TL;DR
- A boiler room has an operator you can walk up to. Organisation-wide intelligence that decides and executes has no equivalent, and it should.
- The typical mid-market enterprise is embarrassingly unprepared: healthcare practices still debating what a data lake is, manufacturing CIOs still only provisioning cloud budgets.
- Forward Deployed Engineering is an admission that your data and know-how matter. If they matter that much to the vendor, they matter more to you.
- IT departments should be replaced wholesale by AI departments, and those people belong on the shop floor rather than in the IT room.
On a balmy Monday afternoon, imagine walking into the plant manager’s cabin and saying that they don’t need a boiler operator anymore because AI has become good at doing that job.
If I were the plant manager, I’d ask, “So, if the damned thing blows up, should I call you or chat with your bot?”
I am an insider in AI. I am not saying AI should not go to some sacred avenues inside a plant. It will. But the control of the control systems should rest with the plant manager. The intelligence should not be packed away and stashed into interfaces whose keys you will not possess.
A boiler room needs an operator. What about organization-wide intelligence that does, ratify, decide, and execute everything? Who is that operator? Where do they sit and whose payroll funds them? Are they a human whose throat is accessible or are they an email and a chat box away when you really want them three feet away so that you can imagine choking their visible neck?
Healthcare practices still debate what a data lake is; CIOs in manufacturing firms still only provision cloud budgets. They don’t know an MCP from an API. The typical enterprise is embarrassingly inadequate to handle AI.
I am not talking about change management. That is being dealt with by the brute-force called “Forward Deployed Engineering.” The self-serving framing accelerates deployment, picks the parts that are valuable, and leaves. (Side note: We like diffusion better than deployment and we keep the data and IP with our clients so that they get the upside of AI valuation too.). What happens after deployment? Next plant outage? New practice location with a new payer? Are we waiting for the glorious return of the AI maintenance team, like a lovelorn wife waiting for the returning veteran, so the fights can now be face to face?
The world is making a case for open-weights & local models that you can deploy within your premises. I will make a case for a team that goes with it. IT departments have to be replaced wholesale, by AI departments. They are too comfortable provisioning, when they have to be managing intelligence and maintaining it. And they have to be sitting in the shopfloor and not in the IT rooms.
The FDE model is the failure of the IT department. Who knows better than a company’s workflow that the one who does it and the colleague who enables it? And yet, the constraints of the talent market have prevented the industrial campus in Cincinnati from having its own data scientists. The role that is genuinely missing alongside the engineer is the industry insider, which I set out in FDX: The executive that FDEs need to meet.
The market has changed. The software engineer is not in as much demand as the forward deployed engineer is. See LinkedIn. You’d find this mentioned everywhere, next only to pronouns.
Forward Deployed Engineering is tech conceding that your data, and knowhow matters a lot. If it matters a lot for them, it matters more for you.
Its time not just to bring back manufacturing onshore. Bring AI on-premises. Bring the team that brings AI on-premises, on-premises. Without it, you will be paying in tokens for running the factory and yet not have the keys to the machine that was built by the tokens you paid for.
Where Moative fits
Moative works as an AI operating partner with mid-market companies, and the distinction in this essay is structural to how we engage. We prefer diffusion to deployment, which in practice means the configuration, the rules and the IP stay with the client rather than with us, so the capability compounds inside your company instead of walking out with a contractor. That is also why we instrument a process before automating it, as set out in Instrumenting Work to Avoid Expensive AI Rollouts. If you are deciding whether to build an internal AI team or keep renting one, we are glad to give a straight opinion.
Frequently asked questions
Why should AI control systems stay with the plant rather than the vendor?
Because accountability has to sit with someone who is physically present and answerable when something fails. If the intelligence lives behind an interface whose keys you do not hold, you have outsourced control of your own operation without outsourcing the consequences.
What is wrong with the Forward Deployed Engineering model?
It accelerates deployment, extracts the valuable parts and then leaves. It answers the question of how to get AI running and not the question of who maintains it through the next outage, the next site or the next payer.
Should IT departments become AI departments?
Yes, and the change is more than a rename. Provisioning infrastructure is a different job from managing and maintaining intelligence. The skills, the reporting line and the physical location all have to change.
Why should the AI team sit on the shop floor?
Because nobody understands a workflow better than the person doing it and the colleague enabling it. Distance from the work is what forces companies to import outsiders to explain their own operation back to them.
What does running AI on-premises actually require?
More than open-weight models. It requires people on site who can deploy, monitor and maintain them, and the data and IP staying with the company rather than with a vendor.
What is the risk of not building this capability?
You end up paying in tokens to run your own factory while holding none of the keys to the system those tokens built.