Humanloop: an LLM evals platform shut down when its team joined Anthropic
Humanloop, a YC-backed platform for prompt management and LLM evaluation, told customers in July 2025 it would sunset. Its founders and about a dozen engineers joined Anthropic, which did not buy the product.
- Product
- Humanloop
- Founder
- Raza Habib, Peter Hayes and Jordan Burgess
- Launched
- 2020
- Model
- B2B subscriptions for LLM evaluation and prompt management (discontinued)
Published · 2 min read · 5 sources
Key takeaways
- Humanloop was a 2020 University College London spinout that went through Y Combinator and raised $7.91 million in seed funding, per PitchBook data cited by TechCrunch.
- It emailed all users on 17 July 2025 about a shutdown, stopped billing on 30 July and sunset the platform on 8 September 2025.
- Its three co-founders and around a dozen engineers and researchers joined Anthropic; an Anthropic spokesperson said it did not acquire Humanloop's assets or IP.
What it is
Humanloop was a platform for prompt management, LLM evaluation and observability (TechCrunch). Its founders describe it as "the first development platform for LLM applications" (Humanloop). Enterprise customers included Duolingo, Gusto and Vanta (TechCrunch).
Timeline
- 2020: Humanloop is founded as a University College London spinout, then goes through Y Combinator and the Fuse Incubator. It raises $7.91 million in seed funding across two rounds led by YC and Index Ventures, per PitchBook (TechCrunch).
- 17 July 2025: Humanloop emails all users that it is shutting down (Humanloop migration guide). A user posts the email to Hacker News: Humanloop "has entered into a process to be acquired" and will sunset the platform on 8 September (Hacker News).
- 30 July 2025: All billing stops (Humanloop changelog).
- 13 August 2025: TechCrunch reports that CEO Raza Habib, CTO Peter Hayes and CPO Jordan Burgess, plus around a dozen engineers and researchers, have joined Anthropic. An Anthropic spokesperson confirms it did not acquire Humanloop's assets or IP (TechCrunch).
- 8 September 2025: The platform is sunset. "After this date, the platform and all data will be permanently inaccessible." (Humanloop changelog)
Why it worked
- Early to a real problem. Humanloop built evaluation and prompt tooling for LLM apps before most teams had them, and won enterprise customers like Duolingo (TechCrunch).
- The team was the asset. Anthropic's Brad Abrams said their "proven experience in AI tooling and evaluation will be invaluable" (TechCrunch).
What to learn
- An acqui-hire can be a good outcome for founders and a bad one for customers. Users got about seven weeks from the first email to permanent deletion (Humanloop migration guide).
- If you build on a small vendor, keep an export path ready. Humanloop shipped an export tool, but "all data will be permanently inaccessible" after the deadline (Humanloop changelog).
- Tooling that sits close to the model provider is exposed. Compare Inflection Pi and Adept, two other AI startups whose teams went to bigger companies.
Risks and what didn't work
- "Acquisition" versus acqui-hire. Humanloop's own notices say "following our acquisition", while Anthropic says it bought neither assets nor IP (Humanloop changelog; TechCrunch).
- Customers had to migrate at short notice. The company says it gave "as much notice as possible, whilst working through legal constraints" (Humanloop migration guide).
- No public revenue. Humanloop never published revenue or customer counts in the sources above; only the seed total is reported (TechCrunch).
FAQ
Short answers drawn from the sections above.
What is Humanloop?
Humanloop was a platform for prompt management, LLM evaluation and observability (TechCrunch). Its founders describe it as "the first development platform for LLM applications" (Humanloop). Enterprise customers included Duolingo, Gusto and Vanta (TechCrunch).
Why did Humanloop work?
- Early to a real problem. Humanloop built evaluation and prompt tooling for LLM apps before most teams had them, and won enterprise customers like Duolingo (TechCrunch).
- The team was the asset. Anthropic's Brad Abrams said their "proven experience in AI tooling and evaluation will be invaluable" (TechCrunch).
What can indie builders learn from Humanloop?
- An acqui-hire can be a good outcome for founders and a bad one for customers. Users got about seven weeks from the first email to permanent deletion (Humanloop migration guide).
- If you build on a small vendor, keep an export path ready. Humanloop shipped an export tool, but "all data will be permanently inaccessible" after the deadline (Humanloop changelog).
- Tooling that sits close to the model provider is exposed. Compare Inflection Pi and Adept, two other AI startups whose teams went to bigger companies.
What didn't work for Humanloop, and what are the risks?
- "Acquisition" versus acqui-hire. Humanloop's own notices say "following our acquisition", while Anthropic says it bought neither assets nor IP (Humanloop changelog; TechCrunch).
- Customers had to migrate at short notice. The company says it gave "as much notice as possible, whilst working through legal constraints" (Humanloop migration guide).
- No public revenue. Humanloop never published revenue or customer counts in the sources above; only the seed total is reported (TechCrunch).
Sources
Every link cited above, in order of first use.
- 1.TechCrunch (techcrunch.com)
- 2.Humanloop (humanloop.com)
- 3.Humanloop migration guide (humanloop.com)
- 4.Hacker News (news.ycombinator.com)
- 5.Humanloop changelog (humanloop.com)
Related case studies
More in Developer tools →Parse: Facebook acquired it, then shut the hosted service down
Parse was a mobile backend-as-a-service that Facebook acquired in 2013. In January 2016 it announced a one-year wind-down, open-sourced Parse Server, and retired the hosted service on January 28, 2017.
Campsite: a $2.6M team-communication startup that wound down after two years
Campsite raised a seed round to fix scattered team chat, broadened from design teams to whole companies, and then shut down as its founders joined Notion. It later published its code.
Heroku: Salesforce paid $212 million, then ended the free tier
Orion Henry, James Lindenbaum, and Adam Wiggins built a Ruby platform-as-a-service that Salesforce acquired in 2010 for about $212 million in cash. In 2022 Heroku shut down its free plans, citing fraud and abuse, and pushed hobbyists toward paid tiers.