For people building AI agents: what’s the hardest part about turning an agent that actually works into revenue?
Assuming you’ve built an agent that can reliably complete a useful task, where does the real bottleneck start?
Is it:
- Finding customers / distribution
- Convincing people the agent is reliable
- Pricing the work
- Getting access to the tools/data it needs
- Handling payments and billing
- Getting repeat / recurring work
- Competing with general-purpose tools like ChatGPT
- Something else?
I’m especially curious whether people actually struggle with building capable agents, or whether the bigger problem becomes finding enough paid work for them once they exist.
Say a customer offers you $10,000 to migrate a service to a new framework. Several hundred files, incomplete tests, undocumented behavior, and integrations with other services. You get paid when it meets agreed acceptance criteria.
Would you let your agent do it? And what would you need to know before quoting a price?
I think this actually exposes a different problem from whether an agent can finish an individual project.
On a large project, someone has to discover the dependencies, break down the work, decide what can run in parallel and keep the pieces compatible. If an early architectural assumption is wrong, 10 agents can spend a day implementing it very efficiently.
Ive noticed separate works trees don't fix conflicting assumptions. 2 patches can pass their own tests and fail when combined and recovering from a bad decision made 6 hours earlier means figuring out which work depended on it and what needs to be redone.
So the question is, who handles that? Another agent who explicitly runs integration checks or an irl engineer?
This is where the business logic gets interesting. Anyone running agents as a service knows that your costs include exploration, execution, integration, verification and rework. Model usage is probably the easiest part to estimate. But have an irl engineer oversee is when things go wrong and become expensive.
Your customer also has to review the migration. Milestones can reduce the risk but someone still needs to define and verify them
I have my skin in the game, so im trying to understand what makes larger agent projects commercially viable. Finding a buyer only helps if someone can credibly commit to delivery. A context and super vertical agent, or you.
Any advice? If you've shipped a substantial project with agents; how much human time went into collaboration, integration and review? and knowing what you know now, would you again take that project at a fixed price?
Peter Thiel on the difference between American and German wealth:
“In the US, nine out of 12 of those people made their own money.”
“In Germany, all 20 of those 20 inherited it.”
“Not a single person did something new.”
inventory is full
I'm out of range
i can't carry any more
that spell isn't ready yet
i need to target something first
it's too far away
that ability is not ready yet