The agent is already inside your workflow
When a founder asks me "where can we add AI to our business?", I gently push back. It's the wrong question. It treats AI as a thing you bolt on — a feature, a chatbot, a line item. That framing almost always leads to a demo that impresses everyone in the room and then quietly dies, because it was never load-bearing.
The better question is the one I actually go in to answer: where is there already an agent inside your workflow — a piece of repetitive judgment a human is doing today that a machine could do just as well, or better, every single time?
Why "find" beats "add"
Every operation runs on a thousand small decisions. Is this claim a real denial or a contractual adjustment? Does this lead look like it'll convert, or is it noise? Has this document got everything it needs before it moves to the next desk? Each of those is a tiny model a person carries in their head and applies all day.
Those decisions are where the agents are. They're not glamorous. They're the unglamorous workhorse calls that get made hundreds of times a week, with rules that are mostly consistent and mostly knowable. That consistency is exactly what a machine is good at — and exactly what burns out a smart human who'd rather be doing the hard 10%.
How I look for them
I don't start with the technology. I start by watching the work. The pattern I follow is simple and I run it the same way every time:
- Audit. Shadow the team, read the tools, trace a unit of work from start to finish. Write down every place a person applies a repeatable rule. Rank them by how often they happen, how clear the rule is, and what it costs when it's slow.
- Build. Take the one or two with the best ratio of value to risk and ship them into production — in your stack, not a sandbox. An agent that lives in a slide deck has never saved anyone an hour.
- Hand off. Leave a runbook and monitoring so the agent is something the team owns, not something they depend on me to keep alive.
The agents that survive this are never the flashiest ones. They're the ones that quietly remove a recurring tax on someone's day — and keep removing it long after the excitement of the demo has worn off.
The test that matters
There's one question I hold every candidate agent against: does it earn its keep? Not "is it impressive," not "is it AI." Does it pay for itself in time, accuracy, or capacity — reliably, in production, this month? If the honest answer is no, I'll tell you that, and we won't build it.
That's the whole job, really. Find the agent that's already inside the workflow. Give it a good home. Make sure it earns its keep. Everything else is theatre.