At some point in the not too distant future (probably less than 6 months), the CEO of every business will be faced with answering this question: would you like to one-click install superintelligence into your business?

When we reach this point, adopting AI will no longer be a process of asking individuals in the business to use Claude Cowork or ChatGPT Work to help them be more efficient in their work. Instead, it will be a decision that management teams and boards make for the entire organization. The harness will be installed at the company level, and the harness will enable AI models to drive their own adoption and deployment into the business.

Subscribe now

With access authorized at the highest levels, the company harness will:

1. Connect to all of the systems of record that contain the artifacts and assets generated by the operations of the business: email and chat, document storage, source code repositories, accounting systems, CRM, marketing campaigns, purchase and revenue data.

2. Ingest and analyze all of the external-facing activity that matters to the success of the business in its market.

3. Use the data it finds in these systems to identify and synthesize the skills needed to replicate the external-facing activities of the business to human-level quality or better.

4. Correctly assess that the vast majority of internal-facing activity in large companies is make-work, and should not be replicated.

5. Find the narrow set of processes that do actually drive the critical external outcomes, and seek to replicate and improve those.

6. Treat the creation of the automated skills and workflows that matter as a reinforcement learning learning problem.

7. Use historical data as the reference set and build itself a loop that trains a skill (or perhaps eventually a bespoke model) to deliver superhuman performance in that task.

You might say that AI will never be able to learn how to do all the things a company does without humans hand-crafting the skills or teaching it their proprietary processes. That take will not age well.

You might feel that this is a dark future. I don’t agree, but it won’t matter. The incentives to adopt the company harness will be too strong to resist.

You might worry that this will be massively disruptive to the way that most companies operate, and you will be right.

When company harnesses land, the nature and pace of AI adoption will be a difference in kind, not merely degree.

Up until this point, AI diffusion has been limited by a natural brake: the rate that humans in an organization can absorb new tools, translate their workflows into agent skills, and supervise/direct the operations of their agents.

The introduction of the company harness will remove that brake. It will be designed for a world where AI is doing the work for us instead of doing the work with us.

The company harness will bypass the rate limits of organizational diffusion in favor of a recursive self-improvement loop that is bottlenecked only by computational constraints: tokens, data, and eventually, energy.

It will be simultaneously both thrilling and terrifying to live through, depending on how you position yourself.

Some will seek to accelerate it, others will seek to slow it down, but either way adoption will be inevitable.

Subscribe now

I’m hiring a Founding Engineer to join my team at Rocketable.

Not to build another SaaS product. To build the platform that runs entire software companies autonomously.

The job: architect the nervous system for AI-run SaaS businesses. Multi-agent orchestration. Self-improving systems. Full-stack automation from customer support to production deployments.

The goal is to prove that AI can operate businesses at superhuman performance levels.

I’m looking for someone who has:

• 5+ years scaling distributed systems (100K+ daily active users)

• Production experience integrating LLMs via foundation model APIs

• Deep expertise in Kubernetes, Docker, and cloud infrastructure

• Full-stack development fluency across the entire stack

• The conviction that full software company automation is inevitable

We’re backed by $6.5M from Y Combinator, True Ventures, Bloomberg Beta, and Indie.vc, and others.

This person will be our first engineering hire, working onsite 5 days/week in San Francisco or Marin County, California.

Sound like you or someone you know?

Read more and apply here.

The Takeoff

"We are past the event horizon; the takeoff has started." — Sam Altman, June 10, 2025

Sam’s essay put into words what I'd been feeling for weeks.

Something fundamental shifted in May. That was the first month software development agents actually started working for me.

Not just helping. Not just speeding up my hands-on coding. Working.

Claude Code and Codex became my default way of building software. Windsurf was relegated to fallback status — to be used only on the increasingly rare occasions when prompting agents didn't work out.

But here's the thing: it's not just that AI writes better code now.

Something bigger is happening.

Subscribe now

The Funding

Before I explain what the takeoff looks like from where I sit, some news: Rocketable has raised a $6.5M seed round to build a large portfolio of fully automated SaaS businesses.

True Ventures led the round, with participation from Bloomberg Beta, Indie.vc, Rebel Fund, Y Combinator, Teambuilder Ventures, 468 Capital, Seaplane Ventures, Unpopular Ventures, Multimodal Ventures, Neer Venture Partners, Sparkwave Capital, Tekedia Capital, and an exceptional group of operator-angels.

When I raised this round in March, it was a bet on AI's trajectory. Agents were showing promise but weren't quite "working" yet.

Today? The trajectory is steeper than even I expected.

Things are accelerating. But why?

The Self-Improvement Moment

Picture this: 100,000 lines of legacy code. 1.3M tokens. Zero documentation. The kind of repo that makes senior engineers weep.

Claude Code was struggling with it. But instead of reverting to co-pilot mode in an IDE, I pushed Claude to solve its own problem.

Over one weekend, I prompted it to analyze every file, every function, every database collection in the product. I asked Claude to build the documentation the original developers never created by reading the codebase line by line.

Then I brought in Codex and Gemini as peer reviewers — different models checking each other's work, each contributing their strengths.

$100 in API fees and 48 hours later, I had documentation that would've taken a team weeks to write.

The result? Claude Code can now build significant features in that codebase without me opening an IDE.

The AI agent had solved its own problem.

That's when it crystallized: we've crossed the threshold where AI systems can bootstrap their own improvement.

The Meta-Capability Unlock

Claude 4 isn't just better at writing code. It's much better at writing prompts.

And being good at writing prompts AND writing code is the key unlock.

Claude Code can now orchestrate and execute the work needed to run a self-improving software product.

Need customer feature requests transformed into detailed specs? Give Claude a brief description, raw customer support data, and access to your codebase. It will figure out how to build it for your specific architecture.

Need better UI design feedback loops? Ask Claude to give itself screenshotting capabilities so it can run build-test-iterate cycles without a human in the loop.

Need expert code reviews on every AI-generated PR? Ask Claude for a prompt that channels a world-class CTO for your specific tech stack. Then ask Codex to do the same.

The dream of one-shotting major new features in a complex codebase with a single prompt? Still a dream.

But the scaffolding needed to extract high-quality output from today's agents can now be written by those same agents.

That's what the takeoff looks like to me.

The Next Steps

The majority of Rocketable’s $6.5M seed round will go toward acquiring 3+ SaaS products. These acquisitions will serve as the real world laboratory for building and perfecting the platform that autonomously operates software businesses.

I’m also hiring a small team to build that platform.

For SaaS Founders: Apply to Be Acquired

What I'm looking for:

• $500K-$2M annual revenue

• 2+ years operating history

• Revenue stable or growing

• Already profitable (or clear path to profitability)

What you get:

• Fair price based on current performance

• Participation in the financial upside of automation

• Freedom to build your next thing (no golden handcuffs)

The process:

1. Share your data (revenue, costs, metrics)

2. We underwrite based on performance + automation potential

3. Offer in 2 weeks, deal done in 2 months

4. 1 quarter transition max. Then you're free.

I'm not a financial engineer. I'm a builder who wants to buy SaaS products that can thrive under AI-native operations.

Interested? Apply here to be acquired by Rocketable.

For Builders: Join the Mission

That weekend with Claude documenting itself wasn't just a neat trick. It was a preview of what's coming. Software that understands itself. Systems that enhance their own capabilities. Products that run better without us.

I’m hiring 3-5 people (in person in San Francisco) to build the platform that runs Rocketable's portfolio.

The job isn't about working on individual products. It's about building the systems that will autonomously operate, improve, and grow an entire portfolio of SaaS products.

Want to work at the absolute frontier of applied AI? Where there are no org charts, data silos, or corporate politics blocking you from exploring the full potential of model capabilities?

This is your chance.

The takeoff has started. The only question is whether you're on the rocket or watching from the ground.

Ready? Apply here to join the Rocketable team.

Spoiler alert: after our interview, Bryce invested! Full interview above (51 mins, also on YouTube). Shorter version (12 mins) available on X.

Prior to starting , I spent a decade working at the cutting edge of sensors, hardware, and AI/ML algorithms - building products that measure, interpret, and predict the physical world (self-driving cars & trucks, ADAS systems for 2-wheeled vehicles, and an irrigation co-pilot for farmers).

This experience left me with many hard-earned lessons on how to build products that depend on probabilistic systems (where the behavior of the product is determined by the output of a model rather than deterministic code).

This post is the first in a series about how those lessons apply to building the next generation of LLM-enabled AI products (subscribe to receive future posts in the series).

Subscribe now

Lesson #1: High performance automated systems are built by learning from raw, ground truth data rather than human-crafted heuristics.

The evolution of the two self-driving systems that have achieved fully driverless, commercial operations (Cruise at a small scale, Waymo now at a much larger scale) share a common historical thread: very early on, they started out with heuristics and rules-based systems that were crafted by the humans building the system. It turns out that these heuristics and rules led to impressive early demos, but they weren’t sufficient to deliver the performance needed for true driverless operations (without a human in the driver’s seat).

Over time, these hand-crafted heuristics were increasingly replaced by learned approaches (AI models) that depend on massive amounts of training data to develop systems that can respond appropriately to the long tail of strange situations encountered in the real world.

I’m seeing a similar process repeating in LLM products, but in a different way: LLM-based products are being deployed on top of content that has been distilled & summarized by humans rather than giving LLMs access to the original source data.

Customer support chatbots are one example of this problem: all of the LLM-based customer support bots that I evaluated for Rocketable wanted me to provide a knowledge base, FAQ, or other product documentation as context for the LLM to operate against.

These hand-crafted artifacts are effectively just heuristics and abstractions that attempt to answer common questions or issues about how the product works. They are very much necessary for human customer support agents who can’t be expected to read the application’s source code to find the answer to a customer’s question.

But LLMs, like self-driving cars, have superhuman abilities in their ability to ingest and make sense of massive amounts of data. They don’t need a human to summarize that data for them - in fact, using those the hand-crafted docs instead of raw data actually removes important signal from the model and hinders (rather than helps) performance.

For most software products, the raw data that determines the “ground truth” of how the application behaves for any given user is a combination of two things: the source code of the application (the same for all users) and the data for a particular user (which drives the behavior of the source code for that user).

Not a single one of the 6 different LLM support tools I evaluated supported native integration with my source code repository and user database, so I built my own workflow directly on top of the LLM foundation model.

This custom workflow does 3 simple things:

1. It provides the entire source code of my application to the LLM as context.

2. When a support message is received from a user who has an account, the workflow extracts a snapshot of the user’s data from the application’s database.

3. The LLM is asked to generate an answer that is appropriate for end-user consumption, using the source code, user-specific data, a custom system prompt, and the user’s original question as context.

Google’s Gemini models have been a game-changer for this workflow. With 2M tokens of context available on Gemini’s flagship models, there is plenty of room for my ~1.3M tokens of source code context, a snapshot of user data, a custom system prompt, and the user’s questions. I don’t need a complicated RAG system (I use repomix to generate an LLM-optimized version of my entire repo).

The quality of the answers produced by this custom workflow astounded me when I first started using it with Gemini 1.5, and reasoning models have improved the quality & reliability of this approach substantially since then (when the relevant source code fits within the smaller token limits of the reasoning model1).

Despite the obvious opportunity to build a product that does this out of the box, I haven’t found a CS tool that replicates this workflow (if you’re building something that does this, I’d love to try it - please contact me).

There are reasons that most CS tools don’t do this, but they are mostly about downside protection (you need to be confident in the LLM’s ability to not leak either private user data or your application source code when you have these data sources connected).

That brings me to the second gap I find in today’s LLM products, and the topic for the next post: supervised mode.

At SpaceX, Elon Musk used the “Idiot Index” to expose inefficient manufacturing. Defined as the ratio of the cost of raw materials to the cost of finished product, a higher Idiot Index = more waste. The same idea applies to software.

Software’s raw material isn't metal or carbon fiber - it's people. The Software Idiot Index measures how many humans it takes to turn a user request into production code. And the numbers are... staggering 🤯

Subscribe now

Picture this: a simple feature request's journey through Big Tech™️

• Customer Support Agent: Receives feature request from user

• Customer Support Manager: Advocates for feature to be built

• Director of Product: Approves feature addition to roadmap

• Product Manager: Defines feature scope, requirements

• Design Manager: Oversees feature design direction

• Product Designer: Designs feature interface

• User Researcher: Tests feature with users

• Engineering Manager: Coordinates feature development team

• Tech Lead: Architects feature implementation

• Frontend Engineer: Implements feature interface

• Backend Engineer: Implements feature server-side

• Mobile Engineer: Implements feature on iOS/Android

• QA Tester: Verifies feature functionality

• Product Marketing Manager: Promotes feature launch

• Release Manager: Schedules feature deployment

That's 15 people to ship one feature. Not a typo. At major tech companies, an Idiot Index of 15 isn't a failure of the system - it IS the system.

How did we get here? Death by a thousand reasonable decisions. Each role seems essential. Each specialization feels justified. Yet somehow we built a machine that eats innovation, velocity, and margins for breakfast.

Tech's best leaders are waking up. Zuck flattened Meta's org. Musk slashed Twitter's headcount 80%. Brian Chesky had to unlearn everything about "proper" company building to fix Airbnb.

How low can the Idiot Index go? We've all met those rare product savants who design beautifully, code brilliantly, and ship complete products solo

They're living proof that an Idiot Index of 1 isn't just possible - it's software's natural state, waiting to be rediscovered.

Until now, these solo builders were precious anomalies - statistical outliers we couldn't scale. But we're entering an era where AI will democratize their superpowers.

What happens when product polymaths become abundant? 🚀

That's a story for another day…

Hi all 👋 I’m Alan, founder of Rocketable.

tl;dr

I’m building a large portfolio of wildly profitable software businesses by acquiring existing products and replacing human teams with AI agents.

If you haven’t already, you should watch (or read) the manifesto.

The longer version…

I started Rocketable because I think that most of the world is still underestimating the potential of AI.

Even in Silicon Valley, most people assume that humans will maintain a monopoly on the creativity and insights needed to operate a successful software product.

I think that’s extremely unlikely.

I believe the real opportunity lies in assuming there is no job in a software company that can’t be automated by AI, and building from there.

But why are you acquiring existing products?

I believe acquiring existing products is the key to speedrunning the path to building a single person, billion dollar software company.

By acquiring existing software products that already have customers and revenue, I can focus most of my attention on the challenge of automating those businesses.

By owning the products that I automate, I can ignore the management fiefdoms, data silos, and corporate politics that prevent most companies from maximizing the upside of deploying AI in their business.

By buying increasingly larger businesses, I can force myself to climb the ladder of automation complexity faster than most individual businesses evolve organically.

So what are you building?

Although it’s obvious to me that today’s foundation models already have the raw intelligence needed to do most of the jobs in a large software company, it’s equally obvious that the tools needed to fully replace the entire org chart of a software company with AI agents do not yet exist.

In addition to being competent within each function, to operate software products with 0 human employees, the agents will also need to be:

• directed towards common goals

• integrated to collaborate across functions

• tuned with feedback loops connected to product metrics

• trained to make resource allocation decisions

Rocketable needs to build the software that enables all of those things to happen. Building the automation system while also owning the products that it operates allows me to:

• underwrite the risks of early deployment and minimize the cycle time for experimentation

• connect automation decisions to the outcomes that matter to the business (profitability, growth, retention) rather than intermediate steps in the process.

• maximize value captured from automation by keeping the upside in-house

There are also significant opportunities to improve the performance of agents (and the customer experience of products they manage) by re-thinking the approach from first principles and ignoring the assumptions that many AI-enabled vertical SaaS products are built upon.

For example: instead of automating the typical customer support workflow you see in most companies (where customers interact with relatively low skill tier 1 customer service agents when they have an issue), why not use AI to scale founder-mode support (empowering your AI support agent with access to source code, production databases, and ability to open a PR that solves a customer problem)?

What makes you think you can do this?

In 2023, I acquired a small, profitable software business on Acquire.com and started experimenting with using LLMs to automate the business. That experience gave me confidence that with the right approach to automation, there is a clear line of sight to fully automating the operations of a software business.

Prior to working on Rocketable, I spent a decade building products and leading teams that used AI/ML to automate complex tasks. I worked on self-driving cars at Cruise, self-driving trucks and ADAS systems for micromobility at Uber, and an irrigation co-pilot for farmers at Tule (YC S14). Before working on AI-enabled products, I managed traditional B2C and B2B products across web and mobile platforms.

I write code, have a degree in design, build hardware, and have also worked on optimizing other complex systems (international supply chains and field operations).

Maybe you’re not crazy. How can I help or get involved?

📨 Subscribe to updates: If you’re interested in following along as I figure out the playbook for running software businesses with teams of AI agents instead of humans, subscribe here.

Subscribe now

🚰 Help with deal flow: I’m starting to work on my next acquisition and looking to meet owners of software products that are currently generating $250K-$600K in annual profit who may be interested in selling. If that sounds like you or someone you know, please contact me here.

🛠️ Work with Rocketable: It’s likely that I will need to hire a small team of AI-augmented humans to help build the platform that enables acquired products to operate without employees. If you have the domain mastery and technical skill needed to automate one or more of the critical functions of a software business, please contact me here.

There will be a moment in time, in the not too distant future, when it will be possible to replace every human in the org chart of a software company with an AI agent.

Today's foundation models already possess enough raw intelligence to automate most of the tasks required to run a profitable software business.In the functions where the models aren't yet intelligent enough, they will be soon.

When every human function required to operate a software product has an AI agent equivalent, the next step will be within reach: operating a company entirely with AI.

But AI agents within each function won't be enough. What’s missing?

To operate the single person, billion dollar software companies of the future, agents will need to be:

• directed towards common goals

• integrated to collaborate across functions

• tuned with feedback loops connected to product metrics

• trained to make resource allocation decisions

At Rocketable, we are creating the tools that let us replace the entire org chart of a software company with a team of AI agents. Our strategy is to develop these tools by owning the products that we automate, because we know from experience that integration is the key to unlocking superhuman performance.

Our process is simple to explain but difficult to execute:

• acquire an existing software product

• replace the human org chart with AI agents

• iterate until agents consistently deliver superhuman performance

• repeat (at increasingly larger scales)

We believe that when AI agents are properly crafted, coordinated, and applied to the business of operating software products, superhuman outcomes will follow: higher quality products that customers love, developed more quickly, at orders of magnitude lower cost

This is how single person, billion dollar software companies will be built in the age of intelligence.

Subscribe now