How large a project can one person take on with AI? I think the answer depends on what that person is selling: compute or a delivered result.
I'm Young, an independent worker with 20 years across healthcare, education, and long-term care. Over the past two years, I've used AI to run more than 15 projects at once. I've written about managing those projects and the system behind them.
An AI evaluation platform I was discussing gave me a concrete way to test this question.
Why "don't charge by the token" stuck with me
In August 2026, a16z published an article titled "You're Not a Model, Don't Price Like One" (source: a16z, 2026-08). Its point stayed with me: if all I sell is model usage with a markup, the value of my work falls with the price of that usage. I need to be accountable for the problem solved.
At the time, I was in talks about an AI evaluation platform. It was a direct test of that idea.
The project on the table
I was discussing the project through an AI platform partner and would serve as its external PM.
The other side provided high-end compute and internal models, so there was no extra compute cost for me to cover. The main deliverable was code, with a clear handoff contact.
That left one central question: could I build the platform, get it right, and deliver it as one person?
How I divide the work
A project like this traditionally points toward hiring engineers and forming a team. My working structure is one person plus AI:
- I make the decisions, plan the work, and answer to the client. Those responsibilities stay with me.
- AI does much of the implementation, batch work, and repeated engineering. That is where it saves time.
- I verify the output, then check the checks. A green test isn't enough if the test cannot catch the failure it claims to cover. I wrote about how I check whether my AI follows its rules.
This structure keeps the person making the decisions close to the work. It also avoids the fixed cost of maintaining a team for a clearly scoped project.
That is what "don't charge by the token" means in my own work. The client is paying for a platform delivered, not for a count of AI calls.
Where the model fits
| A traditional team | One person plus AI |
|---|---|
| Hiring, management, and fixed staff costs | Work starts when the project starts |
| Decisions pass through several people | The decision-maker does the delivery work |
| Price by people or hours | Price by the result delivered |
| Time needed for a team to settle in | An AI delivery process I've used for two years |
Teams still matter. This approach fits projects with clear scope, a need to move quickly, and one person who can own the result from start to finish. It doesn't fit work that needs many people on site or several specialists present at once.
Questions I get
Can one person keep quality up on a project like this?
Only with a clear split of responsibilities. AI handles volume. I handle decisions and verification.
What work suits this approach?
Clearly scoped work that benefits from a single accountable contact. Work requiring substantial on-site staffing is different.
How do you show that capacity?
Through the work delivered. In the same month, I was also handling security delivery for a medical system and an eight-chapter course production line.
I'm Young, an independent worker with 20 years across healthcare, education, and long-term care. The evaluation project and partner remain anonymous. If you have a project you assumed needed a team, book a free conversation with me and we can look at its scope together.