Software with AI sprinkled on top isn't agentic

There are three things people mean when they say "AI software," and the differences matter more than the label.

Chat on your data

The first is a chat window pointed at your warehouse. Ask a question, get an answer. Useful — genuinely — but it's a faster analyst, not a different kind of system. Someone still has to ask.

AI-written software

The second is using AI to build ordinary software quickly. Describe a dashboard, get a dashboard. Also useful. But what comes out is the same software you would have gotten from an engineering backlog, minus the wait. It sits there until someone opens it.

Agentic software

The third is different in kind, not speed: software that takes a standing instruction and keeps executing it in the background.

Say it once — "when a damage claim form comes in, check it against the policy and past claims, and draft the response" — and the work keeps happening. Not because someone asked that morning, but because a form arrived at 8:41 am and the instruction was standing.

The pattern shows up everywhere in retail operations:

Trigger	Standing instruction
A form comes in	Check policy, draft the response
The 5th of the month	Study aged stock, rank the moves
A what-if gets asked	Replay last quarter under the new plan
A store makes a request	Pull traffic and peer comps, write the memo

None of these need a prompt. They need context — the comp plan, the promo calendar, the store list, the exception history — and that context accumulates from every build that came before.

That's the real dividing line. Chat answers questions. AI-written software waits to be opened. Agentic software carries standing instructions on accumulated context, and every build makes the next one smarter.

How it works walks the build path end to end, and the background loops show five standing instructions running at one retailer.