Diogo Almeida (TypeSafe AI)

Almeida earned his bachelor’s degree from Rensselaer Polytechnic Institute and later obtained a master’s degree from Georgia Tech. He has about a decade of AI research experience and previously worked at Google Brain.⁠[1][2]

Almeida worked at OpenAI for about four and a half years. On March 4, 2022, the paper “Training language models to follow instructions with human feedback” (the InstructGPT paper) was submitted to arXiv, with Almeida listed as the fourth author; he is also one of the authors of the GPT-4 Technical Report and worked on the RLHF research behind ChatGPT.⁠[3][4][4]

He later wrote that the methods he helped build at OpenAI made language models good at following instructions and talking with people, and that this work became the research behind ChatGPT; at the time he thought chat models might lead to artificial general intelligence, but he came to believe that something important was missing.⁠[5][5]

In 2024, Almeida left OpenAI and founded TypeSafe AI with Sasha ShengSasha ShengSasha Sheng is a software engineer and entrepreneur, co-founder and chief operating officer of the artificial intelligence company TypeSafe AI. She worked on mobile apps and machine learning systems at Yahoo and Facebook (Meta), won a number of AI hackathons after leaving Meta, and in 2024 co-founded TypeSafe AI with Diogo Almeida and Erik Gafni.Open the full entry and Erik Gafni, becoming its chief executive; the company spent about two years developing in stealth. He has said that the question driving him over the past four years was: models have been superhuman at chat for years, so where is all the automation?⁠[1][1][6][5][5]

In 2026, about four months before TypeSafe launched its product, Almeida gave a talk titled “AI: Too Good to Be True, Too Bad to Be Useful” at the AI Council conference in San Francisco. He argued that nearly all production large language models are trained with RLHF or a variant of it, and that RLHF optimizes for human preference rather than for doing the task correctly, which makes models impressive at assistance work but poor at automation work that looks easier.⁠[3][3][3][3]

On September 15, 2026, TypeSafe AI came out of stealth, and Almeida announced its first System One model, Jev, in a post under his own name; the same day the company announced a $40 million seed round led by DCVCDCVCDCVC (Data Collective) is an American deep tech venture capital firm founded in 2010 by Matt Ocko and Zachary Bogue, with offices in Palo Alto and San Francisco, California. It invests in companies that use computing and AI to solve real-world problems in fields such as energy, climate, biomedicine, defense, space, and industry; as of October 2026 it had more than $4 billion in assets under management across 13 funds.Open the full entry. In a Q&A published by DCVC, he said that large language models work well because they have been optimized for human preference, which is also why they still need humans in the loop; he believes that in a world where artificial general intelligence has automated most of the economy, over 99% of AI calls will be made by and for software.⁠[5][5][7][7][7][7]

On September 28, 2026, Andreessen HorowitzAndreessen HorowitzAndreessen Horowitz (a16z) is a venture capital firm founded in 2009 by Marc Andreessen and Ben Horowitz. It invests from the seed stage through the growth stage and provides operational support to its portfolio companies.Open the full entry released a podcast conversation between Ben HorowitzBen HorowitzBen Horowitz is an American entrepreneur, venture capitalist and business author who co-founded Andreessen Horowitz (a16z) with Marc Andreessen in 2009. He previously led Loudcloud’s transition into Opsware and its 2007 sale to Hewlett-Packard. His writing and investing address startup management, organizational culture and the relationship between technology and cultural industries.Open the full entry, Martin CasadoMartin CasadoMartin Casado is a computer scientist, entrepreneur and venture capitalist who is a general partner at Andreessen Horowitz, where he leads infrastructure investing. A major proponent of software-defined networking, he co-founded Nicira and served as its chief technology officer, and after VMware acquired the company he ran its networking and security business; he also co-founded the nonprofit MuralNet.Open the full entry, and Almeida. Almeida argued that coding agents help people write software faster, but that the software they produce works much as it always has; his goal is technology that can reliably “do what I mean.”⁠[8][8][8]

On October 9, 2026, TypeSafe AI announced an $870 million Series A led by Andreessen Horowitz at a $7.5 billion valuation, with Martin Casado joining the board. a16z said it had long been a fan of Almeida and shared his view that “the true bitter lesson” lies in returning the primitives of intelligence to builders.⁠[9][9][10][10]

TypeSafe AI: Official website: https://typesafe.ai/ (opens in a new window).⁠[5]

X: Public account: https://x.com/CompleteSkeptic (opens in a new window)⁠[8]