Post

Why a loop? An “agentic flow” can be trigger -> action ? I ask because I’ve seen previous descriptions using loop but I’m not sure why the idea of continuous execution makes it more “agentic”

I would only add that it can run as many tools or take as many turns as it needs. It “autonomously” decides when to stop.

That to me sounds like just bad computer science. In what other context would you say "this algorithm can take indefinite time and resources for a given task" and someone would be ok with it? Tasks have values, they need to take bounded time and cost.

So you don’t consider an LLM deciding when to stop on its own to be part of the definition of an agent? This is the current behavior of ADK (Google), Agents SDK (OpenAI), LangChain, and others.

Hmm. The industry’s been using “user agent” to describe software agents responsible for retrieving and facilitating user interaction with web content for a very long time. Web browsers, download managers, feed readers, email clients, etc.

Agreeing with the general premise of your post, I’d venture that “user agent”, “software program”, “software system”, or even “software agent” are more apt phrases than “LLM agent” in most (perhaps all?) scenarios.

Amusingly, Wikipedia only lists “Web browsing” under “Agentic AI” applications, other than linking to the MIT study spotlighting “95% of enterprise generative-AI pilots fail to deliver measurable P&L impact”. User agent and software agent wiki pages have myriad real world examples listed.

i hate the term agent because it’s paired with the term autonomous which is just so completely unreliable and discongruent with just about every experience in technology today. what we have now is a probabilistic rules engine at best.

Includes this note about why agents as human replacements is my least favorite definition - because unlike AI agents, humans have agency!

The worst offenders for constantly redefining agents with new, vague and inconsistent definitions are OpenAI themselves simonwillison.net/2025/Sep/18/...

Well, if you want vague and inconsistent, there is the glaring I in the AI bit. What is intelligence with respect to AI...? (rhetorical question). Tech loves to fuck with language.

Like your definition, however, be better not to call them agents at all. Another case of anthropomorphism with AI (implying human characteristics they don't have) If something is called an agent, it must, in some sense, have agency, which, as you say, they don't.

I tried the "don't call anything agents" thing for a couple of years, but it feels counter-productive at this point - plenty of people are "building agents" now where what they mean is "LLMs achieving goals by running tools in a loop" and I want to be able to have conversations with them about it

I also like being able to say "you do realize the one thing agents don't have is agency, right?" just to stir things up!

I think this is a standard weak/strong double use thing. the agency agents are supposed to have is in deciding what to do each loop call in service of the goal they're given. that's a very limited and derivative form of agency but it's too strong to say there's nothing resembling agency in the mix

I’ve always seen it as meaning “a thing I’ve delegated agency to”