What an AI agent actually is (and what is hype)
If it cannot fail observably, it is not an agent: it is a demo.
- Published
- 29 June 2026draft
- Category
- AI in production
- Reading
- 7 min
This piece is published as an outline: the statement and the structure are settled, the full text is coming.
The minimum definition
An agent decides which tool to call and with what arguments, and lives with the consequences. Everything else is a prompt with good PR.
The failure test
The question that separates product from demo: when it gets something wrong, do you find out? If the only symptom is an unhappy customer, there is no agent.
Typed tools, not instructions
Why a validated schema beats three paragraphs asking the model to behave.
Marked in red ink
The claims going around about agents, and what survives contact with a real system.
Write to me
If you are thinking about building something with AI and want to talk it through with someone who has put it into production, write to me. I answer myself.
hola@giron.ai