# Why an AI Business Still Needs 100 People

The strongest AI businesses won’t win with one “super-agent,” but with factories of narrow agents, strict evals, orchestration, and people handling exceptions. Their real moat is accumulated domain skills, rules, error history, and execution speed.

Author: Maxim Tolmachev
Published: 2026-08-29
Language: en
Canonical: https://maximtolmachev.com/blog/why-an-ai-business-still-needs-100-people/
Original source: https://t.me/mtolm/101

“There won’t be a one-person company”: why an AI business with 300 agents employs 100 people

Today I listened to a talk by Vsevolod Ustinov, who is building Plurio.ai, an AI-Native company in User Acquisition that has raised $4.5M.

While the market is still debating whether “ChatGPT will replace people,” the team is building a real agent factory for performance marketing.

I picked out three major product shifts that are changing the rules of the game right now:

1️⃣ From selling tools to taking responsibility for results

An AI agent or chatbot quickly becomes a commodity. The wow effect of connecting Claude to a database wears off after a couple of days.

Clients don’t need yet another Copilot or dashboard where an employee has to sit, click buttons, monitor prompts, and take responsibility for the result.

The future belongs to the Outcome-as-a-Service model (AI Factory): a company sells neither software nor specialist hours, but a fully managed business function with financial responsibility for the outcome.

2️⃣ The death of the “digital CEO” myth: automation only works from the bottom up

Trying to build a single “super-agent” that sees the whole business from above, understands the context, and makes strategic decisions is still pure utopia.

A working architecture is built strictly from the bottom up: a complex process is broken down into dozens of narrow, repeatable operations (collect data ➔ find burned-out combinations ➔ turn off campaigns ➔ relaunch creative ➔ check the budget).

The winning setup is: lots of narrow agents + strict evals (tests) + an orchestrator + a human handling rare exceptions.

3️⃣ The team paradox: why an AI business employs 100 people

Right now, it’s fashionable to dream about a one-person company worth a billion dollars. But if you’re responsible for clients’ real multimillion-dollar budgets, you need a reliability infrastructure.

The 100 people in an AI Factory aren’t doing the work manually. They’re building the factory:
• developing domain skills;
• investigating model failures and edge cases;
• calibrating rules and building automated tests;
• maintaining SLAs for clients.

Here’s the leverage: one well-tuned system serves hundreds of clients. Large fixed investments go into creating the system, while the marginal cost of onboarding a new client approaches zero.

💡 The main takeaway

The main moat in AI isn’t the foundation model—in six months, competitors will have the same one or something stronger.

The real asset is the library of domain skills, digitized rules, history of analyzed errors, and speed of implementation.

Thanks to Vsevolod Ustinov for the honest talk, and to the organizers for putting together such a great conference about AI!
