More business school insights for your agent orchestrator
On the first post of what is quickly becoming a series, I have pointed out the similarities between orchestrating AI coding agents and managing a company, today we continue taking inspiration from another trending post How I estimate work as a staff software engineer to point out more similarities.
This time we do the reverse though: take business insights and make them specific to orchestration. Still with me? Good.
- It is not possible to accurately estimate software projects
You can't estimate how long your orchestrator will run, or how many tokens () it might need, and neither can the orchestrator. - Map everything out in advance ... doesn’t work: programmers must be empowered to make architectural decisions, because they’re the ones who ... code
You might think being super-descriptive on how a task should be completed is the best aspect, it isn't. Focus on the end state, let the agents find the way to it. - Start with the estimate, and then go and figure out what kind of software
Set your time (tokens) budget upfront to manage the quality of your software, not the other way around.
The whole How I estimate work section of the post is ready to copy and paste for your planner agent instruction set, go and take it if you need it.