I Founded a 20-Particular person AI Company With Big Competitors — Here is Methods to Out-Ship a Team 3 Times Your Size

You know the competitor with thrice your individuals and thrice your price range. Sooner or later, that hole makes you marvel, “Do I have to hire the way they hire to keep up?”

Most of the time, you don’t.

Y Combinator CEO Garry Tan advised CNBC that firms are reaching as a lot as $10 million in income with groups of fewer than ten individuals. “You don’t need a team of 50 or 100 engineers,” he mentioned. Base44, a bootstrapped app builder with eight staff, offered to Wix for $80 million six months after launch.

Small groups can win with an AI-first playbook

As the founding father of a 20-person AI firm with a lot greater opponents, I consider AI may also help small groups punch above their weight. But this doesn’t imply utilizing AI for its personal sake.

Instead, the most effective AI-native playbook entails validating AI efficiency knowledge, holding AI accountable and utilizing AI in ways in which don’t bury people in time-consuming new assessment duties.

The methods beneath require self-discipline reasonably than giant groups or massive budgets, and you can begin them right this moment.

Verify AI-reported efficiency knowledge

Every AI software offers you a dashboard of hours saved and output produced. I ignore these numbers as a result of they measure quantity. A rising quantity line can imply progress or a rising pile of busywork.

When the code analytics agency GitClear studied 211 million modified strains of code, it discovered copy-pasted code exceeding refactored code for the primary time in 2024 and tied the shift to AI coding assistants. More strains shipped, and extra of them had been duplicates somebody must keep.

Tracking how shortly you hit roadmap targets is a greater method to assess AI’s influence. A function slated for late within the third quarter that ships in mid-April has pulled you roughly 170 days ahead. Log deliberate versus precise dates to your subsequent three roadmap objects, and maintain ignoring the dashboard.

Build prototypes as an alternative of prolonged function paperwork

In my expertise, the quickest agent-era groups spend extra time deciding and fewer time documenting.

At my firm, we used AI to eradicate prolonged proposal paperwork. When somebody has a brand new concept, they use AI to prototype it quick. Then the staff performs with it and makes a go/no-go choice in an hour. You can do that, too.

Write a one-page contract earlier than an agent writes code

Once a function is accredited, we give the coding agent a lean, one-page doc naming what the function does, the information it touches, the interfaces it exposes and the methods it may fail.

A selected, well-written contract dramatically will increase code high quality and reduces your time-to-ship. A poorly written or lacking contract slows you down: Auditing and fixing dangerous code prices greater than the agent saved.

A 2025 Faros AI examine of 10,000-plus builders discovered groups with excessive AI adoption finishing 21% extra duties whereas code assessment time rose 91% and bug charges rose 9%.

Require the one-page contract this week for any agent-generated work.

Let automated gates and brokers do preliminary critiques

The high frustration with AI instruments within the 2025 Stack Overflow Developer Survey, named by 66 p.c of builders, was options which might be “almost right, but not quite.”

Most groups counter these buggy options by including extra human assessment at each stage. I believe that intuition will be counterproductive.

Consistent automated testing, constructed as a sequence of deterministic gates, frees people from the primary line of assessment. That can imply requiring a compiler, kind checker, linter and a full check suite in entrance of each merge, then letting brokers assault every change like an adversary.

We do that at my firm, and the result’s cleaner code. My growth staff spends extra time architecting checks and defining outcomes than combing by means of code. Make one gate non-negotiable, beginning along with your subsequent merge.

Rank work by what it modified for customers

AI made output low cost, and low cost output breaks scoreboards. Commits, tickets, drafts and decks inflate in a single day whereas nothing modifications to your prospects.

I name this consolation work. It seems to be and feels productive however doesn’t advance any objective, and AI multiplies it as a result of iterations are almost free.

Attach a verifiable check to each objective {that a} stranger may examine. “Checkout conversion crossed 4% and held for 30 days” is verifiable. Anyone can look it up. “Improved the funnel” is imprecise and may’t be validated.

Rank every undertaking towards these checks as an alternative of by variety of artifacts, and your staff learns quick that transport solely counts when it impacts a objective.

Use AI to your hardest choices

The widest hole between a small staff and an enormous one is how shortly they will make good choices. Big groups have extra knowledge and experience, so that they extra usually select the suitable path. Small groups can shut the hole through the use of AI for choice help.

To get an unbiased reply and catch AI fabrications, ask a number of fashions the identical query. Each has its personal map of the world, so the locations the place they agree counsel a consensus.

Then have one mannequin, or a skeptical teammate, attempt to discredit that consensus. Consensus tells you what’s widespread. Verification tells you what’s true.

At my firm, questions that when took a senior researcher per week now come again that morning. Run your subsequent class wager or pricing name by means of this earlier than the assembly that decides it. It is the throughline of what we’re constructing: as extra patrons and their AI assistants ask whether or not they can belief a advice, the benefit goes to whoever can confirm what the mannequin solely guesses at.

Discipline is the moat

A competitor with thrice your individuals carries roughly 9 occasions the coordination price. Fred Brooks named the legislation a half-century in the past in The Mythical Man-Month. “Adding manpower to a late software project makes it later.”

Every new rent coordinates with everybody already there, so working relationships multiply quicker than headcount. Five individuals have 10 of them. 100 have almost 5,000.

Shopify runs hiring on this math. CEO Tobi Lütke advised his groups to reveal why AI can not do a job earlier than asking for extra headcount.

The groups that rebuild round these six shifts win the last decade. The ones that wait will spend it catching as much as groups a 3rd their measurement.

You know the competitor with thrice your individuals and thrice your price range. Sooner or later, that hole makes you marvel, “Do I have to hire the way they hire to keep up?”

Most of the time, you don’t.

Y Combinator CEO Garry Tan advised CNBC that firms are reaching as a lot as $10 million in income with groups of fewer than ten individuals. “You don’t need a team of 50 or 100 engineers,” he mentioned. Base44, a bootstrapped app builder with eight staff, offered to Wix for $80 million six months after launch.