TL;DR
- Enterprise AI today automates tasks inside departments: closing the books, managing denials, running ads. That gives you a faster horse, not autonomy.
- Autonomous operations need someone who sees how work moves across the whole organization from output to outcome, and product companies structurally can't see that.
- At Moative, we build agentic workflows for midmarket businesses in vertical industries and incubate AI-native companies in them. We keep seeing the same thing: it's faster to run an AI-native operation than to sell AI into one.
- Until megaplatforms arrive that can hold full context and collapse departments outright, the fastest path to autonomy is in-house engineers stitching discrete automations together, not vendor FDEs.
The holy grail of the enterprise is a company that runs itself.
You're shaking your head. Regulators won't allow it, boards won't allow it, the workforce won't allow it. Fine. But the most efficient enterprise is still the one that runs itself, and every dollar of AI spend is moving toward it.
To be precise: I mean autonomous operations, not autonomous strategy. Strategy stays human. Operations are within reach now, and that's where the prize is: a lower cost base, and customers who stay longer because things just work.
We automated thinking first, and it made things worse
Everyone uses Claude or ChatGPT to think. You can see it: documents are longer, words are fancier, meaning is murkier.
That's because thinking is the connective tissue between chunks of real work. Once the real work at the lower and middle levels gets automated reliably, most of that in-between thinking disappears, along with the decks.
We're in the faster-horse phase
Look at what's getting automated in every industry: book closure, denial management, ad operations. Discrete, adjacent steps, each automated well.
F.W. Taylor would not be impressed.
We're replacing workers with AI, but the AI is doing work whose boundaries were drawn by the limits of the department it sits in. A century of management science asked how to make the worker productive. Nobody has asked what makes the department redundant. Part of the reason is that every team has the authority to add and nobody has the authority to remove, which I set out in The Cost of Taking Things Out.
We're breeding faster horses. Soon we'll have to ask whether we're running a race that matters.
The FDE is the right idea in the wrong building
The forward deployed engineer movement does one job well. It covers for what the product can't yet do in the real world, learns how the work actually gets done, and feeds that knowledge back into the product.
But look at where the knowledge ends up: in the vendor's product. That is the design, not a side effect, and I set out what it costs the client in Don't shortsell your know-how to AI vendors.
Autonomous operations need the opposite. They need a full view of the organization: how one output turns into the next input, across every department, until an outcome is reached. A product vendor's FDE sees one slice of that. Its job is to push its product's bottleneck downstream, where it becomes someone else's problem.
Run it, don't sell into it
At Moative, we see this over and over. If the goal is an autonomously administered hospital, it is faster to run an AI-native hospital than to build AI for hospitals.
An AI product company automates whatever its context allows. An AI-native operator has no such limits. It brings in tools for the discrete jobs, builds the agents it knows it needs, and puts people only where AI can't yet be trusted. We see our job as that of someone who enables an organization to embrace AI across the board.
What comes next
A phase is coming where context gets captured reliably. Add long-running agents and cheaper compute, megaplatforms will emerge that do everything and force departments to collapse.
That moment isn't here yet.
Until then, the fastest path to autonomy is not buying more point solutions with more vendor FDEs. Vendor FDEs are like scaffolding. They will go when their part of the work is done. Why the scaffolding comes down as models get better is the argument in The Dawn of Enterprise Agency Engineering. What you need is your own FDE organization that sits in every department and between them. Your FDE team should treat products as scaffolds until they can be replaced with better ones.
Autonomy is being built the way Sagrada Família is: while it's in use, swapping one scaffold for the next.
So which inter-department ritual in your company would you collapse first, and what's actually stopping you?
Where Moative fits
Moative works as an AI operating partner with mid-market companies in vertical industries. We build agentic workflows inside them, and we incubate AI-native companies in the same industries, which is where the lesson in this essay comes from. On an engagement in healthcare revenue cycle management, we start with how work moves between departments rather than the task inside one, because the handoffs are where autonomy is won or lost. The engineers who stitch it together should end up being yours, sitting with the work, which is the case I made in Your IT Department should not be in the IT room anymore. If you are deciding which handoff to collapse first, we are glad to give a straight opinion.
Frequently asked questions
What does it mean for a company to run itself?
It means autonomous operations, not autonomous strategy. Strategy stays with people. The day-to-day work moves from step to step without people carrying it across. The payoff is a lower cost base and customers who stay longer because things just work.
Why isn't automating tasks inside departments enough?
Because the department drew the boundaries of the task. Automating book closure, denial management or ad operations makes each step faster, but the handoffs between departments stay where they were. You get a faster horse, not a company that runs itself.
What is wrong with vendor forward deployed engineers?
Nothing, for the job they do. They cover for what the product can't yet do, learn how the work gets done, and feed that back into the product. The problem is where that knowledge ends up, and that a vendor's FDE only ever sees its own product's slice of your organization.
Why is it faster to run an AI-native operation than to sell AI into one?
An AI product company can only automate what its context allows. An operator has no such limit. It brings in tools for the discrete jobs, builds the agents it needs, and puts people only where AI can't yet be trusted.
What will make departments collapse?
Platforms that capture context reliably, combined with long-running agents and cheaper compute. Once a system can do the work end to end, department boundaries stop making sense. That phase is coming, but it is not here yet.
What should a company do until then?
Build its own FDE team that sits in every department and between them. Treat vendor products as scaffolding, useful until something better replaces them. Then pick one handoff between departments to collapse first, and find out what is actually stopping you.