Anthropic scoops up Humanloop team for enterprise AI tools

AI talent wars just took another corporate turn, and this time, the software didn't even make the cut.

Anthropic has scooped up the co-founders and most of the team behind Humanloop, an enterprise platform built for prompt management, LLM evaluation, and observability. But it is an acqui-hire in the purest sense: Anthropic confirmed it did not acquire Humanloop's assets or intellectual property, opting instead to bring over the people who built it.

Humanloop's three co-founders — CEO Raza Habib, CTO Peter Hayes, and CPO Jordan Burgess — are now at Anthropic, along with around a dozen engineers and researchers.

An acqui-hire for brains, not IP

Humanloop started in 2020 as a spinout from University College London before going through Y Combinator and the Fuse Incubator. According to PitchBook, the startup raised $7.91 million across two seed rounds led by YC and Index Ventures.

Over the past few years, the team built a solid reputation serving enterprise names like Duolingo, Gusto, and Vanta, helping them test, fine-tune, and monitor production AI applications.

Last month, Humanloop informed customers that it would be shutting down in preparation for an acquisition. Anthropic left the software and code behind, focusing entirely on hiring the talent.

In generative AI, that math makes sense. Codebases get rebuilt quickly, but seasoned engineers who understand how enterprises actually test and evaluate safety at scale are rare.

Building out the enterprise plumbing

This move comes at a high-stakes moment for Anthropic's enterprise push.

Earlier this week, Anthropic struck a deal with the U.S. government's central purchasing arm to supply AI services across executive, judicial, and legislative branches for just $1 per agency for the first year. That was an explicit attempt to undercut OpenAI's similar budget push.

At the same time, Anthropic has been pushing longer context windows and leaning hard into agentic workflows and coding. But selling into federal agencies and enterprise giants takes more than raw model benchmarks.

These buyers need evaluation workflows, bias mitigation, and observability guardrails. As Brad Abrams, API product lead at Anthropic, noted, Humanloop's tooling and evaluation experience will be directly plugged into Anthropic's safety and utility efforts.

Why it matters

We think this deal highlights a clear shift in the AI landscape: raw intelligence is no longer enough to win the enterprise war; the developer tooling ecosystem is where the battle is heading.

For enterprise builders and developers, the immediate downside is that Humanloop's stand-alone product is winding down, forcing existing customers to migrate elsewhere. But long term, having seasoned evaluation specialists inside Anthropic means Claude's native developer dashboard and safety controls could become far more polished.

For the broader industry, this signals a squeeze on stand-alone AI middleware. If prompt tracking and LLM evaluation can simply be absorbed into base model providers, third-party tooling startups need to defend why they should exist independently.

Anthropic made a smart play by avoiding messy legacy software integrations and grabbing pure domain expertise. Now, the pressure is on Anthropic to prove that adding Humanloop's brains will translate into enterprise-ready features before OpenAI and Google outpace them in the same corporate boardrooms.