You're one person. What happens if you get hit by a bus?
The fair version of this question is: what am I left holding if you disappear? The answer is everything, and that is deliberate.
What gets built runs on your hardware or your cloud account, under your credentials, in standard tools — Python, Postgres, Docker, ordinary web technology. No proprietary runtime of mine, no license that expires, nothing phoning home to me. Every project ships with written documentation of what it does and why each decision was made, and the code is yours the day it is delivered, not licensed to you.
The practical test is this: could a competent contractor you hire off the street pick it up? For everything I build, yes — and I will tell you before we start if any part of a project would fail that test.
Is this going to cost people their jobs?
No, and I would rather not take the work if that is the goal.
What gets removed is transcription — the hours somebody spends reading numbers off one screen and typing them into another. That is not a job, it is a symptom, and the person doing it is almost always capable of considerably more than that.
The return here is capacity, not headcount. The same team carries more work, which means you take orders you would have turned away and stop hiring to cover growth you could already have absorbed. If your plan is to cut staff, you do not need me — you need a spreadsheet and a difficult morning.
Everyone is selling AI right now. Why you?
Because almost none of them have had to make payroll inside the kind of company they are selling to.
The hard part of putting AI into a business is not the model. It is knowing which of the twenty things people do by hand actually matter, which exist for a real reason, and which have survived since 2014 because nobody asked. You learn that by running an operation, not by reading about one.
And there is a system in production you can ask about — one that replaced eight disconnected systems at a 455-person company, switched over only after it ran in parallel against the hand-calculated payroll until the numbers matched. Most firms pitching you have a demo.
We already tried AI and nothing came of it.
That is the normal outcome, and it usually is not because the tool was bad.
Most AI projects fail at the edges rather than in the middle — the model works fine, but nothing feeds it reliably, nothing does anything with the answer, and no one owns it after the pilot. The work is in the plumbing and the ownership, which is unglamorous and is most of what I actually do.
If you have a failed attempt sitting there, bring it to the first call. It is usually recoverable and it tells me more about your business in ten minutes than an hour of questions would.
How much of our time will this take?
Less than an internal project, considerably more than zero.
Discovery needs a few hours across two or three weeks — mostly me watching people do their work and asking why. During a build, expect an hour or two a week from your named decision-maker, plus short bursts from whoever knows a particular process.
The one non-negotiable is that named person. Not a committee. It is the single biggest predictor of whether a project lands, and I do not start without one.
What if it does not work?
Nothing goes live until it has been proven against the way you do it today.
At the manufacturer, payroll ran in parallel against the hand calculation for weeks and only switched over once every number matched. That pattern is standard here: the old way keeps running until the new way has demonstrably produced the same answer.
Fixed-price work is fixed price. If a build overruns because I misjudged it, that is mine, not yours.
Do we have to move to the cloud?
No, and for some of you the point is the opposite.
Where the work can live in the cloud and there is no reason it should not, it does — it is cheaper and simpler. Where the data cannot leave the building, whether for compliance, contract obligations or plain preference, it runs on hardware you own, with models whose weights sit on your own disk. Both are normal and the decision is made on your constraints, not my convenience.
We are not a manufacturer. Does this still apply?
Yes. The industry is not the thing these engagements have in common.
A dental group hands work from scheduling to insurance to billing. A distributor from order to pick to invoice. A law firm from intake to matter to trust accounting. Every one of those hand-offs is a person carrying information between systems, and it is the same problem wearing different vocabulary.
Manufacturing is where the deepest proof happens to be, not the limit of where this works.