# Isara (company)

> Source: https://aiwiki.ai/wiki/isara
> Updated: 2026-10-27
> Fact-checked: 2026-10-27
> Categories: AI Agents, AI Companies, Finance AI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Isara (company)." aiwiki.ai, 27 Oct 2026. https://aiwiki.ai/wiki/isara
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Isara** is a San Francisco artificial intelligence startup that is developing software to coordinate large numbers of [AI agents](https://aiwiki.ai/wiki/ai_agents) working together on analysis and forecasting problems. It was founded in June 2025 by Eddie Zhang and Henry Gasztowtt, whom The Wall Street Journal described in March 2026 as two 23-year-old AI researchers.[1] Zhang previously worked on safety research at [OpenAI](https://aiwiki.ai/wiki/openai), and Gasztowtt studied mathematics and computer science at the University of Oxford.[5][6] The company's website, isara.io, carries the name Isara Laboratories.[13]

In March 2026 The Wall Street Journal reported that Isara had raised $94 million and that OpenAI was backing the company at a $650 million valuation.[1] Other investors named in coverage of the financing were Amity Ventures, Michael Ovitz and Stanley Druckenmiller.[2][3] The Journal reported that Isara had demonstrated its software coordinating 2,000 AI agents to forecast the price of gold at a tech conference.[1] Coverage of the financing said the company planned to sell predictive modelling software to investment firms first.[2][3]

## Overview

| Field | Detail |
| --- | --- |
| Name | Isara (website title: Isara Laboratories)[13] |
| Founded | June 2025[1] |
| Founders | Eddie Zhang, Henry Gasztowtt[1] |
| Headquarters | San Francisco, California[1] |
| Focus | Software that lets large numbers of AI agents communicate and divide work on complex problems[1][2] |
| Initial market | Investment firms (predictive modelling software)[2][3] |
| Reported funding | $94 million (reported March 2026)[1] |
| Reported valuation | $650 million (reported March 2026; not specified as pre-money or post-money)[1] |
| Investors named in coverage | OpenAI, Amity Ventures, Michael Ovitz, Stanley Druckenmiller[1][2][3] |
| Staff | About a dozen researchers hired besides the founders, as of March 2026[1] |
| Website | isara.io[13] |

## History

### Research before the company

Zhang and Gasztowtt worked together on AI-based policymaking research before they founded Isara. Both are authors of "Position: Social Environment Design Should be Further Developed for AI-based Policy-Making", a position paper published in the proceedings of [ICML](https://aiwiki.ai/wiki/icml) 2024.[7] Zhang, listed with a Harvard University affiliation, was the first author, and Gasztowtt was listed with an Oxford University affiliation. The seven other authors included David C. Parkes, Milind Tambe and Yiling Chen of Harvard and Stephan Zheng.[7] The paper introduces Social Environment Design, a general framework for using AI in automated policymaking that connects [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning), economics and computation, and computational social choice. The framework includes voting on policy objectives and is meant to support analysis of government and economic policy through AI simulation.[7]

TechFundingNews and The Next Web both described this ICML 2024 paper, which explored how AI systems could cooperate to improve policymaking, as the intellectual foundation for Isara.[2][3] Christ Church, Gasztowtt's Oxford college, wrote that the company's work "builds on research into how networks of AI systems can cooperate to tackle complex decision-making challenges".[5]

The two continued to publish together in 2024. Gasztowtt was the first author and Zhang the last author of "Large Legislative Models: Towards Efficient AI Policymaking in Economic Simulations" (October 2024). The paper proposes using pre-trained [large language models](https://aiwiki.ai/wiki/large_language_model) as sample-efficient policymakers in socially complex multi-agent reinforcement learning scenarios and reports efficiency gains over existing methods across three environments.[8] Both were also among 51 authors of a December 2024 proposal for a cooperative AI policymaking platform. It outlined a multimodal foundation model combining economic time-series data with natural-language policy data, mechanisms for eliciting diverse perspectives, and a web platform for data-driven policymaking.[9] Zhang's personal website lists Gasztowtt among students he has mentored.[6]

### Founding

Isara, a San Francisco company, was founded in June 2025.[1] By March 2026 the founders had hired about a dozen other researchers from companies including [Google](https://aiwiki.ai/wiki/google), Meta Platforms and OpenAI, according to The Wall Street Journal.[1] The Next Web, writing nine months after the founding, reported that the company had no product in market.[2]

### 2026 financing

On March 25, 2026, The Wall Street Journal published an exclusive report, headlined "OpenAI Backs New AI Startup Seeking Bot Army Breakthroughs", on Isara's financing. It stated that Isara had recently raised $94 million from investors, with OpenAI backing the company at a $650 million valuation.[1] The Journal described the company's goal as building software that allows AI agents to communicate and solve complex problems in industries such as finance and biotech.[1]

Follow-up coverage named Amity Ventures, Michael Ovitz (described by The Next Web as the former Creative Artists Agency chairman) and hedge fund manager Stanley Druckenmiller among the investors.[2][3] Reports differed on OpenAI's role. VKTR wrote that OpenAI led the investment, while The Next Web described OpenAI as being among the investors.[2][4] Neither The Next Web nor TechFundingNews named the round type or gave a separate figure for OpenAI's own contribution. In its analysis section, The Next Web wrote that "investing $94 million at a $650 million valuation" was inexpensive insurance for OpenAI, using the size of the whole round.[2][3]

On March 30, 2026, Christ Church, Oxford, published a news item about Gasztowtt's role as a co-founder. It repeated the $94 million raise and the $650 million valuation.[5]

## Founders

| Founder | Education | Earlier work |
| --- | --- | --- |
| Eddie Zhang (publishes as Edwin Zhang) | BS and MS in computer science, University of California, Santa Barbara; left a computer science PhD at Harvard[6] | Safety research at OpenAI from July 2024; visiting researcher at the MIT-IBM Watson AI Lab, December 2022 to May 2023[6] |
| Henry Gasztowtt | BA in Mathematics and Computer Science, Christ Church, Oxford (matriculated 2021, graduated 2025)[5] | AI policymaking research with Zhang (2024)[7][8] |

### Eddie Zhang

Zhang's personal website says he studied computer science at the University of California, Santa Barbara, earning a bachelor's degree (2019 to 2022) and a master's degree (2022 to 2023). It also lists a visiting research position at the MIT-IBM Watson AI Lab from December 2022 to May 2023 and a research internship at Meta in 2022.[6] He began a computer science PhD at [Harvard University](https://aiwiki.ai/wiki/harvard_university) in September 2023 and, according to the site, later dropped out.[6] The site lists safety research at OpenAI beginning in July 2024 and describes his work there as covering "safety, alignment, and applying AI for social good".[6] TechFundingNews described him as a former OpenAI AI safety researcher who had held a visiting research position at MIT.[3]

Zhang was the first author of "Transcendence: Generative Models Can Outperform The Experts That Train Them", presented at [NeurIPS](https://aiwiki.ai/wiki/neurips) 2024.[10] The paper trains an autoregressive transformer to play chess from game transcripts. It reports that the model can sometimes outperform every player in its training data, and that low-temperature sampling can enable this effect.[10]

Zhang is listed with an OpenAI affiliation as one of four authors of "Why Language Models Hallucinate" (September 2025), with Adam Tauman Kalai, Ofir Nachum and Santosh Vempala. The paper argues that language models [hallucinate](https://aiwiki.ai/wiki/hallucination) because training and evaluation procedures reward guessing over acknowledging uncertainty.[11] He was also one of four authors of OpenAI's August 2025 post "Collective alignment: public input on our Model Spec". For that project OpenAI surveyed more than 1,000 people about how its models should behave and compared their views with the [Model Spec](https://aiwiki.ai/wiki/model_spec).[12]

### Henry Gasztowtt

Gasztowtt matriculated at Christ Church, Oxford, in 2021 and graduated in 2025 with a BA in Mathematics and Computer Science, according to the college.[5] Coverage of the 2026 financing described him as a computer science student at Oxford.[2][3] Before Isara he was first author of the "Large Legislative Models" paper and a co-author of the Social Environment Design position paper with Zhang.[7][8]

## Technology and plans

Public information about Isara's system comes mostly from press coverage of the 2026 financing. The Wall Street Journal described the product goal as software that allows AI agents to communicate and solve complex problems.[1] The Next Web reported that the architecture is meant to let hundreds or thousands of specialised agents communicate, divide tasks, align on goals and produce a combined output, and that the founders describe this as a shift from isolated tools to coordinated teams.[2] Christ Church described the approach as part of a move from single AI systems toward coordinated networks of specialised agents, a field it said is often described as [multi-agent AI](https://aiwiki.ai/wiki/multi_agent_system).[5]

According to the Journal, Isara demonstrated its software coordinating 2,000 AI agents to forecast the price of gold at a tech conference.[1] Christ Church described the same work as early versions of the technology coordinating around 2,000 AI agents to analyse and predict market trends.[5]

| Area | What was reported |
| --- | --- |
| First commercial market | Investment firms, as buyers of predictive modelling software[2][3] |
| Secondary markets | Biotechnology and geopolitical analysis[2][3] |
| Longer-term goal | Training agent swarms to track geopolitical shifts and forecast economic trends, as described to The Wall Street Journal[2] |
| Product status (March 2026) | No product in market[2] |

As of October 2026 the isara.io homepage showed only the name "Isara Laboratories", a "Limited access" status and the note "Materials are shared privately."[13]

## Funding

| Reported | Amount | Valuation | Investors named | Sources |
| --- | --- | --- | --- | --- |
| March 2026 | $94 million | $650 million (not specified as pre-money or post-money) | OpenAI, Amity Ventures, Michael Ovitz, Stanley Druckenmiller | The Wall Street Journal; The Next Web; TechFundingNews[1][2][3] |

## Reception and classification

The Next Web called the technical challenge "real". It wrote that getting thousands of agents to communicate without cascading errors, conflicting objectives or compounding hallucinations is a problem the academic literature has barely begun to address at scale. It added that frameworks such as [LangChain](https://aiwiki.ai/wiki/langchain), [CrewAI](https://aiwiki.ai/wiki/crewai) and [AutoGen](https://aiwiki.ai/wiki/autogen) typically coordinate small numbers of agents on relatively structured tasks.[2] The same article argued that the gap between a demonstration forecasting gold prices and a production system that investment firms would rely on for capital allocation was large. It interpreted OpenAI's investment as a way to gain exposure to outside approaches to multi-agent coordination and to keep a relationship with Zhang and his team.[2] VKTR concluded that the harder task for Isara was proving the technology works at the scale its valuation implies.[4]

The Next Web placed Isara in a pattern that The Information has labelled "[neolabs](https://aiwiki.ai/wiki/neolab)": research-heavy AI startups founded by alumni of large AI labs.[2] Debarghya Das's "The Ultimate Neolab List" includes Isara but marks its inclusion as "Borderline", with the explanation "Agent-orchestration software for finance; smaller raise and less compute-heavy than a model lab."[14]

## See also

- [List of neolabs](https://aiwiki.ai/wiki/list_of_neolabs)
- [Neolab](https://aiwiki.ai/wiki/neolab)
- [Multi-agent system](https://aiwiki.ai/wiki/multi_agent_system)
- [Swarm intelligence](https://aiwiki.ai/wiki/swarm_intelligence)
- [AI in finance](https://aiwiki.ai/wiki/ai_in_finance)
- [Agentic AI](https://aiwiki.ai/wiki/agentic_ai)

## References

1. Berber Jin, ["OpenAI Backs New AI Startup Seeking Bot Army Breakthroughs"](https://www.wsj.com/tech/ai/openai-backs-new-ai-startup-seeking-bot-army-breakthroughs-a0b1fedc), The Wall Street Journal, March 25, 2026.
2. Cristian Dina, ["OpenAI backs a nine-month-old startup building swarms of AI agents at a $650 million valuation"](https://thenextweb.com/news/openai-isara-ai-agent-swarms-650m-valuation), The Next Web, March 27, 2026.
3. Vignesh R, ["OpenAI backs Isara's $94M round at $650M valuation to build 'AI agent swarms'"](https://techfundingnews.com/openai-backs-isara-ai-agents-94-million-funding/), TechFundingNews, March 27, 2026.
4. Michelle Hawley, ["OpenAI-Backed Isara Raises $94M to Build AI Agent Swarms"](https://www.vktr.com/ai-news/openai-backed-isara-raises-94m-to-build-ai-agent-swarms/), VKTR, March 27, 2026.
5. Christ Church, University of Oxford, ["Christ Church graduate co-founds startup backed by OpenAI"](https://www.chch.ox.ac.uk/news/christ-church-graduate-co-founds-startup-backed-openai), March 30, 2026.
6. Eddie Zhang, ["Eddie Zhang"](https://eddie.win/) (personal website), accessed October 27, 2026.
7. Edwin Zhang, Sadie Zhao, Tonghan Wang, Safwan Hossain, Henry Gasztowtt, Stephan Zheng, David C. Parkes, Milind Tambe and Yiling Chen, ["Position: Social Environment Design Should be Further Developed for AI-based Policy-Making"](https://proceedings.mlr.press/v235/zhang24cl.html), Proceedings of the 41st International Conference on Machine Learning, PMLR 235:60527-60540, 2024. Preprint: [arXiv:2402.14090](https://arxiv.org/abs/2402.14090).
8. Henry Gasztowtt, Benjamin Smith, Vincent Zhu, Qinxun Bai and Edwin Zhang, ["Large Legislative Models: Towards Efficient AI Policymaking in Economic Simulations"](https://arxiv.org/abs/2410.08345), arXiv:2410.08345, October 10, 2024.
9. Aiden Lewington et al., ["Creating a Cooperative AI Policymaking Platform through Open Source Collaboration"](https://arxiv.org/abs/2412.06936), arXiv:2412.06936, December 9, 2024.
10. Edwin Zhang, Vincent Zhu, Naomi Saphra, Anat Kleiman, Benjamin L. Edelman, Milind Tambe, Sham M. Kakade and Eran Malach, ["Transcendence: Generative Models Can Outperform The Experts That Train Them"](https://neurips.cc/virtual/2024/poster/94284), NeurIPS 2024 poster. Preprint: [arXiv:2406.11741](https://arxiv.org/abs/2406.11741).
11. Adam Tauman Kalai, Ofir Nachum, Santosh S. Vempala and Edwin Zhang, ["Why Language Models Hallucinate"](https://arxiv.org/abs/2509.04664), arXiv:2509.04664, September 4, 2025.
12. OpenAI (Tyna Eloundou, Mitchell Gordon, Eddie Zhang and Sandhini Agarwal), ["Collective alignment: public input on our Model Spec"](https://openai.com/index/collective-alignment-aug-2025-updates/), August 27, 2025.
13. Isara Laboratories, ["Isara Laboratories"](https://isara.io/) (company website), accessed October 27, 2026.
14. Debarghya Das, ["The Ultimate Neolab List"](https://debarghyadas.com/neolabs/), accessed October 27, 2026.