Misha Laskin

Misha Laskin is a Russian-born computer scientist, co-founder and chief executive officer of Reflection AI, and a former staff research scientist at Google DeepMind who led reward modeling for the Gemini project.
Nextomoro
8 min read
Misha Laskin

Misha Laskin

Misha Laskin is a Russian-born computer scientist and reinforcement-learning researcher. He is the co-founder and chief executive officer of Reflection AI, the 2024 San Francisco startup building autonomous coding agents and a planned open-weights frontier language model. He was previously a staff research scientist at Google DeepMind where he led the reward-modeling effort for Gemini, and a postdoctoral scholar at the University of California, Berkeley under Pieter Abbeel on reinforcement-learning research that produced the CURL and RAD papers.

At a glance

Origins

Laskin was born in Russia and emigrated with his family to Israel at age one. His parents, both chemists who had completed PhDs at Hebrew University of Jerusalem, moved the family to Washington State when he was nine. He has described his upbringing in the Tri-Cities area of southeastern Washington, the region built around the Hanford Site of the World War II Manhattan Project, as the source of his early interest in physics and in the published lectures of Richard Feynman.

He completed undergraduate study at Yale University with a double major in physics and literature, and then a PhD in theoretical many-body quantum physics at the University of Chicago.

Career

Laskin's first venture after the Chicago PhD was a Y Combinator-backed startup that built inventory-prediction software for retailers planning seasonal apparel runs. He has described the period as a consulting business that produced revenue without a true product, and as a transitional phase during which he began contacting researchers at OpenAI about deep-learning methodology. Those contacts introduced him to Pieter Abbeel at UC Berkeley, who took him on as a postdoctoral scholar in September 2019.

The Berkeley postdoc under Abbeel from September 2019 through February 2022 produced Laskin's principal academic publication record on visual reinforcement learning. With Aravind Srinivas and Abbeel, he co-led the April 2020 CURL paper at ICML 2020, which introduced contrastive unsupervised representation learning for sample-efficient reinforcement learning from pixels. With Kimin Lee, Adam Stooke, Lerrel Pinto, and Abbeel, he co-led the April 2020 Reinforcement Learning with Augmented Data paper at NeurIPS 2020, which demonstrated that simple data augmentations could produce state-of-the-art results across the DeepMind Control Suite.

In February 2022 Laskin joined Google DeepMind as a staff research scientist, working initially on the General Agents team led by Volodymyr Mnih on agent-research methodology. The principal academic artifact from this period is the October 2022 In-context Reinforcement Learning with Algorithm Distillation paper, presented at ICLR 2023 with Laskin as lead author. Algorithm Distillation introduced a method for distilling reinforcement-learning algorithms into causal-transformer sequence models that learn entirely in-context without parameter updates. Following the ChatGPT release in late 2022, Laskin transitioned within DeepMind to the RLHF program for Gemini, where he led the reward-modeling effort that trains the preference models underlying the post-training pipeline.

In March 2024 Laskin and Antonoglou, the AlphaGo and AlphaZero co-developer with whom he had collaborated on Gemini's RLHF stack, departed to co-found Reflection AI. The company operated in stealth through 2024 and emerged in March 2025 with $130 million in cumulative funding from a $25 million seed round and a follow-on $105 million Series A co-led by Lightspeed Venture Partners and CRV. The first product, Asimov, is an autonomous coding agent that ingests not only source code but also project documentation, internal communications, and engineering notes to build a model of how a software system was developed.

In October 2025 Reflection AI raised a $2 billion Series B at an $8 billion valuation, led by NVIDIA with $800 million invested and additional participation from Lightspeed, Sequoia Capital, and Eric Schmidt. The valuation represented an approximately fifteenfold step from the company's previous $545 million level seven months earlier. Laskin used the public communications around the round to position Reflection AI as "America's open frontier AI lab," with the planned 2026 release of an open-weights frontier language model framed as a US-domiciled response to DeepSeek.

Affiliations

Notable contributions

Laskin's body of public work centers on visual reinforcement learning at Berkeley, agent research and reward modeling at Google DeepMind, and the founding of Reflection AI.

  • CURL: Contrastive Unsupervised Representations for Reinforcement Learning (April 2020). Co-lead author with Aravind Srinivas and Pieter Abbeel at UC Berkeley. Presented at ICML 2020. CURL introduced contrastive learning into the reinforcement-learning pipeline for sample-efficient training from pixels and produced the first image-based algorithm to nearly match the sample efficiency of methods using state-based features on the DeepMind Control Suite.
  • Reinforcement Learning with Augmented Data (April 2020). Co-lead author with Kimin Lee, Adam Stooke, Lerrel Pinto, and Pieter Abbeel. Presented at NeurIPS 2020. The paper demonstrated that simple data-augmentation techniques applied to existing reinforcement-learning algorithms could match or exceed prior state-of-the-art across fifteen environments in the DeepMind Control Suite.
  • In-context Reinforcement Learning with Algorithm Distillation (October 2022). Lead author with thirteen co-authors at Google DeepMind including Volodymyr Mnih and Satinder Singh. Presented at ICLR 2023. Algorithm Distillation introduced a method for distilling reinforcement-learning algorithms into causal-transformer sequence models that learn in-context without parameter updates.
  • Gemini reward modeling (2022 to 2024). Reward-modeling research lead at Google DeepMind on the RLHF stack underlying the Gemini family of frontier models, working alongside Antonoglou on the broader RLHF program.
  • Reflection AI founding (March 2024). Co-founder and chief executive officer alongside Antonoglou. The company has shipped the Asimov autonomous coding system, raised approximately $2.13 billion across multiple rounds, and announced a planned 2026 open-weights frontier-model release positioned as a US-domiciled alternative to DeepSeek.
  • Asimov (November 2024). Reflection AI's first product, an autonomous coding system that builds a software-development model from source code, project documentation, and internal communications.
  • Public-talk record. No Priors interview with Sarah Guo and Elad Gil on Asimov and Reflection AI's research direction; Manifold conversation with Steve Hsu on the path from theoretical physics to superintelligence; Sequoia Capital Training Data podcast on the AlphaGo moment for LLMs.

Investments and boards

The entries below are limited to AI, semiconductors, datacenters, software, and energy.

  • Reflection AI (AI): Co-founder and chief executive officer, March 2024 to present. San Francisco-based autonomous-AI lab. Cumulative funding approximately $2.13 billion through April 2026, including a $25 million seed round, a $105 million Series A co-led by Lightspeed Venture Partners and CRV in early 2025, and a $2 billion Series B at an $8 billion valuation in October 2025 led by NVIDIA.

No other public investor activity on record in AI, semiconductors, datacenters, software, or energy as of May 2026.

Network

Laskin's longest-running professional relationship is with his Berkeley postdoctoral advisor Pieter Abbeel, with whom he co-authored the CURL and RAD papers from 2019 to 2022. Aravind Srinivas (CURL co-lead) and Adam Stooke and Kimin Lee (RAD co-authors) are part of the same Abbeel-led research environment that also produced John Schulman's prior generation of policy-gradient research.

His Google DeepMind period from February 2022 through March 2024 produced his closest current professional relationship with Antonoglou, the AlphaGo and AlphaZero co-developer who is now his Reflection AI co-founder. Volodymyr Mnih, the General Agents team lead and a co-author on the Algorithm Distillation paper, is the senior collaborator from his early DeepMind period. The Gemini RLHF program connected him to senior DeepMind research staff including Koray Kavukcuoglu, Demis Hassabis, and Shane Legg.

Among Insurgent-lab founder peers, his Reflection AI position runs in parallel with Mira Murati and Barret Zoph at Thinking Machines Lab, Ilya Sutskever at Safe Superintelligence, and Dario Amodei at Anthropic.

Position in the field

As of May 2026, Laskin occupies a structurally distinctive position among Insurgent-lab chief executives through the combination of a theoretical-physics doctoral background, a Y Combinator startup phase that preceded his entry into AI research, the Berkeley postdoc under Abbeel, the DeepMind staff-scientist period, and the rapid Reflection AI valuation acceleration to $8 billion in seven months.

Industry coverage has characterized Laskin as the public face of the "America's open frontier AI lab" framing, with Antonoglou as the AlphaGo-credentialed research counterpart. The October 2025 NVIDIA-led round placed Reflection AI among the highest-valuation Insurgent labs in the 2024 to 2025 cohort, behind only the largest scale (Safe Superintelligence, Thinking Machines Lab) on capital base.

The physics-PhD-to-AI arc is more common among senior frontier researchers than the headline framing suggests. John Schulman (Caltech physics undergrad, Berkeley EECS PhD) and John Jumper (Vanderbilt physics undergrad, Chicago theoretical chemistry PhD) share the trajectory of a physical-sciences quantitative training that preceded a transition into deep learning during the 2015 to 2020 period. Laskin's record is distinctive on the entrepreneurial detour between PhD and Berkeley, and on the speed of the trajectory from postdoc through DeepMind into a Series B at frontier valuation.

Outlook

Open questions over the next 6 to 18 months:

  • Open-weights frontier-model release. The 2026 commitment to release a US-domiciled open-weights frontier model is the central public milestone. Release timing, capability profile against DeepSeek successors, parameter count, and licensing terms will shape the durability of the "America's open frontier AI lab" framing.
  • Asimov adoption and revenue traction. The autonomous-coding product line provides commercial validation ahead of the frontier-model release. Customer base, deal sizes, and competitive position against OpenAI Codex, Anthropic Claude Code, and Cursor will inform whether the autonomous framing differentiates from the broader coding-agent market.
  • Senior research talent recruitment. Continued movement of reinforcement-learning researchers and post-training specialists from DeepMind, OpenAI, and Anthropic into Reflection AI, which had approximately 60 staff at the October 2025 round.
  • Follow-on financing. Reflection AI was reported in March 2026 to be seeking new investors at a valuation above $20 billion, which if confirmed would mark a further step from the October 2025 round.
  • Public commentary cadence. Frequency and substance of podcast and conference appearances on reinforcement learning and US open-source frontier-AI policy as Reflection AI moves toward its first model release.
  • DeepSeek competition. Whether the planned Reflection AI open-weights frontier model matches DeepSeek's V3, R1, and successor capability tier on standardized benchmarks at release, and whether the framing as a US-domiciled open-frontier counterweight produces durable strategic positioning.

Sources

Person Reflection AI
About the author
Nextomoro

Nextomoro

nextomoro tracks progress for AI research labs, models, and what's next.

Read next

Ioannis Antonoglou

Alexander Amini

Mathias Lechner

Ramin Hasani

Daniela Rus

AI Research Lab Intelligence

nextomoro tracks progress for AI research labs, models, and what's next.

AI Research Lab Intelligence

Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to AI Research Lab Intelligence.

Your link has expired.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.