Humanloop: The AI Platform That Transformed LLM Development Before Joining Anthropic
Humanloop was a British artificial intelligence company that helped businesses develop, test and monitor large-language-model applications. Its platform centred on prompt management, evaluation and observability, which became important as companies moved from chatbot experiments to dependable AI products.
Founded in 2020, it grew from a University College London spinout into an enterprise software provider. Customers included Gusto, Duolingo, Vanta, Filevine, Dixa, Fundrise, FMG and Athena. In August 2025, its team joined Anthropic and the platform closed.
Humanloop’s Legal Identity and Background
A UCL Spinout Founded in 2020
Humanloop began with research from the UCL AI Centre. The early team included machine-learning academics, doctoral researchers and technology entrepreneurs. UCL helped commercialise non-patented research, while Y Combinator accepted the startup into its Summer 2020 batch.
The original mission was to make artificial intelligence easier for business teams to use. Its methods helped software identify valuable examples and ask human experts for guidance, reducing manual work.
Humanloop Limited
The UK legal entity is Humanloop Limited, company number 12495436. Companies House records show incorporation on 3 March 2020 as a private limited company. Its registered office is Salisbury House, Station Road, Cambridge, CB1 2LA, with software development as its business classification.
The register continues to show the company as active. Raza Habib, Jordan Burgess and Peter Hayes remain directors, with no resignations shown. This does not mean the platform is still operating.
Quick Info: Humanloop
| Category | Details |
|---|---|
| Company | Humanloop Limited |
| Founded | 2020 |
| Founders | Raza Habib, Jordan Burgess and Peter Hayes |
| Industry | Artificial Intelligence and Software Development |
| Based in | Cambridge, United Kingdom |
| Main Product | LLM evaluation, prompt management and AI observability |
| Funding | Approximately $8 million |
| Key Customers | Gusto, Duolingo, Vanta, Filevine and Dixa |
| Anthropic Move | Founders and team joined Anthropic in 2025 |
| Current Status | Platform closed on 8 September 2025 |
| Website | humanloop.com |
The Founders of Humanloop
Raza Habib
Raza Habib was chief executive and co-founder. He completed a machine-learning PhD at UCL and studied physics at Cambridge. He worked as a founding engineer at Monolith AI and helped build speech systems at Google. Forbes included him in its 30 Under 30 Europe Technology selection in 2022.
Jordan Burgess
Jordan Burgess served as chief product officer and co-founder. His background includes machine learning and engineering at Cambridge and MIT. He previously worked at Amazon and was connected with Bloomsbury AI, later acquired by Facebook.
Peter Hayes
Peter Hayes was chief technology officer and co-founder. He completed a machine-learning PhD at UCL and previously co-founded fintech company Certua. His technical experience shaped the evaluation, tracing and developer tools.
Professors David Barber and Emine Yilmaz were also closely connected with the early academic work. The company identifies Habib, Burgess and Hayes as its three founders, while UCL describes the professors as part of the wider research team.
How the Platform Developed
From Data Labelling to LLM Tools
The first product, Programmatic, helped experts create rules for labelling large datasets. A lawyer, doctor or financial specialist could describe patterns, and the software could apply those rules across much larger collections of data.
By 2022, the business had moved towards large language models. It introduced production monitoring, prompt experiments, model comparisons and fine-tuning, allowing teams to measure feedback and improve applications with evidence.
Prompt Management and Evaluation
The later platform allowed organisations to store prompts, create versions, compare changes and control movement from testing to production. Product managers and subject experts could work beside engineers in one workspace.
Its evaluation system acted like testing for AI. Teams could use code checks, LLM-as-a-judge systems or human reviewers. Versioned datasets allowed comparisons with earlier results, while CI/CD connections helped detect problems before release.
Observability, Agents and Security
Humanloop recorded model calls, prompts, outputs, feedback, costs, token use and response times. Tracing followed retrieval systems, external tools and multi-step agents. The platform also supported structured output, function calling, several model providers and custom models.
For enterprise customers, it promoted role-based access, SSO and SAML, virtual private cloud deployment, US and European hosting, GDPR and HIPAA support. It stated that customer data was encrypted and not used to train models. These are historical features.
Funding and Investors
Investment History
UCL Technology Fund backed the company in 2020, followed by Y Combinator support. In July 2022, it completed a $2.6 million seed round led by Index Ventures, with participation from Y Combinator, LocalGlobe and AlbionVC.
In November 2024, the company stated that total funding had reached $8 million. TechCrunch later cited PitchBook’s $7.91 million figure, making approximately $8 million the safest conclusion.
Humanloop’s Filed Financial Position
Latest Available Accounts
The latest accounts cover the year ending 31 December 2024 and were filed in September 2025. The profit-and-loss account was not submitted, so revenue and net profit cannot be confirmed.
The accounts showed fixed assets of £9,403, current assets of £674,079, cash of £250,708, and net assets of £394,437. Creditors due within one year stood at £289,045. Retained losses were £2,727,227, while the share premium account stood at £3,121,555. The company had an average of five employees, including directors.
The comparative figures cover only nine months of 2023, so they are not directly comparable with a full year. The accounts state that a parent company provided financial support, but do not name it.
Customers and Business Impact
Enterprise Use Cases
It worked with companies developing customer support assistants, legal technology, financial services tools, education products and internal systems. Gusto used it while improving its AI support agent. Duolingo involved language experts, while Filevine used legal specialists.
Customer studies state that Gusto tripled AI deflection rates and saved millions in support costs. Filevine launched six AI products in one year, Dixa became three times faster, FMG claimed a 93 per cent cost reduction, and Athena used the platform across more than 600 executive assistants and 200 projects.
These figures come from company-produced customer studies and were not independently audited. They show the business problem the platform addressed: making AI more reliable and measurable.
The Anthropic Transition
A Talent-Focused Deal
On 13 August 2025, Humanloop announced that its team was joining Anthropic. Raza Habib, Jordan Burgess and Peter Hayes moved across with engineers and researchers. TechCrunch stated that Anthropic did not acquire the assets or intellectual property, and the financial terms were not disclosed.
This is best described as a talent-focused acquisition, often called an acqui-hire. Y Combinator marks the company as acquired, while Humanloop documents use “following our acquisition”. There is no evidence that Anthropic purchased the UK company itself. The legal structure and shareholder treatment remain private.
Platform Closure and Current Status
Humanloop stopped billing on 30 July 2025 and scheduled closure for 8 September 2025. It told users that accounts, prompts, versions, logs, evaluations and settings would be permanently deleted. An export tool was provided, with Langfuse and Braintrust recommended as alternatives.
As of August 2026, the website and documentation remain online mainly as historical records. The shutdown date has passed, and no live customer service should be assumed. Humanloop Limited remains active at Companies House, while its founders are connected with Anthropic.
Final Assessment
Humanloop was a British AI startup with strong academic roots, respected investors and a focus on enterprise reliability. It developed from data labelling into an environment for prompt engineering, evaluation and production monitoring.
Overall, Humanloop Limited is the UK company behind the brand, its platform is closed, and its founders joined Anthropic. Deal value, ownership and US-company links remain private.
Also Read
