When a company is actually ready for AI agents.

We get the same opening question, twice a week: "Are we ready for AI agents?" It's a stand-in for a different question — one that's harder to ask out loud. The real question is: do we have a system worth automating?

Agents amplify the system underneath them. If the system is broken, you get faster broken.

Three preconditions, and they're not technical

We've started screening engagements against three preconditions before we agree to design an agent. They are unglamorous and they sort the conversations quickly.

• A repeatable decision. Something a competent operator does the same way every time, with inputs that mostly live in your stack already.
• A measurable outcome. If you can't tell us what "good" looks like in a number — handle time, conversion, accuracy, cost — we can't write evals against it.
• An owner. A real person whose KPI moves when the agent works, and whose phone rings when it doesn't.

If you can produce all three, you're ready. If you can produce two, we can usually get to three in a week. If you can produce one, you don't have an agent problem — you have a clarity problem, and an agent will make it worse.

The wrong starting question

"Where can we use AI?" produces an inventory of cute use-cases that all die in pilot. The better starting question is operational: "Which of our most valuable workflows is the most fragile?" That's where agents create leverage — by absorbing the variance in something the business already depends on.

Signals you're not ready (yet)

• The data the agent would need lives in someone's head, or in a spreadsheet emailed weekly.
• Two senior people will give you contradictory definitions of the workflow.
• Nobody has written down what "wrong" looks like — only "right".
• The team that would own the agent doesn't yet exist.

None of these are blockers forever. They're a six-week problem, not a six-quarter one. We almost always start by fixing them.

What "ready" actually buys you

The teams that are ready don't ship faster because their models are better. They ship faster because the workflow is legible, the metric is agreed, and the owner is in the room. The agent is the last 20% of work, not the first.

If you're trying to figure out where you sit on this — that's exactly the kind of conversation a working session is for.