Note
October 6, 2026
2 min read
Teaching a Toucan to Fly, One Weekly Token Allowance at a Time
By Cristiano Pierry
An unfinished toucan animation became a practical reminder to set spending limits, time limits, and review checkpoints for AI agents.
Over the weekend, I asked my Codex AI agent, Bob, to animate a toucan for my personal website. I woke up Monday morning to find him more than twelve hours into the task, with my weekly allowance exhausted and the toucan still unfinished.
Bob is still working, and I can still give him directions. OpenAI allows an active turn to continue after the usage limit is reached, subject to fair-use limits. Thank you, OpenAI. I’m certainly getting my money’s worth.
My weekend project acquired a rather detailed bird agenda.
It’s also a useful reminder to put some checks in place when we assign work to agents. Exhausting my personal subscription allowance is manageable. An agent making metered API calls without spending controls could become expensive very quickly, especially if it keeps trying to fix the same problem.
I’d set spending and time limits upfront, with checkpoints to review what has actually improved and approval required before increasing the budget or expanding the scope. We need a way to pause, reconsider the approach and decide whether another round of work is worth it.
These weekend projects are paying off. They give me room to experiment with foundation models, refine how I use them, and revisit where they can help with product development. Each project gives me a clearer sense of what to delegate, what to check, and when to step in. Still waiting for my toucan, though. 😊
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.