How Much Should One Person Be Trusted to Direct?

In April, I wrote a short note imagining a familiar compensation statement: salary, medical, dental, vision, retirement benefits, paid time off. Then one additional line: a monthly AI token allowance.

There was a joke in there about whether unused tokens would roll over. I still expect someone in HR to have that conversation. The part I meant seriously was that access to capable AI tools could become a meaningful part of what a company provides its employees. A token allowance would give people more help to put to work on an idea. The Compensation Statement

A recent post by Mike Taylor took that thought considerably further. He proposed that an unusually capable engineer might manage a $50 million token budget and earn compensation resembling an investment manager's: a 2% management fee and a share of the returns. I like the ambition of giving a person that much room to create something valuable. It raises a question I think we will spend more time on: how much work should one person be trusted to direct?

I expect more people to direct what amounts to an army of AI agents, backed by a budget. One agent might investigate a problem while others build alternatives or test the work. The person directing them would decide which questions deserve the effort and how the results fit together. Their capacity would depend partly on how much work they could guide well beyond what they could perform themselves.

That makes agency and taste especially important. A person with high agency can recognize an opportunity and take responsibility for moving it forward, including when the path is unclear. Taste helps them decide whether the result deserves to reach a customer. They can see when something technically works but feels wrong, when a simpler approach would serve the user better, or when an attractive idea deserves no more time. Both qualities become consequential when a decision can set many agents to work.

This opens a path for people who want to keep building. Someone could take on substantially more responsibility while remaining close to the work, without acquiring a larger reporting organization. A specialist might develop a repeatable workflow, extend it to problems the team previously had to leave alone, and become responsible for improving it over time. I would like to see companies make room for that kind of growth.

The salary comparison becomes awkward at that point. A person's compensation tells us very little about the cost of the next worthwhile experiment they could run. A budget capped at some comfortable fraction of salary might be too small for a useful opportunity. We would need to examine the opportunity itself and what the person is prepared to do with it.

The authority attached to the budget would matter too. Someone needs enough room to choose an approach, discard it, and try again within agreed limits. Requiring approval for every agent task would leave much of that agency unused. The company would need to make the relevant data and tools available, and be clear about which actions require additional review. A useful budget comes with room to exercise judgment.

Consider a designer exploring an onboarding problem. With a modest AI budget, she might direct agents to build several working alternatives, test the account states, and prepare the flows for user research. Her taste would show in which alternatives she keeps and what she asks the team to learn from them. If the exploration exposes a bad assumption early, I would consider the budget well spent, even if none of the prototypes ships.

Suppose she then asks for a larger budget to test the approach across more complicated account types. I would want to see what the first round taught her and why the additional cases matter. She should be able to show where the current design breaks and explain how the next round would help the team make a decision. That gives the budget owner something concrete to assess.

There is also a point where producing more alternatives creates work for everyone else. Someone has to inspect the flows, check what the prototypes imply about the underlying system, and decide which direction deserves implementation. If those people are already at capacity, another batch may sit untouched. Review capacity belongs in the budget conversation. The useful scope of the designer's work depends partly on how much of it the team can absorb.

Even after accounting for the cost, the return may take time to become clear. The onboarding exploration could prevent a poor investment months before anyone can attach a credible financial value to it. The designer should be able to explain what she learned and which decision changed. Once the work reaches customers, the team can examine whether onboarding actually improves. I would be wary of inventing a revenue number to justify useful exploration, or continuing to call something exploration after it repeatedly fails to inform a decision.

For a released change, the evidence should become more concrete. Did more customers complete onboarding successfully? Did support requests fall? A controlled comparison, where practical, would help separate the effect of the change from other things happening in the product. The cost should include the agents and the people needed to review and maintain the result. Time saved becomes useful when the team can put it toward something else; it does not automatically become cash saved.

Employees should share in the value they help create. If someone consistently turns a budget into useful outcomes, that should matter in compensation and advancement. A bonus tied to a verified improvement could be one approach. Broader, sustained responsibility might justify higher compensation or equity. The reward should reflect the contribution over enough time to see whether the result holds up, including the cost of fixing what initially looked successful.

Paying someone a percentage of what they spend would create an awkward incentive. A person who achieves the same result with half the compute has made a valuable decision. Their reward should recognize that efficiency. Nor should the person directing the agents receive all the credit for work that depends on a platform another team maintains or an engineer who catches a mistake before release. Sharing the upside requires an honest account of who contributed.

There is a fairness problem in giving larger budgets only to people who have already demonstrated exceptional results. They needed access somewhere to develop that ability. If the initial opportunity goes only to the people a manager already trusts, the company may keep finding evidence of talent in the same small group.

A baseline allowance gives people room to learn. They should be able to try the tools on real work, make some unproductive attempts, and develop a sense of when the help is useful. Larger commitments can follow evidence. A new employee needs a reasonable way to produce that evidence before the company asks them to prove they deserve the tools.

The designer who makes onboarding meaningfully better should have a path to take on the next difficult problem, with a larger budget if the work warrants it. Her scope could grow through the agents she directs and the quality of the decisions she makes. Her compensation should grow with a sustained contribution. That would give the token allowance a much more consequential place in the compensation statement I was imagining.

This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.