What the AI Investors Get Right About Agents, and What They Miss About the Power Industry

Earlier this year, Union Square Ventures published a thesis called "The Rebel Alliance," and it is a very good articulation of where enterprise AI is heading. Their argument, in brief: the AI era will not be owned end to end by a few giant model companies. Like every computing wave before it, it will unbundle into an ecosystem of specialized layers, and the force driving that unbundling is AI agents, software that doesn't just answer questions but goes and does things across many companies' systems. Agent traffic on the web has already passed human traffic. Credit where it's due…the essay is worth reading in full.

But read it from inside the power industry and one assumption jumps off the page. Everything in that thesis assumes the connections between companies already exist. Agents "hop" between enterprises. Information "flows" between systems.

In our industry, there is nothing to hop across.

An AI agent cannot parse a phone call. It cannot chase a thread with the subject line "RE: RE: FW: updated delivery??" It cannot act on a lead time that lives in one salesperson's head. The most capable agent ever built, pointed at our supply chain as it exists today, would do what every new procurement hire does in their first month: sit there, blocked, waiting for someone to reply.

The gap we've all learned to live with

We all know the sequence. A utility needs a transformer. Someone emails a distributor. The distributor emails multiple reps. The distributor calls the manufacturer. The reps call their manufacturers. A week later a lead time comes back, already stale, marked up along the way, and trapped in an inbox where no schedule, no system, and no other stakeholder can see it.

For most of our careers, that gap was survivable. Load was flat. Lead times were weeks. Pricing was steady and labor was abundant. None of that is true anymore. Load growth is the steepest it has been in generations. Critical equipment lead times are measured in years. Every week of coordination delay now lands somewhere that matters: a data center that can't energize, a substation that slips a season, a crew standing down because material didn't show.

And here is the irony worth sitting with. Within each of the great interconnections that power this continent, every synchronous machine is locked to a shared frequency. Sixty times a second, thousands of generators owned by hundreds of companies hold together as one system. Our industry solved physical coordination 80 years ago (literally), at a scale nothing else on earth matches. Commercially, we never built the equivalent.

So when the industry asks which AI tools to buy, we believe it's the wrong first question. The better question is whether, when capable agents arrive, they will have anything here to work with.

Infrastructure before intelligence

Our industry has seen this pattern before. Electric motors existed for decades before factories electrified, because the value was never in the motor. It was in rewiring the factory around it.

Agents are the motors. The rewiring is a shared, structured, permissioned layer where our commercial information, the capacity, lead times, pricing, specs, and schedules, actually lives and moves between organizations in real time. That layer does not exist in electric power. So we are building it.

And we made a sequencing decision worth being direct about, because it is the heart of the strategy: Utilen is built for humans first.

Not as a stepping stone. Not as a compromise. The coordination problem is already costing us real money today. Intermediary layers alone can add up to 30% on top of what buyers actually pay for equipment, before counting the schedule slippage that comes from decisions waiting in inboxes. A platform where a utility sees real capacity, live lead times, and current pricing in under a minute, and where a manufacturer reaches real demand directly, pays for itself on human workflows alone. No AI required. If the agent economy took a decade longer than anyone expects, the case would still stand.

But every quote, order, and schedule that flows through the Coordination Engine™ has a second property that email will never have: it is structured, permissioned, and machine readable. The same infrastructure that makes a person faster today is exactly what an agent needs to act tomorrow. We don't have to choose between solving today's problem and preparing for the next decade. Done right, they are the same build.

The choice in front of us

There is a reasonable instinct going around: the AI wave is noisy, the vendors are everywhere, maybe the wise move is to wait for the dust to settle.

Waiting is a fine strategy for picking tools. Tools are swappable; buy them late and little is lost. Coordination infrastructure works differently, for a reason this industry understands in its bones: networks reward the connected. The value of interconnection was never in any single line. It was in joining early enough to shape where the lines ran. The organizations that join a shared commercial layer early get their workflows, their vendors, and their data reflected in how it works. The ones that wait will eventually connect to something shaped around everyone else's operations.

And the grid does not have a decade to spare. We are being asked to execute the largest buildout since the railroad with a coordination system built for a slower century. Every quarter we keep running it on email threads is capital handed to intermediaries and schedule handed to delays.

The investors have the direction right. Agents are coming, and the value will flow through the layers that connect us. What they miss is that in our industry, that layer has to be built first. We'd rather build it than wait for it.

The physical layer of our industry figured out synchronization eighty years ago. It is time the commercial layer caught up.

Sean Dunham is the co-founder and CEO of Utilen, the Coordination Engine™ for the electric power industry, based in Minneapolis. The Union Square Ventures essay referenced above, "The Rebel Alliance," is available at blog.usv.com. For the engineering behind the synchronization point: NERC's Balancing and Frequency Control Reference Document describes North America's four major interconnections, each operating as a single synchronized system.

‍