Where AI adoption actually starts

Ask a founder where AI should land first and almost all of them say engineering. It is the instinct: put the new technology next to the technical people and let them figure it out.

Then six months pass and nothing has shipped.

The reason is not that the technical team is bad at this. It is that the work worth automating is mostly somewhere else, and nobody is pointing at it.

Engineering is the hardest room, not the easiest

Developers already have opinions. They have been using these tools since before anyone called it adoption, they have a workflow they like, and they are right that they could probably build something themselves.

So a technical team tends to hold the work. Not out of obstruction, out of competence, and out of a reasonable belief that this is their territory.

Meanwhile the executive team is reading about what everyone else is doing and getting impatient. The pressure builds from both sides and lands on the same small group of people, who are busy.

The result is a lot of trying and not much shipping.

The test. Ask your technical lead what the AI rollout put into production last quarter. If you get things with names attached, engineering has this handled and you should leave them alone. If you get a list of experiments, the work is stuck, and more pressure will not unstick it.

The departments with no developers get nothing

Compliance. Legal. Finance. Sales. Customer support. Nobody is going to second an engineer into those teams to build them an AI employee, because that engineer is needed elsewhere and the request never climbs high enough on the priority list.

There is also nobody selling into them. Every vendor in this market pitches the CTO, because that is where the budget sits and where a demo lands. The AI strategy deck that recommends starting with engineering was written by someone who has never watched a compliance analyst work.

So those teams get the consumer chat window and nothing else. They paste in, they copy back, and everyone calls it AI adoption.

This is where the work actually is. Not because those people are less capable, but because their work is more legible. A compliance analyst can tell you the exact shape of the document they produce, which fields come from where, and which rules never bend. That is a specification. Hand it to someone who can build and it becomes a real system in days.

An engineer's work is much harder to specify, which is exactly why they are better off doing it themselves.

The trade that makes it work

I have been inside this for a while at Scytale, a compliance and security business in Tel Aviv. Their VP of R&D, Eyal Cafri, describes where they started:

"Our AI use was scattered. Omri built the system Claude now runs on, in every department, and we ship faster at higher quality."

The important word there is every. Not engineering. Every.

The pattern has held across departments and it is always the same trade. They bring the vertical expertise. I bring the technology. Neither half works alone.

A compliance consultant knows what a defensible answer looks like and spots instantly when an output is subtly wrong. I know how to make the thing produce that output reliably, how to give it the memory to know the business, and where to put the stops. Put those two in a room for a few hours and you get something neither would have produced apart.

The stops matter more than anything else here. These departments touch regulated work, client commitments, money and filings, and every one of those is a place where a wrong answer is expensive and hard to walk back. So the build starts at the human review step and the rest is arranged around it. The AI employee does the drafting and the assembly. A qualified person still reads it and still signs it, and skipping that is designed to be difficult rather than easy.

The test. Point at any step where output leaves the building: an email to a client, a filing, a number that goes into a report. Ask who reads it before it goes. If the answer is a person's name, you have a system. If the answer is that the AI handles that one, you have a liability with good formatting.

Where to start instead

If your technical team has the AI rollout and it is moving, leave it alone and go help someone else.

If it is not moving, more attention on engineering is not the fix. Walk into the least technical department you have, sit with one person for an hour, and ask them to show you what they did yesterday. You will find the same thing every time: a document assembled by hand from four places, a rule that lives in one person's head, an hour a day going into something that has a shape.

That is the best work in the business, and it has been sitting there the whole time because nobody technical was ever pointed at it.