A firm that sells judgement about technology should be careful about which technologies it leads with. The vocabulary of our industry has become untrustworthy. “AI automation” now means everything and nothing. The landing pages of our competitors contain rocket emoji and promises to save you twenty hours a week; the proposals beneath contain Zapier flows and a few API calls that could have been written by any competent developer five years ago. We find this embarrassing and are trying to do something different.
This is a short note on what we prefer to say instead, and why.
The vocabulary problem
In any market, when a category becomes a way to charge a premium, the words get stretched. “Organic” in the grocery aisle, “gourmet” in the ready-meal section, “artisan” on a loaf of bread baked in a factory. The word becomes a claim about positioning rather than a claim about the thing. Buyers who are paying attention stop trusting the word; sellers who want to keep charging the premium invent a new word; the cycle repeats.
“AI” is going through that cycle at unusual speed. Eighteen months ago it was a real technical distinction — you either were building on language models or you weren’t. Today it is a marketing category that covers everything from genuinely novel reasoning systems to a regex with a decent wrapper. A buyer who has been approached six times this quarter by firms “using AI to solve your operational problems” has no way, on the surface, to distinguish the sixth pitch from the third.
So we default to being quiet about it. We use AI where it is the right instrument for the job. We don’t use it where it isn’t. We try not to tell you which parts of our work involve it unless you specifically ask, for the same reason your builder doesn’t tell you which parts of the job required the yellow drill bit.
Your builder doesn’t tell you which parts of the job required the yellow drill bit.
What we’d rather talk about
The things that are actually worth knowing about an operations firm, in our view:
Whether the person you are about to hire has seen your problem before. Most businesses’ operational problems are not novel; they are common problems dressed in the specifics of that business. A firm that has seen your particular failure mode ten times already is worth a great deal more than a firm with all the right technology and no prior exposure.
Whether what they build will run after they step back. A great many engagements end with the firm holding the only working copy of the system, the only set of API keys, and the only knowledge of how the pieces fit together. This is a failure mode dressed as ongoing revenue. A firm you could fire tomorrow without the business breaking is worth more than a firm you can’t.
Whether they will tell you when the work isn’t worth doing. Most operations engagements we see could, with hindsight, have been replaced with a simpler system, a half-day conversation, or sometimes nothing. Firms that will recommend the cheap answer — the answer that doesn’t pay them — tend to do better work on the engagements they do take.
Whether they have opinions, and whether those opinions are testable. A firm that says “it depends” to every question is hiding. A firm that makes confident claims it cannot back up is dangerous. The useful thing is a firm with strong positions, clearly argued, that you can disagree with. If you can imagine them being wrong, and they can describe what evidence would change their mind, that’s a firm worth working with.
What we’re willing to say about the technology
Not nothing. We use language models on most engagements. We host some of them ourselves and call others as APIs. We build orchestration in a mix of n8n, TypeScript, and occasionally Python. We can describe any of it in detail when asked and regularly do so in Local versus cloud AI for UK SMBs and in conversation.
We try to talk about the technology in the same register as a competent tradesperson talks about their tools — matter-of-fact, secondary to the craft, occasionally opinionated about specific choices. A good joiner will tell you why they prefer one glue over another, but only if you ask. A bad one leads with the glue.
What we are asking of you
Read our work, read our writing, and decide whether the reasoning holds up. We would rather be chosen for what we think than for what we build on.
If you find that our service pages read like they have been written by a firm that has done this work before, and our writing makes specific claims that can be checked, then we are the kind of firm we are trying to be. If they don’t, tell us — we’d rather be told.
What we would ask you to be suspicious of, from us and from anyone else in this market:
- Claims of hours saved per week without a specific workflow attached.
- A pitch that leads with the technology rather than the problem.
- Demos that use canned examples you were not in the room to choose.
- Proposals with vague scope and precise fees.
- A firm that can’t name a situation in which they would decline the engagement.
None of these are absolute; there will be firms who violate one or two of them and do good work. But the combined signal is useful, and it points, usually, toward firms whose primary competence is selling.
The firm we’re trying not to be
We are not a reseller of a specific tool. We are not a training provider selling “AI transformation”. We are not selling you a future in which AI replaces twenty per cent of your headcount. We think most claims of that shape will look embarrassing in five years.
We are a small firm that takes operational problems apart, designs better systems for running them, and occasionally uses language models as part of those systems where the models are clearly the right instrument. The work is mostly ordinary. The discipline is in doing it well and not pretending it is something more magical than it is.
We would rather be a boring firm that is right than an exciting firm that is eventually not.