Article · · 4 min read
Most teams didn't choose their AI tools together. One arrived with a new hire, one came bundled with existing software, another just stuck. Nobody owns them as a set, and they don't always agree.
If two tools answer the same question differently, surely one is wrong? So the instinct is to standardise on one, or find something that makes them all speak with one voice. One brain for the business.
It's the wrong goal. What has to be the same isn't the tools' voice. It's the job you give them.
This is a hypothetical example, not a client story.
A small lettings office has a backlog of repair messages. Two people each ask a different tool to put them in priority order.
The lists don't match. One puts a leaking washing machine near the top. The other leads with a boiler that's been out for days in a flat with a new baby in it. The office manager has two ranked lists and no idea which to act on.
The easy diagnosis is that the tools are inconsistent. The real problem is upstream: nobody wrote down what the ranking was for, what it could draw on, or who signs it off. Two loose questions got two reasonable answers to slightly different questions.
A job is not a prompt and it is not a workflow. It's three things you can write on an envelope before anything runs.
Purpose. What decision or action the work has to enable, and for whom. The maintenance coordinator needs to decide which repairs get a contractor this week. That sentence alone closes most of the gap.
Evidence. What the work may draw on, what it must show, and what it must mark as missing. Use the tenancy records and the last inspection notes. For each repair, show the property, the fault and the date it was reported. Flag anything the records don't cover. Confident wording isn't evidence; the reviewer should be able to point from each claim to its source.
A human gate. One named person, one moment, one basis for saying yes. The coordinator checks the top of the list against the safety and vulnerability notes before any contractor is booked. Not a group chat where everyone comments and nobody decides.
Put the same three lines in front of every tool doing that work. That's the "one" in one clear job, not one tool, one login or one house style.
Set the job up properly and the lists still won't be identical. That's fine, as long as both serve the same purpose, show the same evidence and land on the same person's desk.
Where they still differ, read the disagreement. It usually means one of these:
Sometimes one answer is simply unsupported; prefer the one you can check. Either way, the difference told you where to look. Two tools that agree can share the same bad assumption.
The one-brain picture assumes a single box you type into. Much of what would save that lettings office time isn't conversational:
Each needs the same three lines.
You don't need a new layer over your tools to make them agree. What we've described is writing, mostly yours, because it depends on what the work is for and what you're prepared to be wrong about.
We won't quote you hours saved or an error rate; anyone who does is guessing about your team. What we can tell you is the shape of the work, and that it's smaller than it sounds.
Pick one job that already runs, costs real hours and draws complaints: the repair list, enquiries that reach the wrong person, renewals nobody chases.
Write three lines:
Give those same lines, unchanged, to the tools you already have, and compare what comes back.
If you have several tools and no clear job for any of them, The Catalyst is the conversation for it. One real problem, worked through with your team, and the keys handed to you at the end.
The goal was never for your tools to sound alike. It was for someone in your office to look at what came back and say: yes, I'll stand behind that.