Guild Raises $44M to Build the Agent Control Plane

Every company building with AI agents hits the same wall. The first agent works great. Then there are five, then fifty, and suddenly nobody knows what they're doing, what they cost, or how to shut one down before it burns through the quarterly budget overnight.

James Everingham, CEO of Guild.ai, just closed $44M in total funding to fix exactly that. In this TBPN interview, he breaks down what the former Meta Dev Infra team is seeing from early customers — and why the control problem is landing on every engineering org's desk at the same time.

Everingham's framing is blunt:

"It's like gremlins. The first one's fine until they start multiplying and taking over and pulling levers in your infrastructure."

The pattern he's describing is already familiar to anyone running agents in production. Last year was experimentation — one agent, an interesting demo. Now companies are scaling, and the problems that follow are predictable: security wants to know what the agents can access, compliance needs audit trails, and the infrastructure team needs a deterministic layer sitting underneath non-deterministic systems.

The timing is striking. Everingham says it's not just one segment of the market. Small companies and large companies, tech-forward and traditional — they're all arriving at the same phase simultaneously. The models got powerful enough to operate tools, and now everyone is staring at the same governance gap.

Highly regulated businesses are slower by design, but even there the calculus is shifting. The competitive pressure to move fast is outweighing the instinct to wait, as long as there's a control layer to reduce the risk.

The top request from early customers is not sophisticated orchestration or fancy agent design. It's cost visibility.

"One engineer blew through their entire budget in 12 hours and they didn't know."

The story is specific: a company had a monthly inference budget built for single-player use. Engineers were running agents on separate servers or their laptops, with no centralized view. One person torched the whole month's spend before lunch. Circuit breakers — the ability to say "turn this thing off if it gets too hungry" — are table stakes.

But it's not just raw spend. Companies want to measure what they're getting for that spend. Impact measurement, not just cost reporting.

The other pattern Everingham is seeing: companies building small internal versions of an agent control plane themselves, then realizing it's not their core competence. They don't want to maintain it. They don't have the specialized expertise to do it well. And vendor neutrality matters — nobody wants to lock into a single model provider when the competition is leapfrogging every few months.

A few use cases from early customers stand out. One team built an onboarding agent that makes the codebase sentient — say "check out this feature" instead of "check out these files," get a system diagram, have a conversation with the codebase. Another built a risk analysis agent that analyzes diffs for likelihood of taking the system down, letting low-risk changes through automatically and blocking high-risk ones. That kind of tooling can eliminate the holiday code freeze entirely.

Then there's the natural language interface. Everingham described typing a plain-English instruction into Guild:

"Hey, don't make it so it's not possible for duplicate bugs to exist in our company. And it just went and did it behind the scenes."

The system installed the agent, configured it, and hooked it into the workflow — no engineering ticket required.

Guild is building something they call Agent Hub: a public surface where companies can discover, distribute, and deploy agents, structured a lot like GitHub.

Companies can keep internal agents private or push them externally. Third parties can distribute agents as products. Within minutes, someone can authorize an agent, set up a workspace, and have it running.

The founding team are self-described open source believers, and they're carrying that ethos into how the agent marketplace works. The bet is that agents need the same kind of discoverability and distribution layer that code got with GitHub — a place to find what works, fork it, and run it in your own environment.

The $44M in total funding — $30M in the latest round plus $14M in seed, both raised within four months — is led by GV (Google Ventures), with NFX, Kla Ventures, Scribble Ventures, Acrew Capital, and Web Investment Network participating.