Eric Steinberger

Eric Steinberger is an Austrian artificial intelligence researcher and technology executive. He is the co-founder and chief executive officer of Magic, the 2022 San Francisco artificial-intelligence company building foundation models for autonomous software engineering on the proprietary Long-Term Memory architecture, with Sebastian De Ro as chief technology officer. He was previously a deep-reinforcement-learning researcher at Meta AI Research, the co-founder of the climate-education non-profit ClimateScience, and a student researcher at TU Wien and the University of Cambridge.

At a glance

• Education: University of Cambridge, computer-science undergraduate program (departed after one year, 2019 to 2020) before founding ClimateScience and joining Facebook AI Research.
• Current role: Co-founder and chief executive officer of Magic since 2022.
• Key contributions: Author of "Single Deep Counterfactual Regret Minimization" (2019); co-author with Noam Brown of "DREAM: Deep Regret Minimization with Advantage Baselines and Model-Free Learning" (NeurIPS 2020); deep-reinforcement-learning researcher at Facebook AI Research from 2019 to 2021 working on imperfect-information-game agents; co-founder and chief executive of ClimateScience from 2019; co-founder and chief executive of Magic from 2022, leading the LTM architectural research line and the August 2024 announcement of the LTM-2-mini model with a 100-million-token context window.
• X / Twitter: @EricSteinb
• LinkedIn: ericsteinb
• GitHub: EricSteinberger
• Personal site: steinberger-ai.com
• Google Scholar: Eric Steinberger

Origins

Steinberger was born in Vienna, Austria. He has described, in podcast interviews including the Sequoia Capital Training Data episode 12 of August 2024, beginning self-directed study of physics and computer science through MIT OpenCourseWare at age 14, with a stated long-term commitment to artificial general intelligence research from that period. The early-teenage period overlapped with secondary school in Austria, where, by his own account, he wired the school's computer lab into an informal training cluster for machine-learning experiments.

The first publicly documented research affiliation is a 2017 student-research role at TU Wien (Vienna University of Technology) on robotics, followed by short student stints at MIT on digital agriculture and at the University of Luxembourg. The same period produced the PokerRL public framework on GitHub, the codebase that would later anchor his collaboration with Noam Brown at Facebook AI Research.

Career

Steinberger entered the University of Cambridge undergraduate computer-science program in 2019, departing after the first year to focus on Facebook AI Research and on ClimateScience. The Facebook AI Research role ran from 2019 through approximately 2021 and produced two of his most-cited papers, "Single Deep Counterfactual Regret Minimization" (2019, solo author) and "DREAM: Deep Regret Minimization with Advantage Baselines and Model-Free Learning" (2020, with Noam Brown and Adam Lerer). The work sat in the imperfect-information-game-theory line that Brown had pioneered with the Libratus and Pluribus poker systems, and Steinberger has publicly credited the FAIR period as foundational to his subsequent transition into language-model research.

Parallel to the FAIR work, Steinberger co-founded ClimateScience in 2019 with Isabel Key, a fellow Cambridge student. The non-profit, registered as a UK charity (number 1190993) and headquartered in London, produces free climate-education courses, videos, and children's books in multiple languages. Steinberger has retained the chief executive role through the Magic period, with the organization characterized in industry coverage as the world's largest climate-education platform.

In 2022 Steinberger co-founded Magic with Sebastian De Ro, the former chief technology officer of the German business-process-management firm FireStart. The pair had met through ClimateScience volunteering. The founding thesis was that long-context-window scaling, rather than parameter scaling alone, is the principal architectural lever for autonomous software-engineering capability and a credible path toward artificial general intelligence. The company is headquartered in San Francisco.

The Magic fundraising trajectory has been aggressive. The $23 million seed of summer 2023 was followed by a $117 million Series A in February 2024 led by NFDG Ventures (the Nat Friedman and Daniel Gross investing partnership) with CapitalG and Elad Gil. The $320 million round of August 2024 added Eric Schmidt, Atlassian, Jane Street, Sequoia Capital, and others, bringing cumulative capital to approximately $466 million at a reported $1.5 billion valuation. The August 2024 LTM-2-mini announcement coincided with the funding disclosure and reported a 100-million-token context window with a sequence-dimension algorithm approximately 1,000 times more efficient than transformer attention at that context size.

Affiliations

• TU Wien: Student researcher (robotics), 2017.
• MIT: Brief student-research role (digital agriculture), approximately 2018 to 2019.
• University of Cambridge: Computer-science undergraduate, 2019 to 2020 (departed after one year).
• Meta AI Research (FAIR): Deep-reinforcement-learning researcher, 2019 to approximately 2021.
• ClimateScience: Co-founder, chief executive officer, and trustee, 2019 to present.
• Magic: Co-founder and chief executive officer, 2022 to present.

Notable contributions

Steinberger's public work spans imperfect-information-game research, climate-education non-profit work, and the founding of Magic.

• "Single Deep Counterfactual Regret Minimization" (2019). Solo-authored paper introducing a single-network variant of the deep counterfactual-regret-minimization algorithm for imperfect-information games. The paper was the entry point for the FAIR collaboration.
• "DREAM: Deep Regret Minimization with Advantage Baselines and Model-Free Learning" (2020). Co-authored with Noam Brown and Adam Lerer at Facebook AI Research. Presented at the NeurIPS 2020 Deep Reinforcement Learning workshop. Cited approximately 80 times.
• "Chain of thought monitorability: A new and fragile opportunity for AI safety" (2025). Multi-author position paper co-signed by Steinberger alongside researchers from across the frontier-lab cohort, on the safety implications of chain-of-thought reasoning visibility.
• ClimateScience (2019). Co-founder and chief executive of the UK climate-education charity, with multilingual courses, videos, and children's-book content.
• Magic (2022). Co-founder and chief executive of the LTM-architecture lab. Public artifacts include the August 2024 LTM-2-mini announcement and the HashHop benchmark released alongside it.
• Public-talk record. No Priors episode 79 with Sarah Guo and Elad Gil (August 2024) on Magic's coding-AI thesis; Training Data episode 12 with Sonya Huang at Sequoia Capital (August 2024) on Magic's counterintuitive approach to AGI; ClimateScience YouTube channel content over the 2020 to 2025 period.

Investments and boards

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

• Magic (AI): Co-founder and chief executive officer, 2022 to present. Cumulative funding through April 2026 approximately $466 million at a reported $1.5 billion valuation.
• ClimateScience (Software / Education): Co-founder, chief executive, and trustee, 2019 to present. UK registered charity (1190993).

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

Network

Steinberger's longest-running professional partnership is with Sebastian De Ro, his Magic co-founder. The pair met through ClimateScience volunteering, and De Ro brought the German enterprise-software operating credential from FireStart that complemented Steinberger's research background. The Magic senior team includes Russ Salakhutdinov as chief scientist, the Carnegie Mellon professor of computer science and former director of AI research at Apple from 2016 to 2020.

The Facebook AI Research period from 2019 to 2021 produced the long-running collaboration with Noam Brown, formerly of Carnegie Mellon and FAIR, currently at OpenAI. Adam Lerer at FAIR was the third author on the DREAM paper. The ClimateScience partnership with Isabel Key is the other long-running professional relationship of public record.

The Magic investor base provides the broader network anchor. Eric Schmidt, Nat Friedman, Daniel Gross, Elad Gil, and partners at CapitalG, Sequoia Capital, Atlassian, and Jane Street have all participated in successive rounds. Among coding-AI peer founders, his Magic position runs in parallel with Jason Warner at poolside, Eric Ho at Goodfire, Misha Laskin at Reflection AI, and Aidan Gomez at Cohere, with overlap on the autonomous-software-engineering thesis with Sam Altman at OpenAI and Dario Amodei at Anthropic.

Position in the field

As of May 2026, Steinberger occupies a structurally distinctive position among insurgent-lab chief executives through the combination of the European-prodigy research origin, the Facebook AI Research deep-reinforcement-learning credential, the dual climate-education and AI-research career paths, and the LTM architectural bet at Magic. The career arc through TU Wien, MIT, Cambridge, and FAIR maps onto a research-first founder profile uncommon among the broader coding-AI insurgent cohort.

Industry coverage has consistently characterized Magic under Steinberger's leadership as one of the principal commercial coding-foundation-model labs globally, alongside Cursor (Anysphere), poolside, Cognition AI, Codeium / Windsurf, Reflection AI, and the in-house coding-AI products at Anthropic, OpenAI, and Google DeepMind. The competitive frame is unusual: Magic's LTM thesis differs architecturally from the transformer-and-scale orthodoxy that dominates the rest of the cohort, and the company has positioned its products toward autonomous engineering rather than developer-co-pilot augmentation.

The Magic valuation at the August 2024 round, approximately $1.5 billion, is moderate compared to peer insurgent labs (Safe Superintelligence, Thinking Machines Lab). The funding scale is sufficient to support the LTM research program but is below the largest insurgent rounds. The ClimateScience leadership produces an additional public-profile dimension uncommon among AI founders, with industry coverage frequently noting the dual mandate.

Outlook

Open questions over the next 6 to 18 months:

• LTM-3 and successor model release. Capability profile of the next-generation LTM model relative to peer commercial coding-AI offerings, and published benchmark positions on standard coding evaluations.
• Magic commercial product launch. Whether Magic launches a broadly available autonomous-coding product comparable to Cursor or GitHub Copilot, and how the LTM architectural premise translates to deployment economics.
• Atlassian strategic relationship. Whether the Atlassian investor position produces a commercial integration with the Atlassian developer-platform products (Jira, Bitbucket).
• Series-D successor financing. Whether Magic raises a follow-on round at a higher valuation, the lead investor profile, and the cumulative capital level relative to insurgent-cohort peers.
• ClimateScience scale. Whether the climate-education organization continues to expand under Steinberger's leadership while Magic's commercial intensity increases.
• Public-talk and research-paper cadence. Whether Steinberger continues the public-thesis schedule on the LTM premise and the AGI-via-coding line, and whether Magic publishes additional technical papers on the LTM architecture.

Sources

• Eric Steinberger LinkedIn profile. Public profile listing the Magic, ClimateScience, FAIR, and educational roles.
• Eric Steinberger Google Scholar profile. Citation metrics and published papers, listed under the cam.ac.uk verified email.
• Eric Steinberger personal site. Self-published research and biographical information.
• Eric Steinberger on X (@EricSteinb). Public X account.
• Eric Steinberger Sequoia Capital founder page. Lead-investor founder profile with biographical context.
• Founder Eric Steinberger on Magic's Counterintuitive Approach to Pursuing AGI. Sequoia Capital Training Data episode 12 (August 2024) hosted by Sonya Huang.
• No Priors Ep. 79 with Magic.dev CEO Eric Steinberger. August 30, 2024 podcast episode hosted by Sarah Guo and Elad Gil.
• Eric Steinberger Council for Inclusive Capitalism profile. Biographical context on the MIT OpenCourseWare self-study, the FAIR period, the Cambridge year, and the ClimateScience founding.
• Eric Steinberger PokerRL framework on GitHub. Public framework for multi-agent deep reinforcement learning in poker.
• Single Deep Counterfactual Regret Minimization paper. 2019 solo-authored paper.
• DREAM paper on arXiv. 2020 paper with Noam Brown and Adam Lerer at FAIR.
• Magic AI Secures $117 Million to Build an AI Software Engineer. Maginative coverage of the February 2024 Series A led by NFDG Ventures.
• TechCrunch: Generative AI coding startup Magic lands $320M investment. August 2024 funding-round coverage.
• Magic blog: 100M Token Context Windows. August 2024 LTM-2-mini announcement and HashHop benchmark.
• ClimateScience official site. Organizational details, founders, and history.
• Magic on Wikipedia. Wikipedia entry on the company history and founders.