Do not chase “agents”: for now it is mostly hype, and companies lose money on them
26 episodes in 12 months · 41 total · first in July 2024
There is huge hype around agents, but few real working use cases: agents glitch and do the wrong thing, and the whole idea of building your own agents is, in essence, utopian. The bulk of companies will simply lose money on them.
Most of my tasks have long been solved right in the chat, in Codex or Claude Code — without a separate “agent”. Before you invest, ask what measurable result it will give.
Happening now. At the end of 2025 I stopped using ChatGPT agents and reminders: they glitched all the time (extra episode, December 2025). And in 2026 more and more of my tasks moved into the ordinary chat (No. 098).
Where to start
Do not waste time on “personal agents”: most tasks are solved right in the chat or in Codex.
Before investing in agents, ask what measurable result they will give.
Not trying at all is not an option either: it is expensive and unlikely to pay off quickly, but it lets you keep pace with the times.
“Again, this thing, which costs a lot of money, which will hardly bring much profit now, but which will let you gradually, well, live with the present, as I always, always call it, yes.”
When I said it
Quotes
“Or when you, in your company, put huge money into building these very agents that a huge number of people are making now, which are no use to anyone whatsoever.”
“I think that in general this whole business of building your own agents for yourself, it's, essentially, utopian. I mean, it won't lead anywhere. Right now it lets a large number of people make money.”
“Don't chase the notion of an agent at all. And I've said many times that I don't like this notion. I think it will go away.”
“Claims even from programmers at serious companies like OpenAI and Anthropic that they are using them at the same time, that they have a thousand or two thousand agents working for them, are absurd in themselves.”