Michael Higgins
All writing

Down the stack: why I took a solution architect role

The scarcest profile in AI is not the leader who can talk about it. It is the person who has built it inside a real enterprise. So I went to build.

Working draft, content pending

I have taken a new role. I am leading the AI transformation for Corporate Development at UnitedHealth Group, as lead solution architect and forward deployed engineer.

For the people who know my career, that second title is the surprising part. I have spent the last several years as a product leader. Head of Product. The person who owns the roadmap, runs the reviews, and translates between users and executives. Solution architect and forward deployed engineer is a different job. It means I am the one designing the architecture, standing up the platform, and shipping the agentic systems myself.

This is a deliberate move, and I want to explain the reasoning, partly because writing it down forces honesty, and partly because I think the logic applies to more people than me.

The market told me something

Watch what the frontier AI labs are hiring for. Alongside researchers, the role they cannot fill fast enough is the forward deployed engineer: the person who goes inside a customer's organization and turns a frontier model into a working system, with all the security review, data governance, workflow redesign, and organizational friction that entails.

They invented that role because they learned something the rest of the industry is still catching up to: getting AI to work inside a real enterprise is the hard part. The models are astonishing and getting better on their own. The gap between "the model can do this" and "400,000 people's work actually changed" does not close by itself. Someone has to go close it, by hand, from the inside.

Now look at the other side of the market. There are thousands of product leaders who can talk fluently about AI. Strategy decks, vision memos, roadmaps with "agentic" on them. That profile is becoming commodity. The scarce profile, the one I could not stop noticing, is the person who has personally architected, deployed, and operated AI systems inside a large regulated enterprise and lived with the consequences.

I decided I would rather be the scarce thing than the fluent thing.

What the job actually is

"Forward deployed engineer" still gets blank looks outside AI circles, so here is the plain version of what I will be doing.

I sit inside the business function, not adjacent to it. I map how the work actually happens, decision by decision and document by document, the same way I used to shadow users as a researcher. Then I design and build the systems: the agent architectures, the data boundaries, the evaluation harnesses that tell us whether an AI's judgment can be trusted, the human-in-the-loop checkpoints where it cannot. I sit in the security and governance reviews and answer for the design. And I teach, because a transformation that only the architect understands is not a transformation.

Half of that list is engineering. The other half is design research and change management wearing an engineering badge. Which is exactly why I think people with my background belong in this role, and why I wanted it.

Going down the stack on purpose

In the last essay I described my career as a capability stack: visual language at the base, then user research, systems thinking, product strategy, and AI product leadership at the top. Twenty years, bottom to top. Each layer earned before the next.

This move adds a layer I could not have built from the product chair: the technical layer. Architecture. Agent design. Evaluation. The unglamorous plumbing of enterprise AI, where identity, data boundaries, and audit trails live, and where most AI projects actually die.

Here is the thing about stacks, in companies and in careers: the deepest layers are where the leverage lives. Applications come and go at the top. Infrastructure decides what is possible.

I have spent two decades operating above the technical layer, working alongside it, respecting it, but not owning it. Every year that passes, that gap costs more. AI product decisions are becoming architecture decisions. What the system is allowed to do, where the human sits in the loop, how you evaluate an agent you cannot fully predict: these used to be implementation details a PM could delegate. They are now the product.

I do not want to supervise that layer secondhand anymore. I want my hands on it.

Leading is not the thing I gave up

I should be precise about what this role is, because "stepped out of product leadership" can read as "stepped down."

I am leading a transformation program: multiple workstreams, a platform roadmap, governance, and the org change that comes with putting agentic AI inside a corporate function. That is leadership work by any definition. The difference is that I am also the one in the architecture, building. The breadth of a program lead with the hands of an engineer. That combination is the whole point. It is also, not coincidentally, exactly what the forward deployed role was invented to be.

And corporate development is a nearly perfect place to do this. The work is dense with judgment, document-heavy, confidential, and high-stakes, which makes it a proving ground for the hardest questions in enterprise AI: what you automate, what you augment, where the human must stay, and how you prove the system is trustworthy enough for decisions that matter.

The honest part

There is real risk here, and I would rather name it than have you wonder whether I have thought about it.

Careers have ladders, and I am stepping off a well-lit one. Senior product leadership rewards continuity: bigger scope, bigger org, bigger P&L. A detour into hands-on technical work can read, to a certain kind of resume screener, as a step sideways or worse. "Hiatus" is a word that has a way of becoming permanent if you are not paying attention.

I am paying attention. This is an immersion with a horizon, not an exit. I am also doing the things that keep the throughline visible: writing, certification work in AI architecture, and building in public where I can.

But the deeper answer to the risk is that I think the ladder itself is moving. The AI product leaders who matter in five years will be the ones who built during these years. Not exclusively, but disproportionately. When the tools change this much, the people who go down the stack and touch them directly come back up with judgment that cannot be faked or delegated. If I return to a product seat, and I expect I will, I will be a categorically stronger product leader for having built the thing I used to specify.

Twenty years ago I learned to see systems by making things: layouts, identities, interfaces, by hand, badly at first. Everything I know about design came from that contact with the material. AI is the new material.

I am going to go get my hands on it.

Contact

Let's build AI people actually use.

I'm interested in AI deployment, architecture, and product leadership roles where design judgment, enterprise scale, and human outcomes meet.