The Undeclared Dependencies
A series on AI agents as a new kind of user: one that does not do what you expect, but what your system allows.
- Paths are infinite. Acceptable outcomes fit on one page ES
- The best diagnosis of your product lives on someone else's servers ES
- Your support tickets are down. That's the problem ES
- List the channels you don't control ES
- Your invisible attrition isn't random ES
- You don't need an AI team ES
- Perfect exit, zero voice
- The agent isn't hostile. It behaves like a fuzzer
- Agents barely buy anything. That's not the point
- What the system allows was never what the company wanted
- The agent doesn't absorb your org chart. It executes it
- Two questions about your own company you probably can't answer
- Your users were your consensus protocol
- Nobody tests what else is allowed
- The subsidy you never knew you were receiving ES
- What an agent sees of your company is your worst channel