in production after three months and internal customers coming back for more. The inside-out transformation has a series of meetings and an AI policy document.

No Spec, No Feature

Now the argument you’ll hear from every engineering team: “AI produces more code of lower quality, and review can’t keep up.”

How much truth is in that? A lot. And don’t try to deny it. Faros AI telemetry from 2026: 98% more pull requests, 91% longer review times, and no significant change in delivery pace at the company level. The METR study: experienced developers with AI were 19% slower - and after the study they still believed they’d been 20% faster.

We automated code generation and left testing in 2019

How do you tackle it? The constraint isn’t the model, it’s specification discipline. At Open Mercato every feature has a spec in the repo. A feature without a spec doesn’t exist - automated triage raises an alarm when a change arrives without one. We have 100+ pull requests open at any time. At first that hurt. Then we counted how many of them should never have been merged, and it stopped hurting.

A junior with a good spec writes architecture-aware code. A senior without a spec writes AI spaghetti, just faster.

And one principle that changes the conversation with every CTO: code safety comes from WHAT you check. Not from how the code was produced. You’re a CTO with thirty mid-level developers? You don’t read every line they ship either. You check architecture, security, everything that touches the core.

7. The 10% Mirror

Monday, 9:00, a room full of people saying “that won’t work here.” What do you do?

Write down YOURSELF the list of everything that could go wrong. Before the team does. We do this at every meeting with clients’ IT departments. We say: “let’s not talk about transforming all of IT, let’s list what could fail.” And every single time the key sentence drops: “finally, someone reasonable.” You’ve moved from “who’s right” mode to “let’s see” mode.

Mirroring

Take 10% of real work - customer tickets, for instance. The same ticket is resolved by a human and by an agent. Nothing goes to production; the AI result is set aside. The human compares and clicks: green or red. One month of data collection.

What usually comes back? AI cleanly handles about 30% of cases. And there’s a subset it does brilliantly. That’s where you dig.

But notice what happened structurally. The defenders of the status quo became the authors of the data. It’s no longer your idea - it’s their result. IT walks into the boardroom with numbers instead of vibes.

8. Three Baptisms and Two Stories

I once heard a framing that sold me instantly. There are three ways to convert an organization to AI, and they map onto the history of Christianity in Europe.

1. The Gospel: you walk around preaching that this is the moment.

2. The Baptism of Mieszko - our first duke of Poland, who converted in 966 because it simply paid off: the new faith brings bonuses, promotions, visibility ;-)

3. Fire and Sword: the mandate, no volunteers.

Companies that win use all three. But in order. The Gospel in the proof phase. Mieszko’s baptism after the first wins. The mandate last, when there’s proof, payoff and a beaten path.

What you do on Monday

Eight concepts. Don’t implement eight. Implement three first.

1. Customer Zero. One weekly task, one assistant, one number. This week.

2. The Pirate Ship. Three people pulled out of projects, one process in production within a quarter. Don’t dissolve it before 10%.

3. The 10% Mirror. On your own support. One month. Then you show the client the same numbers.

And one question to close, the same one I ask every board. Factories had electricity for 40 years before someone took the belts down from the ceiling. They weren’t short on technology - they were short on the nerve to rebuild the floor.

You have models, agents and a budget today. Which belts in your company are still hanging from the ceiling?

Find one. Take it down this quarter.

Every concept here comes from my training “How to break your team’s resistance to AI SDLC” which is part of AI Tech Leaders Training (in Polish only for now).

Sources for the numbers: JetBrains AI Pulse (01.2026), DX (400+ companies, 2026), Faros AI via arXiv 2605.01160 (05.2026), METR (2025), Paul A. David, “The Dynamo and the Computer” (1990)

Subscribe now

Between 2018 and 2024, 50% of M&A deals in tech services were about buying developer capacity (Zeema report, “AI as a Valuation Filter in Tech Services”, 2026). That era is over. It will never come back.

If you run an agency, a software house, or any professional services firm - this article is for you. I’ll show you what’s actually dying, what survives, and a concrete path from one to the other.

I’m not guessing. I co-founded Divante and Callstack (two solid exits). I wrote a book about the first 5 years of building a tech company. 65,000 copies sold. This is the playbook I use myself, right now, building Open Mercato.

Let’s go.

Your Revenue Is Lying to You

Adil Husain wrote a brutal piece in June 2026 (”The Offshore Labor Arbitrage Is Over”). One sentence stuck with me:

A billed hour and a licensed product look identical on the scoreboard.

Pakistan reports $3.8B in IT exports. Sounds great. But the export ledger doesn’t distinguish between selling software and renting people. Systems Limited, Pakistan’s largest listed IT company, spends less than 0.1% of revenue on R&D. It’s a body shop with a tech label.

Now look at your own P&L. Same question. How much of your revenue is IP - and how much is rented time?

Here’s why this matters. AI compresses the value of codifiable work everywhere, not just in expensive markets. When a workflow drops from 100 hours to 2 hours, the savings from moving those 2 hours offshore collapse. The arbitrage disappears - not because offshore workers got worse, but because there’s nothing left to arbitrage.

Husain’s data: entry and mid-level knowledge-work postings in high-income economies fell 14-41% from 2022 to 2024. McKinsey (2025) found 51% of organizations already reducing entry-level roles. HFS Research declared in April 2026 that the labor arbitrage model “passed its shelf life.”

Infosys is the clearest proof. In FY23 they hired 50,000+ graduates. In FY24: 11,900. A 76% collapse - with stable revenue. Same output, structurally smaller junior layer.

That demand cut doesn’t stay at Infosys. It travels down the supply chain. To every vendor, every staff-augmentation firm, every freelancer. Including you.

Lesson: if your business model is “smart people, cheaper than the client’s own”, you’re selling exactly what AI is deleting.

The 30% Valuation Haircut Nobody Wants to Discuss

Back to Poland for a second, because it’s a perfect lab for this shift.

Software Mind - sale paused. STX Next, Intive - funds stuck with assets bought at boom prices. The polite word is “unfavorable market conditions”. The honest word is: nobody pays for capacity anymore.

Michał Przybylski (ex-Deloitte, now Zeema) says it straight: 2025 saw nearly the highest global valuations in six years - but deals are driven by access to AI specialists, data, and unique competences. Not headcount.

And yet Callstack sold at ~130M USD in exactly this market. Why?

Five assets:

1. World-expert position in React Native - one co-founder, Mike Grabowski, was a core creator of the framework. You can’t buy authority.

2. Above-market margins - proof of unique value, not a red flag.

3. Own technology powering the services- open-source libraries and tools, not just hours.

4. Succession done before the sale - founders moved to the board years earlier. The company passed the vacation test.

5. Built to Sell from day one - clean financials, auditable P&L, a platform for further acquisitions.

Notice something? None of these five assets is “we have 200 great developers”.

Lesson: buyers don’t pay for people. They pay for what remains when the people go home - IP, position, and a company that runs without you.

The Bitter Lesson Is Coming for Your Process Slides

Ethan Mollick wrote a piece in July 2025 (”The Bitter Lesson versus The Garbage Can”) that every agency owner should read twice.

Quick version: Organizations are messy “garbage cans” - undocumented workflows, informal networks, processes nobody understands. Classic consulting sells the mapping and untangling of this mess.

But AI research learned a “Bitter Lesson” (Richard Sutton, 2019): encoding human expertise into systems loses to raw compute plus general learning. AlphaZero beat chess with zero chess knowledge.

If that lesson applies to work, companies won’t need their processes mapped. They’ll define good output and let AI find its own path through the chaos.

Scary? Yes. But read it again from the other side.

If process knowledge becomes worthless, what’s left? Knowing what good output looks like in a specific domain. Owning the examples. Owning the evaluation.

Which brings us to the most important business concept of 2026.

Owning the Hill: The New Definition of IP

In June 2026, Satya Nadella wrote that companies must turn “workflows, domain knowledge, and accumulated judgment into AI systems that improve with each use” - with private evals measuring what actually matters to the business. He calls this loop “a hill climbing machine” - the new IP of the firm (MBI Deep Dives, “Owning the Hill”, June 2026).

Two examples make it concrete.

• Fin (formerly Intercom) replaced frontier models with its own Apex 1.0, trained on proprietary support data. Result: higher resolution rates, 65% fewer hallucinations, dramatically cheaper. Salesforce acquired them days after the announcement.

• Cursor built CursorBench - a private benchmark from real user requests - and trained its own Composer models against it. Composer 2.5 scores 63.2% for $0.55 per task. Opus 4.8 at max effort barely beats it while costing ~14x more. SpaceX just bought Cursor for $60B.

The pattern: private eval + proprietary data + compounding loop = a moat no frontier model release can erase.

And here’s the flip side for you as a buyer of AI. If you rent intelligence, the provider controls the meter. A 2026 paper ”Test-Time Compute Games” shows providers are financially incentivized to burn more compute on your queries even when it adds nothing. Without your own evals, you can’t tell. You’re just guessing.

Most agencies have no evals, no benchmarks, no proprietary data strategy. They have “we’re really good at Next.js”.

Lesson: in the AI era, your moat is not what you know. It’s what you can measure, that nobody else can.

The K-Framework: From Renting Time to Owning Assets

OK, diagnosis done. Now the path. This is the core thesis of my book “The 5” and the K-Framework - the model I used at Divante and Callstack, and that I’m running again at Open Mercato.

The idea is simple: services give you cash flow and market context. IP gives you valuation and a moat. You need both - in the right sequence.

Nine steps:

1. Find the Tornado. A market or technology growing so fast that demand forgives your mistakes. Don’t guess - watch where clients already search.

2. Take the expert position. Content, open source, talks. Authority can’t be bought - Callstack and Divante proved it.

3. Sell services. Cash flow plus something more valuable: a front-row seat to real client problems.

4. Observe repeating needs. Your projects are a market radar. What do clients ask for again and again?

5. Build the IP. A tool, framework, or dataset solving that repeating need. Start small - a “quick drop” on GitHub.

6. Monetize through expert services. At Rigby (my portfolio, Medusa.js ecosystem), their own marketplace framework lifted service rates by 40% in months.

7. Reinvest. High-margin services fund the product. A positive loop.

8. Spin off. Separate product from services when the product is ready to stand alone.

9. The Golden Moment. Succession, sale, or scale - from a position of strength.

The realistic budget for building world-class IP this way: 1,5-3M USD over a few years. Sounds like a lot? You’re funding it from service margins, not VC money. That’s the whole point.

And the AI update to this framework: your step 5 IP should now include **the hill**. Private benchmarks from your niche. Proprietary datasets from client work. Evals that define “good” in your vertical. That’s exactly what Fin and Cursor did - at a scale you can copy in your niche.

At Open Mercato we run this play in real time. ~11 people doing what used to take ~30. The IP is not “AI features” - it’s the framework, the specs, the domain models that make AI code safely. We’re not an AI company. We’re the harness around the AI.

Kill the Hourly Rate Before It Kills You

One more thing, because IP without pricing change is half a strategy.

Markus Hartmann (”Preisfindung in Agenturen”, 2024) nails the structural problem: hourly billing rewards slowness. Finish a project in 2 days instead of 5, and you earn 60% less for the same client value.

Pre-AI, this was a design flaw. Post-AI, it’s a death sentence. Compress a 5-day project to 5 hours with AI, and hourly billing deletes 95% of your revenue. You literally pay a penalty for being good.

Hartmann’s fix, which I fully endorse:

• Price the solved problem, not the hours. Value-based pricing.

• Always three options (A, B, C), priced roughly 1:2:4. The client’s decision becomes “which one”, not “yes or no”.

• Zero discounts. A client who negotiates already wants to buy. Change scope, never price.

• Name the price only after value is clear. Seven-step value conversation first. Early prices anchor low and kill deals.

The market data backs this up: 73% of consulting clients prefer outcome-based pricing, and Gartner predicts 70% of businesses will prefer usage-based pricing by end of 2026.

Lesson: AI makes you faster. Hourly billing makes faster = poorer. Fix the pricing or the efficiency gains go 100% to your clients.

What Survives (And What You Should Build)

Put it all together and the picture is clear.

What dies:

generalist capacity, staff augmentation, hourly billing, “we build any website for anyone”, the affordable mid-level generalist - onshore and offshore alike.

What survives

per Husain and per my own exits:

1. Judgment with liability attached. Someone signs off. Someone is accountable. AI can’t take the blame - you can. That’s worth premium money.

2. Domain asymmetry. Knowledge of a specific market a Silicon Valley team can’t easily acquire. Vertical software built on it, priced in dollars, sold globally.

3. Proprietary compounding loops. Your data, your evals, your hill. Improving with every client engagement.

Notice this maps exactly onto the K-Framework.

Services → observation → IP → premium pricing → spin-off.

The framework didn’t change. The clock speed did. What took Divante a decade, an AI-native team can now attempt in 18 months.

Your Monday Morning Checklist

Enough analysis. Here’s what to do this week:

1. Audit your revenue. Split every service line by AI-replaceability. Anything a client could prompt their way through in a year - that’s your melting ice.

2. Pick one repeating client problem. The one you’ve solved 5+ times. That’s your IP candidate.

3. Start your hill. Collect the data you already produce: solutions, edge cases, quality criteria. Write your first private eval. Ten test cases beat zero.

4. Move one new client to value pricing. Three options, 1:2:4, no discounts. Just one deal. Learn.

5. Apply the vacation test. If you disappeared for a month - does the company have an asset, or just your calendar?

The window is real but short. Every report I read says the same: buyers are selective, not absent. Firms with unique IP and clear specialization sell at near-record valuations. Firms selling capacity sell at a discount - or don’t sell at all.

You can’t choose whether the arbitrage ends. You can only choose which side of the scoreboard you’re on when it does.

So - honest question, founder to founder:

If a buyer looked at your company tomorrow, what would they see - an asset, or a group of people renting out their time?

Subscribe now

Sources: MBI Deep Dives, “Owning the Hill” (June 2026); Adil Husain, “The Offshore Labor Arbitrage Is Over” (June 2026); Ethan Mollick, “The Bitter Lesson versus The Garbage Can” (July 2025); Puls Biznesu, “AI pogrąża polskie software house’y” (May 2026); Zeema, “AI as a Valuation Filter in Tech Services” (2026); Markus Hartmann, “Preisfindung in Agenturen” (2024); Richard Sutton, “The Bitter Lesson” (2019); McKinsey, State of AI (2025); Gartner (2025); HFS Research (April 2026); Tomasz Karwatka, “The 5” and “Karwatka’s Framework” (2025).*

“AI made my team 5x more productive. And my margin went down.”

Sounds absurd? It’s happening right now to thousands of agencies and service firms worldwide.

I just read a report that explains why. “Redesigning the Agency Value Model” by Brian Kessman (Lodestar Agency Consulting). It’s written for ad agencies. But the mechanics apply to every service business: software houses, consultancies, design studios.

What we see on the Professional Services market now

Quick recap: services fund IP

In “The 5” I describe a simple path. I call it the K-Framework.

You start with services. Services give you two things: cash flow and context.

1. You learn what clients really struggle with. You see their problems from the inside.

2. Then you convert that context into IP - intellectual property. Products, tools, frameworks, data.

IP is what builds company valuation. Services pay salaries. IP builds wealth.

I know this because I lived it a couple of times. We built Divante as a services company. We used the cash and the client context to build products. Simple, right? So why do so few firms make this jump?

The Missing Map

Here’s what the report adds to my framework: a map of where firms get stuck on the way.

The authors plot every agency on two axes:

• How you define value: commoditized (”we do everything”) vs. differentiated (”we solve this one expensive problem”)

• How you capture value: hourly billing vs. scalable revenue (products, subscriptions, licenses)

Four positions emerge. Two of them are traps I see constantly - and until now, they had no names.

• Busy by Design. Where most firms start. Broad services, custom scopes, “we can do it all”. Revenue grows only by hiring. Teams are busy. Margins are thin.

• Scaling with Strain. The first trap. You replaced hourly billing with fixed fees. New contracts, new invoices. But you still sell the same custom, labor-heavy work. The report’s key line: changing pricing without changing what you sell is ineffective. The economics don’t move. You just work harder :(

• Expertly Undervalued. The second trap. You have real expertise. Clients respect you. But you still sell that expertise as effort - hours, custom scopes. Everything depends on the founder and a few senior people. Sound familiar? This is where most strong dev-shops and agencies sits today.

• Distinctly Scalable. The goal. Expertise codified into repeatable solutions. Pricing tied to outcomes. Revenue grows without adding headcount. In my language: you converted context into IP. K-Framework is working for you.

Why AI punishes firms in the traps

Here’s the mechanism, and it’s brutal.

When you sell effort, the client buys hours. AI compresses hours. So the client expects a lower price. Your perceived value falls faster than your cost of delivery.

The report calls this being “penalized for efficiency”. Every AI gain lands in the client’s pocket, not your margin.

The numbers back it up. Average agency revenue per employee: around $150,000. At Google: $1.3 million. Agency margins dropped from ~30% in the golden age to ~10% today. The average creative produces almost 5x more output than ten years ago - for the same or less pay per unit.

In “The 5” I argue AI is an amplifier, not an equalizer. It makes strong organizations stronger and weak ones weaker. This report shows the commercial side of the same law. AI amplifies your business model. If the model is “selling hours”, AI amplifies the race to the bottom.

What actually works: three examples from the report

Monks (global, London-based) rebuilt everything around one subscription: “Talent + Machine”. One annual price. People, technology, and - this is the genius part - a built-in year-over-year efficiency rate. They promise clients the work gets cheaper every year. Instead of losing AI gains in negotiations, they sell them upfront and keep the margin. And everything is paid now: training, workshops, roadmapping. No more free “advisory” as business development.

FIG (New York, mid-sized creative agency) didn’t build any product. They changed one thing: project managers stopped tracking hours and started managing scope and assumptions. Result: profitability up about 20%. From a mindset change. Their revenue per employee: $300,000-400,000. Double the industry average.

BOND (Helsinki, 55 people) built an internal AI operating system - their own workflows, templates and logic, independent of any AI vendor. Now clients ask to license it. A 55-person firm competes with organizations ten times its size. That’s IP doing the work.

The test: is it a real solution or repackaged hours?

The report gives a clean, three-part test. An offer is a true solution only if it has:

1. A business-critical client problem - something expensive if left unsolved

2. A repeatable, codified method - frameworks, workflows, tools; not “our senior people will figure it out”

3. A defined outcome - clear direction of impact, even without guarantees

Miss one, and you’re selling a service with a fancy name. And buyers will treat it - and price it - as a service.

One more shift: transparency

When you sell hours, clients demand timesheets. Fair - effort is what they bought.

When you sell solutions, transparency changes meaning. It’s no longer about tracking activity. It’s about clearly defining: the problem, the intended outcome, the method, the boundaries.

The report makes a sharp point: when clients push back on solution-based pricing, the real problem is rarely missing timesheets. It’s a missing, clearly stated objective.

What to do this week

Don’t redesign your whole company. Do this instead:

1. Place yourself on the map. Be honest. Fixed fees with custom scopes = Scaling with Strain, not progress.

2. Find your hidden repeatability. List your last 10 projects. How many followed the same sequence? That pattern is unbuilt IP, hiding in plain sight.

3. Pick one solution to codify. One problem, one method, one outcome, one price. Sell it three times before building anything else.

This is the same direction I push in “The 5”: build a company that runs and grows without you. Hours can’t do that. Hours need you. IP doesn’t.

Subscribe now

An enterprise deal never tells you it’s dead. It just keeps scheduling meetings.

Everyone is polite. Everyone is “excited”. The follow-up is always “after the holidays”. Six months later you realize you were never in a sales process. You were in a book club about your product.

This month we analyzed enterprise sales conversations from Open Mercato’s first 6 months. The pattern was brutal: every deal we lost, we lost in the first two meetings. We just found out months later.

So we wrote down five rules and put them in our internal playbook. One slide. The title says: no exceptions. This article is that slide, unpacked.

Who’s writing this

Fair question. If you’re going to read 1,500 words of sales advice, check the source first.

I’ve been selling software for over 20 years. I co-founded Divante, an e-commerce company we exited at $65M. I co-created Callstack, sold to Viking Global for $130M. Vue Storefront (now Alokai) went through Y Combinator and raised a $40M Series A. I’ve invested in more than 40 startups, including ElevenLabs. I wrote “The 5”, a book for tech founders that sold 65,000+ copies.

And I started as the worst salesman on a street market a small town in Poland. I’m not a natural. Everything below I learned by losing deals, not by winning them.

Today I’m CEO of Open Mercato - an open-source AI engineering platform for enterprise. 1,250+ GitHub stars, 100+ contributors, first client invoices in April 2026. Small numbers, fresh scars. That’s exactly why the lessons are still sharp.

Here are the five rules.

Rule 1: Economic buyer, direct

No pricing, no POC, until we present to the person who owns the budget. Never through intermediaries.

We learned this the expensive way. Our warmest early conversations were with innovation teams and “digital transformation” departments. Great people. Zero signing power. They translated our pitch to their bosses - badly, through three layers of slides. The deal didn’t die. It evaporated.

Compare that with a large insurer we work with. The board sponsor was in the room early. AI was the group’s #1 priority for the next five years. His words, not ours. That deal moved faster than anything else in our pipeline.

So after call one, we push for the CEO. Here’s the trick that works for us: a session with the founders opens that door. A CEO won’t take a vendor call. A CEO will take a founder call. If you’re a founder, that’s an asset. Use it - or hire someone cheaper to do your job.

Harsh version of this rule: if the economic buyer won’t meet you, you don’t have a deal. You have a fan.

Rule 2: Business case in money, meeting two

By the second meeting, we put three numbers on the table. Cost of the status quo. Cost of delay. Investment range. Even rough ones.

Then we ask one question: “Did we get this right?”

That question does the real work. The client corrects our numbers - and by correcting them, they take ownership. It stops being our pitch. It becomes their business case.

One client described their legacy system as “18 years old, a million workflow steps”. Every change costs them serious money and months of waiting. Once you write that down as a number, a five-figure monthly subscription stops sounding expensive. It starts sounding late.

One more thing: your champion never walks into their board alone. They carry your one-pager - the business case in their language, their numbers, one page. If you don’t write it, your champion improvises. And improvised business cases lose to “let’s wait a quarter” every single time.

Rule 3: Name the rival

Every deal has a rival. In-house build, a big suite, or doing nothing. Establish which one in discovery - and pull the right battle-card.

Here’s what surprised us: our #1 rival is not SAP. It’s the client’s own dev team armed with AI coding tools. “Why buy your platform if our people can build it with Claude Code?” That’s a fair question, and we love it. We have a prepared answer, and it’s our best conversation. We wrote a battle-card for exactly this fight before we needed it.

The rival we underestimated? Doing nothing. Status quo has 100% market share and pays no sales team. If the pain isn’t worth money (see Rule 2), status quo wins by default.

The test is simple. Can you picture the rival - a specific person, in a specific meeting? No? Then you can’t answer the objection either. You’ll find out what it was in the rejection email.

Rule 4: License from day one

The business model is part of the pitch, not the epilogue.

This one hurts, because it’s the mistake I watch open-source founders make on repeat. You’re so happy someone wants your product that you postpone the money conversation. You do a “free pilot”. You’ll “figure out commercials later”. Later never comes. What comes is procurement, in month nine, surprised that anything costs anything.

We do the opposite. Our funnel is public and boring. A fixed-price paid workshop. A paid POC of one process. Then an enterprise subscription. Every step costs money. Everyone knows the next step’s price before taking the current one.

The key move: **tie POC success criteria to the license decision.** Result delivered → subscription starts. Not “result delivered → let’s discuss next steps”. We agree on that in writing, before the POC begins.

And never let them build unlicensed. With an MIT-licensed product, this is a real temptation - the code is right there. But a client building unlicensed isn’t a client. It’s a company that already told you the price of your work. Zero.

Rule 5: Momentum or out

Ignition to paid commitment in two weeks or less. Stalling? Raise it openly with the client. A deal that moves slowly is a deal that’s already dead.

Enterprise folklore says 9-18 month sales cycles are normal. Maybe, if you’re Oracle. For a startup, that’s a death sentence with extra steps. You can’t afford to feed zombie deals. Every hour on a stalled deal is stolen from a live one.

Our best proof: at that same insurer, two months of beautiful slides produced polite nods. Then we ran a two-day hackathon on their real process, with their team. We said: “We’ll come with something built, not with slides.” That one sprint created more momentum than the entire preceding quarter.

And when a deal stalls, we say it out loud: “It looks like this stopped being a priority. Should we close the file?” Half the time this revives the deal in 24 hours. The other half, it saves us three months. Both outcomes are wins.

The one rule behind the five

Read the five rules again. They’re really one rule: sell like your time is the scarcest asset in your company. Because it is.

Founder-led sales is not a phase you suffer through. It’s where you design the sales machine that will later run without you. Every rule above exists so that one day a playbook closes the deal, not a founder. Because a company should be an asset that works - not a job that owns you.

Now do one thing. Open your pipeline - today. For every deal, answer three questions:

1. Who is the economic buyer, and have they personally seen the pitch?

2. What is the rival, by name?

3. When is the next *paid* commitment?

Any deal where you can’t answer all three is already dead. You just haven’t been told yet.

I’m Tomasz Karwatka, CEO of Open Mercato - an open-source AI engineering platform for enterprise. If this was useful, subscribe. I write about building global software companies.

Subscribe now

This is a guide for that part.

The thing nobody puts on a slide

For years we believed one rule: more output means more people. AI broke it. The leverage is now big enough that output and headcount come apart. Freeing and unsettling at the same time.

And it splits your company into two speeds. Some people shine - their work automates beautifully and they want to automate it. The intern in the corner suddenly does the work of ten. Others, by role or temperament, can’t multiply their output. A new divide opens up.

Here’s the painful ratio. 400 people, and maybe 10 jumped on the AI wave on their own. You want more to jump. So the real question: are you helping them jump, or shoving them in - and will they swim, or drown?

No framework deck shows you that tension. Hold onto it.

You can’t out-evangelize fear

There are three ways to convert an organization to AI. They map onto the history of religion (thanks to Tomasz Jabłoński from Tutlo for this one).

1. The Gospel. You walk the halls and preach: this is the moment, this is huge, step into the light. Pure evangelism.

2. The pragmatic conversion. Across medieval Europe, rulers adopted Christianity not from a change of heart but because it paid off - trade, alliances, legitimacy. The message turns transactional: do AI because you’ll earn from it - bonuses, promotions, you keep your job.

3. Fire and sword. Force. No volunteering.

The companies that succeed run all three at once. Most run only the first - and feel it isn’t enough. It isn’t.

Because the blocker isn’t a lack of belief. It’s fear about livelihood. You cannot argue someone out of that with a better slide. The logical case lands fine. The emotional one is the wall: if I optimize myself, do I optimize myself out of a job? Address that directly and early, or it runs the room.

The pirate ship vs. the tanker

When a board catches positive FOMO - “are we missing something here? Let’s check” - the energy is great. Keep it. From there, two patterns.

• The pirate ship. “I can’t change the big IT department, so I’ll build a tiny one next to it.” That boat is maybe 10-20% the size of the tanker, reports straight to the CEO, and is staffed with commandos who wanted to be there. They’re in open opposition to classic IT: “we’re pirates, we do it differently.” Old IT stays the source of truth; the pirates build the shiny apps on top. This works - honestly, pretty well.

• Transforming IT itself. Here’s the trap: the people doing the work know it better than their bosses. So they produce a thousand precise reasons it won’t work. A wave of “this fails because…, and that fails because….” Harder road. If you’re unsure, start with the pirate ship. It de-risks everything and buys proof before you touch the tanker.

The move that actually melts resistance

Walk into that hostile room and do the opposite of what they expect. Don’t sell the transformation. Say: let’s not change all of IT - that’s too much, it might fail. In fact, let’s list everything that could go wrong. Out loud.

Naming the real risks first builds trust faster than anything. The room goes, “Yes - and that won’t work either.” Then: “Finally, someone reasonable.” There’s so much hype on the market that simple honesty makes you the credible one.

Then the concrete move - not genius, it just works.

Build a bypass. Your people handle some task - say, customer tickets. You quietly route 10% of it to AI agents in parallel. Nothing goes live. Every ticket gets solved twice: once by a human, once by AI. The human picks green or red. Call it mirroring.

Run it for a month. If AI is useful in 30% of cases, you walk into the board with real numbers, not vibes - and with the subset of cases where AI clearly wins, so the next step is obvious.

Look what happened structurally. You moved the room from “who’s right” to “let’s see.” It’s a pilot. Nothing breaks. And case by case, people stop believing AI is just for the crazies. That’s the conversion evangelism couldn’t buy.

One reason IT is the sharpest example: nobody saw it coming. For years everyone in IT said “we’re the safe ones.” Their work is getting automated first. So there’s real anger in that room. Respect it. Don’t paper over it with a demo.

Hire for agency, manage with contracts

The best AI-native people rarely come through a classic funnel. They show up around the work. Run an open build - a hackathon, a public project - and watch for two days. A person who, on paper, designs UI walks over: “I’ll fix this interface, but completely differently - I’ll write an agent to do it.” On paper they do X. In front of you they do Y and Z too, with joy.

When AI hands out this much agency and teams stay small, culture and mission matter more, not less. Force the best people back through your old funnel and you lose them - they self-select out, because they had ten other doors open.

And the structure shifts from org charts to internal contracts. A small team pitches you like you’re an investor: “this fits, here’s what I’d build.” You agree time and resources for six months and one outcome - revenue, lower cost, higher satisfaction. The weekly rhythm is for unblocking, not interrogation. Trust runs high because the goal is clear and the people pitched their way in.

What to do Monday

Don’t transform your whole company - that’s the mistake that produces a thousand reasons to fail.

Pick one team. One repetitive task. Build a bypass that touches 10% of the work with nothing going live. Give it a month and one question: where does AI actually win? Then bring numbers to the board.

And stop relying on evangelism alone. Run all three at once - preach it, reward it, and eventually mandate it. The winners don’t pick one.

Then sit with the question I can’t answer for you: of your people, how many jumped on the AI wave on their own? And for the rest - are you helping them jump, or shoving them in to drown? Get that right, and most of the transformation takes care of itself.

You know what I’ve never heard a CFO say?

“We need more AI.”

Here’s what they actually say:

• „Why do we need 12 people on this project?”

• „Can we cut the timeline from 6 months to 6 weeks?”

• „I don’t care how you do it. I care what it costs and when it ships.”

And yet - most agencies walk in and lead with their tech stack. “We use Claude. We use Cursor. We built custom agents.”

Business Guys? Nobody cares.

The real product is compression

After many enterprise AI conversations, one pattern became impossible to ignore.

The buyers actually spending money on AI don’t use the word “AI” in their requirements.

They say things like:

• „We want a 3-person team, not a 15-person team.”

• „We want the first version in weeks, not quarters

• „We want to own the system, not rent your people forever.”

That’s not an AI pitch. That’s a compression pitch.

Fewer people. Fewer meetings. Fewer handoffs. Fewer months.

AI is the engine. Compression is the product. If you still sell “AI capabilities,” you’re selling the engine, not the car.

The numbers behind the shift

Deloitte’s 2025 Tech Value survey: 57% of enterprise transformation budgets now go to AI automation. Same work, fewer humans in the loop.

Gartner tracks outcome-based pricing in SaaS. It jumped from ~15% in 2022 to over 30% by 2025. Clients are moving from “pay for effort” to “pay for results.”

And the World Economic Forum projects the AI agent economy from $5.4 billion in 2024 to $236 billion by 2034. That’s a structural shift in how services get bought.

Your clients have already internalized something you might not have accepted yet.

If AI can do the work of 10 people, they will not pay for 10 people. They’ll pay for the outcome. Or they’ll build it in-house.

Those are your two options. Pick one.

So what’s an AI-proof company?

Here’s where most people get it backwards.

An AI-proof company is not one that bolts AI features onto everything. That’s the easy part. AI is becoming a commodity, and commodities don’t protect you.

An AI-proof company is built so AI can’t commoditize it.

I keep saying the same thing to founders: AI alone is not enough. Too many edge cases, too much hidden logic, too much context AI doesn’t have.

But combine AI with your own IP and human context - and suddenly you have something powerful.

Think about it. Most companies use 5-10% of the software they pay for. Absurd waste. Now picture the near future where AI rewrites just that 5-10% - custom, fit to the need, no license, no bloat.

When that lands, what survives? Not the agency selling dev hours. Not the SaaS selling seats nobody uses.

What survives is the company that owns the IP, owns the context, and gets paid for the result. Value-based pricing stops being exotic and becomes the standard. Selling hours becomes a slow death.

That’s the moat. Not “we have AI.” Everyone has AI. The moat is the IP and context AI can’t reproduce on its own.

Open Mercato: compression, made real

Let me make this concrete, because I hate abstractions without a story.

My brother Piotr built Open Mercato - an AI-first ERP/CRM framework - fully with agentic coding. He didn’t write a single line of code himself.

That’s a paradigm shift in how you ship an ERP. And it’s compression on the table, not on a slide:

• Fewer people. The team that ships this is small.

• Weeks, not quarters. The boring 80% is solved before you start.

• You own the system. It’s your IP, not a rented seat you pay for forever.

That’s the exact brief the CFO gave you. Fewer people, faster, owned. We just didn’t say the word “AI” once to get there.

Where this comes from

I’m not theorizing from a deck. After two exits from service companies I see this same story again.

The value sits in the IP and the context, not in the hours billed. Services pay the bills and teach you the market. IP builds the valuation and the moat. Companies that only sell services don’t sell for serious money.

What to do tomorrow

Take your last three proposals. For each one, answer honestly:

Did you sell people and hours - or did you sell a result the client couldn’t get cheaper anywhere else?

If it’s the first, your client will eventually do the math. AI does the work of 10 for the price of 1, and they’ll either squeeze you or build it themselves.

So the real question, the one I’d ask myself:

if you sold your company tomorrow, would the buyer see an asset - your IP, your context, your repeatable outcomes - or would they just see your team’s timesheet?

Curious where you land on this. I argue about it daily on X - come disagree with me there :).

Today I’m building Open Mercato: an open-source AI engineering framework for enterprise workflows and agent orchestration.

Over the last quarter, we’ve had 20+ conversations with enterprise teams about AI.

Two recent ones changed how I think about where enterprise AI is actually going.

Different industries. Different cultures. Different governance.

Same conclusion.

Most enterprise AI programs are failing because they’re trying to bolt AI onto legacy workflows instead of redesigning orchestration itself.

That’s the real unlock.

1. Stop inserting AI into 18-year-old workflows

This is the most common anti-pattern we see:

• Take one step out of a legacy workflow

• Replace it with an AI agent

• Push the result back into the same ancient process engine

One company had a 1,000+ person operation running on an 18-year-old workflow stack.

They automated individual steps.

Savings?

~2 FTEs per step.

That’s not transformation. That’s expensive cosmetics. The leverage is not in optimizing steps. The leverage is in collapsing the workflow itself.

The winning architecture looks different:

• Legacy ERP/workflow becomes the system of record

• A lightweight orchestration layer sits beside it

• Agents drive the actual work

That’s the shift.
Workflow replaced by goal-driven orchestration.

2. Agents don’t want playbooks. They want goals.

Old enterprise thinking: „Follow these 1 000 steps.”

New AI-native thinking:

“Here is the goal.
Here are the tools.
Here are the constraints.
Figure it out.”

The best teams we spoke with independently converged on the same pattern: A few agents instead of giant deterministic workflows.

Example:

• Agent 1 → gather evidence

• Agent 2 → validate and enrich

• Agent 3 → recommend decision

• Agent 4 → execute downstream actions

Each agent owns a goal (Not a flowchart).

The orchestration layer handles:

• auditability

• approvals

• validation

• compliance

This is the real architectural shift happening in enterprises right now.

3. The hardest part of AI engineering is not engineering

One enterprise leader said it perfectly: “The hardest part is explaining the work clearly enough for the agents.”

Clarity.

The teams winning right now are good at:

• defining goals

• splitting phases

• setting constraints

• designing guardrails

• specifying tools and context

Ironically, many traditional process analysts struggle here. Because SOP thinking is sequential.

Agent thinking is intentional.

This becomes a massive new skill:
writing operational briefs for AI systems.

4. The bottleneck moved

One of the most important shifts in software history is happening right now:

Prototype creation collapsed from months to days.

A non-engineering VP at a global company built a working internal tool over a weekend using AI coding tools.

Then central IT spent TWO MONTHS:

• security reviews

• SSO

• observability

• compliance

• enterprise architecture

• cloud migration

That’s the new bottleneck: Hardening.

This is why “vibe coding” simultaneously:

• changes everything

• and solves almost nothing by itself

Enterprise winners will not be the companies that prototype fastest.

They’ll be the companies that productionize fastest.

Huge difference.

5. AI transformation is not committee-compatible

The successful enterprise AI teams we met all shared one trait:

Direct executive sponsorship.

Not innovation theater, 14 steering committees, “AI councils.”

Real air cover.

Small elite teams.
Fast loops + Direct access to C-level leadership.

One exec told his team:

“We were second once. We never recovered. We will never be second again.”

That mindset matters more than model selection.

A 20-person AI-native team with executive protection can now outperform a 400-person traditional IT department in specific domains.

Not because they’re smarter.

Because they can move.

6. Monolithic copilots are mostly dead

Every enterprise tried building the universal AI assistant.

Almost all hit the same wall:

• hallucinations

• unreliability

• impossible governance

• unclear ownership

The better pattern is modular agents.

• Specialized.

• Composable.

• Single-purpose.

Exactly how good software systems have always been designed.

The interviewer agent.
The validator agent.
The policy agent.
The enrichment agent.

Small pieces.
Clean interfaces.
Reusable blocks.

That becomes your compounding advantage.

7. ERP systems are becoming databases with UI

This may be the most important strategic shift: ERP vendors will all ship agents.

Everyone will have them. Those won’t differentiate anyone.

Your advantage will come from the agents that encode:

• your company’s judgment

• your edge cases

• your operating model

• your leadership philosophy

• your customer nuances

ERP becomes the system of record.

The value moves above it.

Into orchestration, agents, company-specific operational intelligence.

That’s the next platform war.

8. Value-based pricing is coming for AI consulting too

One enterprise told us they changed a consulting contract from:
“hours + PowerPoints”

to:

“percentage of measurable business outcome.”

That’s the future.

If vendors can’t get systems safely into production, they won’t matter.

AI made feature creation cheap. Production-grade deployment is now the premium layer.

• Security.

• Observability.

• Governance.

• Compliance.

• Auditability.

• Operational reliability.

That’s where value migrates.

Final thought

Most enterprises mistake is that AI transformation is an automation problem.

No. It’s a re-platforming problem.

The Question is: “How do we redesign orchestration itself?”

That’s the real shift.

At Open Mercato, this is exactly what we’re building for: an open-source AI engineering framework for enterprise orchestration, modular agents, and AI-native workflows - designed to help enterprises move fast without destroying governance, security, or operational control.

The next generation of enterprise software won’t be built around screens and forms. It will be built around agents.

Disclosure: I’m an active contributor to the Open Mercato ecosystem. Not neutral. Very opinionated. And increasingly convinced this is inevitable ;)

The answers were raw. From people actually running companies, selling services, building software, managing teams, and trying not to miss the biggest shift in years.

The main takeaway?

People are not afraid of AI as a technology.

They’re afraid that the market will change faster than they can adapt.

That pattern showed up everywhere.

It’s not “what if AI doesn’t work.”
It’s: “What if it works too well - for our clients, our competitors, and everyone else?”

Fear #1: The collapse of the service business model

This was by far the strongest theme.

Over and over again, people said:

• clients will start doing more in-house,

• companies will insource what used to be outsourced,

• competition will explode,

• prices will drop,

• margins will shrink,

• Time & Material will stop making sense.

In simple terms:

Until now, you sold people, expertise, and delivery capability.
Now the client looks at AI and thinks: “Maybe we can just do this ourselves.”

And this is no longer theoretical. People shared real examples:

• lost enterprise leads,

• pressure on pricing,

• clients experimenting with “vibe coding,”

• extracting process knowledge — and then building without the partner.

Brutal, but real:

The biggest fear today is not “AI will replace us.”
It’s “will clients still pay us?”

Fear #2: We won’t keep up

The second big theme is speed.

“What if we’re too slow?”

People talked about:

• FOMO,

• tool chaos,

• no time to test everything,

• difficulty separating signal from noise,

• fear that competitors will adopt AI faster and smarter,

• building processes for a world that may not exist in 6 months.

This is a strong signal.

Companies are not ignoring AI.
They are actively trying.

But it feels like: the train has already left the station - and the platform is still under construction.

Fear #3: What will clients actually pay for?

This is the uncomfortable one.

A lot of responses pointed to a deeper shift:

Will clients still pay for:

• development hours,

• workshops,

• architecture,

• consulting,

• MVPs,

• analysis,

• expertise delivered in conversations?

If a client can “talk to an LLM,” it creates the illusion that access to answers = access to value.

It doesn’t.

But markets tend to destroy pricing first - and only later realize what they’ve lost.

That tension is everywhere: delivery time is collapsing, but pricing logic hasn’t caught up yet.

So companies are worried about:

• value-based pricing,

• fixed price vs T&M,

• price pressure,

• race to the bottom,

• clients expecting lower costs because “AI makes it faster.”

This will be one of the biggest battles ahead.

Not about technology.
About how value is priced.

Fear #4: Speed vs quality vs responsibility

Many participants are not afraid of using AI.

They’re afraid of using it without control.

Concerns included:

• hallucinations in critical systems,

• unreadable AI-generated code,

• “vibe coding” without accountability,

• shipping PoCs as production,

• broken architectures,

• hidden errors deep in pipelines,

• unclear ownership of decisions,

• lack of awareness that AI is non-deterministic,

• data privacy, security, compliance risks.

This is a big shift.

We’re no longer in the phase of:

“Wow, it wrote a function.”

We’re in the phase of:

“Who is responsible when this system starts making real decisions?”

Different game.

Fear #5: Teams may not handle the transition

This didn’t show up the most - but it was one of the most important.

People mentioned:

• resistance from teams,

• developers not wanting to rely on AI,

• split between enthusiasts and skeptics,

• inability to assess quality,

• over-reliance on AI,

• loss of independent thinking,

• cognitive overload,

• weakening problem-solving skills.

In short: the fear is not only that people won’t use AI - but that they will use it badly.

Too fast.
Too shallow.
Without understanding.
Without ownership.

AI transformation is not just tooling.

It’s a shift in how organizations think.

Fear #6: Fake content, manipulation, and loss of trust

A smaller but very sharp group raised broader concerns:

• “dead internet” scenarios,

• flood of AI-generated content,

• inability to distinguish truth from manipulation,

• fake evidence impacting businesses,

• illusion of competence,

• erosion of trust in media and institutions.

Not the dominant theme - but a very mature one.

Because it asks: “What kind of information environment will we operate in?”

What about the optimists?

They exist. Plenty of them.

Some responses were:

• “no fears,”

• “only opportunities,”

• “this is a golden time,”

• “AI is my advantage.”

But even there, one thing kept showing up:

• “we can’t fall behind,”

• “we need to move fast.”

Even optimists feel the pressure. They just channel it into action.

What this really says about the market

This is not a market afraid of technology.

It’s a market that already understands that AI is not a feature.

AI is pressure on the entire business equation:

• how you sell,

• what you charge for,

• how you defend margins,

• what your moat really is,

• which skills actually matter,

• how fast you can adapt,

• how you avoid commoditization.

That’s why these answers hit hard.

Because they expose an uncomfortable truth: Most companies are no longer asking “should we use AI?” They are asking “will our current business model survive AI?”

One sentence summary

If you had to compress everything into one line:

The biggest fear around AI in the next 12–18 months is not the technology itself, but that AI will accelerate commoditization, change buying behavior, and compress pricing faster than companies can build a new advantage.

What now?

That’s a separate conversation.

But one thing is clear: If you run an agency, software house, consulting business, or any knowledge-driven company, the biggest risk is not AI. The biggest risk is continuing to sell a version of the world that is already losing value.

And that’s not a tech problem. That’s a strategy problem.

The Numbers Don’t Lie

Look at the current landscape:

• 82% of brands have moved work to in-house agencies (up from 58% in 2013).

• 39% of CMOs plan to cut external agency spend this year.

We are seeing a massive “in-housing” trend powered by the fact that internal teams, stayed-up with AI, can now do what used to require a massive external squad.

The AI Coding Revolution

The acceleration of AI-assisted coding has been staggering. It’s no longer about “autofill” for code; it’s about a fundamental shift in production.

As the founder of WiseTech recently put it: “The era of manually writing code is over. Projects that once took 6 months now finish in a day.”

When the production cycle shrinks by 99%, the “man-hour” billing model becomes a suicide pact.

The Productivity Gap: $150k vs. $4.5M

This is the most sobering statistic for any agency owner today. We are seeing a 20-30x productivity gap between traditional firms and AI-native companies.

You cannot compete with a 30x efficiency gap by “working harder” or “improving processes.” You have to change the underlying physics of your business.

The Great Split

The agency world is splitting into two camps.

On one side, we see the giants struggling to turn the ship. Forrester projects that 15% of agency jobs (roughly 47,000 positions) will be eliminated in 2026 alone.

WPP is shedding 7,000 jobs, and the Omnicom-IPG merger is less about “synergy” and more about a $750 million desperate grab for cost savings through labor reduction.

On the other side, there is an unprecedented opportunity. The AI consulting market is projected to surge from $11 billion in 2025 to over $90 billion by 2035.

The Bottom Line

The “High Performers”, the top 6% of companies, are no longer looking at AI for “efficiency.” They are looking at it for transformative change. While 88% of the industry uses AI, only the winners are using it to scale autonomous agents and restructure their entire delivery model.

The agency world is splitting: those who see AI as an existential threat, and those who see it as a once-in-a-decade opportunity. Both are correct. The only difference is how you respond.

Are you building a faster horse, or are you building the engine?

Kill the “Man-Hour” Billing Model

The most dangerous thing an agency can do in 2026 is bill by the hour. When AI reduces a 40-hour task to 40 minutes, your revenue disappears if it’s tied to time.

• Move to Value-Based Pricing: Charge for the outcome, not the effort.

• The “Efficiency Tax”: If you don’t change your pricing, you are essentially punishing your team for being fast.

• Subscription for Outcomes: Transition to recurring revenue models based on maintained “AI Agents” or specific business KPIs (e.g., leads generated, code shipped to production).

The Rise of the “Agentic” Workforce

We are moving from “Human + Copilot” to “Human + Swarm.” In 2025, we used AI to help us write. In 2026, we use AI to execute.

• Vertical AI Integration: Stop selling “General Software Development.” Start selling “AI-Automated Financial Auditing” or “Autonomous E-commerce Logistics.”

• 75/25 Rule: Gartner predicts that by 2030, 25% of work will be done by AI alone. Your agency should be aiming for that mix today. Your “junior” staff shouldn’t be humans; they should be custom-tuned LLM agents.

Focus on “The Glue” (System Integration)

The biggest bottleneck for enterprises isn’t having AI; it’s making AI work with their messy, 20-year-old legacy data. This is where the $90B opportunity lives.

• Data Readiness: Most companies have “garbage” data. Your job is to be the specialized crew that cleans, structures, and feeds that data into RAG (Retrieval-Augmented Generation) systems.

• Custom LLM Fine-Tuning: Generic models are commodities. The value is in the proprietary fine-tuning you do for a specific client’s brand voice or technical stack.

Benchmark Against the “Top 6%”

According to McKinsey, the top 6% of AI high-performers are doing something differently: they aren’t just looking for 10% efficiency. They are looking for 90% transformation.

The Final Verdict

If you are running an agency today, you are actually a software company that hasn't realized it yet. You need to start building internal IP- proprietary tools, frameworks, and agents-that you can deploy across multiple clients. That is how you bridge the gap from $150k revenue per employee to $1M+.

Would you be interested in getting my 40+ pages of research on HOW TO BUILD ANAI-PROOF AGENCY? Let me know in the comments.

This is not a theory for me. I’ve built and sold multiple companies over the last 20 years. I’ve invested in 40+ startups in the last few years. And most of the wins didn’t come from “big trends”. They came from spotting the right wedge early.

Here’s how I actually look for them today at Catch the Tornado:

Follow real budgets, not demo videos

If it trends on X, it’s probably too late or not real. I ignore 90% of AI demos. Product Hunt? Meh… Instead, I spend time talking to enterprise buyers every week. CFOs, CTOs, operators.

Not “what are you excited about?” but… “What did you actually sign a contract for this quarter?” That’s the only signal that matters.

That’s how we ended up early in:

• AI-assisted software engineering → Open Mercato

• AI for finance workflows → Incro / CFO Studio

• AI in healthcare → Doctor Dok

No hype. Just budgets moving.

The more boring it sounds, the better it is

If it sounds sexy, it’s probably crowded. If it sounds boring, there’s money.

Nobody tweets about:

• compliance automation

• financial reconciliation

• internal procurement workflows

But companies spend millions there. The winners are not building “AI for everyone”.

They’re building: “AI that saves $2M/year in one ugly process”. That’s it.

Find the $200/hour human doing repetitive work

This is the cleanest pattern I know.

Look for:

• expensive experts

• doing predictable work

• in high-volume environments

That’s where sub-tornados form.

Examples:

• doctors reviewing cases

• lawyers scanning documents

• finance teams building reports

• engineers doing repetitive coding / QA

AI doesn’t replace them. It removes the 60% of work they hate.

And suddenly: output x3, cost down, demand UP (that last one is key — people underestimate how often AI increases demand).

No proprietary data = no moat

If your product works the same on public data… it’s a feature, not a company.

The best sub-tornados have a built-in data flywheel:

• every customer → more data

• more data → better output

• better output → more customers

That’s how you build something defensible.

That’s also why I like:

• Services + IP + AI

• AI-native workflows

• OSS + enterprise overlays + Expert Consulting

Narrow > broad. Always

“AI” is not a strategy.

“AI for X in Y industry solving Z expensive problem” - now we’re talking.

What I’m watching closely right now:

• AI-native enterprise frameworks (Open Mercato)

• Vertical finance AI (Incro / CFO Studio)

• AI-driven marketplaces (Mercur)

Notice the pattern? None of this is “next ChatGPT”.

All of this is: niche specific, boring, expensive.

My actual quarterly exercise

I literally do this:

1. List the 10 most expensive human workflows in industries you understand

2. Ask: which of these can be 3x faster with AI (not replaced, but augmented)?

3. Check: is someone already building this wedge? “Yes” is OK

That’s it. This simple exercise led me to most of the good bets I’ve made recently.

The AI tornado is loud. Sub-tornados are quiet. And that’s exactly why they’re valuable.

What’s one “boring but expensive” workflow you think is about to break?

ps. When you try to find your Tornado its crucial to know when to Pivot vs. When to Persist.

We are currently witnessing a massive collision. In the era of AI, the old distinction between “services” and “products” is evaporating. Many agencies and software houses are about to face the same decision founders always face: persist with the old model, or pivot into a new one.

The Old Service Model Is Breaking

For the last 20 years the dominant model for agencies and software companies was simple: Sell hours, grow headcount, increase billable utilization.

But AI is starting to compress the cost of producing software dramatically. Tasks that previously took days can now take hours. Sometimes minutes.

Which means something uncomfortable: The old Time & Material model starts to crack. The companies that survive won’t be the ones that just “use AI tools”. They will be the ones that rethink their operating model. This is the real Pivot.

Services and Products Are Merging

One thing I’ve learned after building service companies is that the best ones always behaved like product companies anyway.

They had:

• repeatable onboarding

• internal frameworks

• reusable components

• clear delivery playbooks

• strong definitions of quality

In other words: they productized delivery. AI is now accelerating this trend dramatically.

The boundary between services and products starts to blur.

Many service companies will begin building internal platforms. Some will spin them out into products. Others will build AI-powered delivery frameworks.

And many companies will pivot several times before they find the right model.

Signals You Should Persist

Even in times of disruption, persistence still matters. Here are signals I’ve learned to respect.

Customers who love you exist — there just aren’t enough yet

At Vue Storefront we had maybe 5-8 companies absolutely obsessed with the product before the market caught up. That was enough to start.

If a small group of customers genuinely can’t live without what you built, you probably have a distribution problem, not a product problem.

Distribution problems are solvable. Product-market fit problems are existential.

The market is growing regardless of your efforts

When we were building in the headless commerce space, the market itself was exploding. Even when we made mistakes (we made plenty) the rising tide helped us survive.

Geoffrey Moore calls this the tornado phase. When you’re inside a tornado, persist. Demand moves faster than supply.

Your team still believes

Not the polite “we believe” from all-hands meetings. Real belief.

You see it in people who keep solving problems creatively instead of just executing tasks. Great teams smell opportunity long before spreadsheets confirm it.

Signals You Should Pivot

The opposite signals are just as clear - if you’re honest enough to see them.

You’re pushing, not being pulled

When every deal requires heroic persuasion, something is wrong. Real product-market fit feels like the market is pulling the product out of your hands.

I saw this shift recently at Open Mercato. At some point clients started coming to us. Before that moment, every client meeting felt like pushing a boulder uphill.

Your best people drifting away

When A-players leave for competitors, it’s often a compensation or culture problem. When they leave the industry entirely “to try something new” - that’s a deeper signal. People close to the work often sense strategic problems earlier than founders do.

Your plan for the next year is just “more of the same”

If the strategy is:

• “Let’s just do what we did last year, but harder.”

That’s not a strategy. That’s hope, and Hope is not a strategy.

The Hardest Pivot: Ego

One of the most expensive mistakes early in my career came from ignoring these signals. Everything suggested we should pivot. No one was interested in our offer and… I just added more people to the marketing team… crazy.

Why? Ego. I had told everyone this was going to be big. I had invested years of work. Pivoting felt like admitting failure. That decision cost me almost two years and a lot of money.

Why This Matters so Much Right Now (AI Changes the Game)

This lesson matters even more today. AI is creating massive structural shifts in the software industry.

Agencies will pivot toward:

• productized services

• AI-assisted delivery frameworks

• internal platforms

• verticalized expertise

• value-based pricing models

Some will succeed. Many will pivot multiple times before they get it right. And that’s normal. Every technological shift produces waves of business model experimentation.

My Personal Rule Now

Every 90 days I ask myself three questions about any venture:

1. Are we getting more pull from the market or less?

2. Am I more excited or less excited than 90 days ago?

3. Would I start this company today, knowing what I know now?

If the answer to all three is “less” for two consecutive cycles, it’s time to seriously consider a pivot. Not panic. But be honest with yourself.

Subscribe now

AI is compressing timelines, lowering production costs, and commoditizing execution faster than most agencies are ready to admit.

When code can be generated, designs can be drafted, and workflows can be automated in minutes - billing hours stops being a defensible strategy.

Agencies are being pushed, whether they like it or not, toward value-based models.

Not “we delivered features.”
But “we improved your EBITDA.”

Not “we shipped software.”
But “we changed your operational outcome.”

This is where the path taken by Palantir Technologies becomes a powerful inspiration.

They didn’t sell software licenses - they sold impact.
They embedded into organizations.
They tied technology directly to measurable results.

For agencies navigating the AI era, that’s a serious blueprint.

At Open Mercato, we’re building with the same philosophy in mind.

Not just a framework.
Not just tooling.

But a system that helps agencies move from selling hours to delivering outcomes - with architecture discipline, platform leverage, and AI-assisted execution built in from day one.

The Palantirization of Everything

In “The Palantirization of Everything”, a16z explains something fundamental:
enterprise customers are no longer buying boxed software - they’re buying outcomes.

That’s exactly the logic behind Palantir Technologies.

And it’s the same shift happening right now in AI-native enterprise systems.

So what does this mean for agencies — and for Open Mercato?

“McKinsey with Code” Model

Palantir didn’t win just because of software. It won because of Forward-Deployed Engineers. Engineers who enter the organization and make the software actually deliver value.

Open Mercato operates the same way: elite teams, production-ready apps in weeks, not quarters + deep integration with business reality.

In a world overwhelmed by AI possibilities, people are still the bridge between technology and results.

Agencies should pay attention: consulting + execution + platform leverage is a powerful combination.

The Open–Closed Principle as Protection from the “Service Trap”

a16z warns about something dangerous:

It’s easy to become “Accenture for X” - expensive, fully custom services with no leverage.

Palantir avoids that trap because their engineers don’t build from scratch.
They build on top of Foundry and Gotham.

At Open Mercato, we follow the same discipline: strong core, modular architecture, strict Open–Closed Principle.

Customization is not rewriting the system.
It’s extending the platform.

That distinction determines whether you build enterprise IP - or just bill hours.

Agencies should pay attention: You need to own your IP. Even if it starts small - just an additional module layered on top of existing software. If it’s yours, it compounds. It becomes reusable. It becomes differentiating. That’s how you build an unfair advantage - not by selling more hours, but by owning pieces of the system that no one else can replicate.

AI as the Missing Scalability Layer

Palantir’s model is powerful. But historically, it required rare “unicorn” people - deeply technical, business-aware operators.

That’s hard to scale.

Open Mercato adds something critical: AI-assisted engineering, embedded business context, prompts and structured guardrails for agents

This multiplies team productivity 3x.

You don’t need hundreds of geniuses.
You need a system that amplifies strong operators.

AI becomes the scalability layer.

Agencies Should Pay Attention: With AI-assisted engineering, embedded business context, structured prompts, and guardrails for agents - you no longer need a room full of geniuses. Instead of praying for “10x developers”, you build a system that amplifies good people into high-leverage teams.

AI becomes your scalability engine.

High-Stakes Only - and That’s Good

Palantir operates where stakes are massive:
defense, security, healthcare, billion-dollar decisions.

Open Mercato doesn’t target the exact same sectors - but we also focus on serious, high-impact environments: HealthTech, Financial operations, Critical customer operations, R&D-heavy organizations

Enterprise-grade security, tenant-scoped encryption, architectural discipline.

For a simple CRM? Overkill.

For systems that move EBITDA and drive R&D leverage?
That’s where this model makes sense.

Agencies Should Pay Attention: Are you working on things that truly matter to your clients? Can your work be directly connected to EBITDA growth? If yes - value-based pricing becomes logical. You’re selling measurable business impact.

But if you’re building pet projects… then pricing power disappears. And worse - someone experimenting with AI and vibe coding after hours might deliver something “good enough” for a fraction of the cost.

AI is compressing the value of non-critical work.

So the real question isn’t how to charge more.

It’s this: Are you operating in the layer of the business where impact is undeniable?

If not, the market will eventually price you like a commodity.

Services as Scaffolding — Not the House

Forward deployment it just a mechanism - services are scaffolding.
The platform is the asset. If every project ends as isolated custom code, you are rebuilding the house every time.

If each engagement strengthens your core reusable modules, proprietary IP, refined architecture, AI-enhanced workflows - then every client increases your strategic position.

That’s durable value.

Question

You can feel the shift already. The market is starting to resonate with this new model.

The signal is subtle - but it’s there.

The real question is: are agencies listening? Do you?

In moments like this, founders often ask the wrong question: What’s new?

My bet: a better question is: What doesn’t change?

That’s why Same as Ever by Morgan Housel hits so hard. Technology moves fast. Human nature… not so much.

If you’re building a company in the AI era, some fundamentals are still rock solid:

Trust still compounds (faster than features)

AI can generate code, content, even whole products. But clients still buy from people they trust.

Example: Two agencies use the same AI tools. The one that communicates clearly, delivers on time, and takes ownership wins, every time.

Speed matters - but reliability wins contracts

Yes, AI lets you ship in weeks instead of months. But in enterprise (and I know this game…) “works most of the time” is not a selling point.

Example: An AI-powered logistics platform that is stable and predictable will beat a “cooler” but flaky competitor.

Simplicity scales better than cleverness

AI makes it tempting to overbuild. Resist.

Simple products are easier to: sell, explain, maintain.

Example: The best AI-native systems (this is exactly how we think about Open Mercato) start with 80% done, clean foundation, not a Frankenstein of features.

Long-term thinking is still a superpower

AI cycles will get faster. Hype waves will come and go.

But companies that win will still be the ones that:

• show up consistently

• build real distribution

• nurture ecosystems and partners

Nothing sexy. Very effective.

My takeaway

In times of massive change, your edge often comes from what is boring… and timeless.

→ Trust still wins deals

→ Reliability beats hype

→ Simplicity scales

→ Long-term thinking compounds

Build with AI - but build on human constants.

That’s exactly why one of our strongest bets at Open Mercato is… a partnership program for agencies. They’re the ones who already have the trust, the relationships, and the enterprise access we’re looking for.

Curious - what do you believe will not change in the next 10 years?

I’ve also been one unpaid invoice away from going under.
If you run an agency, you know exactly what that feels like.

Service companies power the tech world. They deliver innovation. They turn ideas into real businesses.

And yet most agencies live on:

• thin margins

• constant pressure

• and a business model that breaks the moment growth slows down

I’ve been there.

Together with my partners, I figured out a different path: service companies powered by their own IP. Not decks. Not theory. Real systems that scale, spin off products, and survive bad markets.

Now AI changes everything.

For many agencies, AI is an existential threat.
For a few, it’s a once-in-a-decade opportunity.

After the success of my book The 5 in Poland (65,000+ copies sold), I want to take this knowledge further - especially to digital agency owners who don’t want to be replaced, commoditized, or squeezed to death.

This Substack is about:

• escaping the hourly-rate trap

• building leverage through IP

• using AI as a force multiplier, not a competitor

• turning agencies into calm, durable, high-value businesses

Just hard-earned lessons from building, almost failing, fixing, and exiting.

I’m starting small. With honest writing. If you’re building an agency and want it to last in the age of AI - this is for you.

Subscribe if it resonates.
We’ll figure out the rest together.

Subscribe now

Most leaders fall into the same trap:
they build companies where they are the sun, and everyone else just orbits around them.

A recent conversation with a friend pushed me to write down 5 rules for building a self-driving organization through hiring.

1. Play above your league

Don’t hire people you can “afford.”
Go after people from top companies who are 2–3 years ahead of you.

Even if you don’t hire them, the conversation is a free audit.
You’ll learn exactly what your company is missing to play in the big league.

2. The “skeletons in the closet” test

Stop painting a rosy picture in interviews.
Show the candidate the biggest mess in your company and ask: “How would you fix this?”

• If it excites them - you’ve got a leader.

• If it scares them - you just saved both of you 6 months of frustration.

3. The 30-day rule: Wow! or Out!

If after a month you’re not thinking:
“How did we ever manage without this person?”
- you probably made the wrong call.

My hit rate for top hires? ~50%.
That’s the game.

4. “PE Yourself” – manage like an investor

Private Equity wins because they have no sentiment.

Look at your business as if you just bought it.
Would you keep “Seb” because he’s a nice guy - even though he doesn’t deliver?
Would you still drive that old leased van instead of investing in real market standards?

Be honest.

5. A self-driving company = your ego on a diet

A real leader is a system architect, not a firefighter.

If every decision has to go through you,
you didn’t build a company - you built a prison.

Push decisions to the people closest to the action.
They have full context. You only see fragments.

Money gives freedom only if you build a system that doesn’t need you.
Otherwise, you just bought yourself a very expensive, very stressful job.

Do you agree with “Hire for attitude, train for skill” - or in your industry are hard skills still sacred?