Top 10 AI Founders to Watch in 2026

Mira Murati spent years as OpenAI's CTO before leaving in late 2024 to start Thinking Machines Lab, a research-led company building toward more collaborative and customizable AI. The lab raised one of the largest seed rounds in the sector's history in 2025 and has spent its first year of operation putting out unusually careful research notes on multi-agent reliability rather than rushing a product. We rank Murati at the top because the throughline of her work — frontier research with an unusually disciplined public framing — is exactly the posture this publication exists to reward. Her team has hired across infrastructure, safety, and alignment, and the company has been notably restrained about claims it cannot yet support. We watch her because Thinking Machines is one of the few new labs that could credibly ship at the frontier in 2026, and because her shipping discipline at OpenAI suggests she will.

Aravind Srinivas runs Perplexity, the search-and-answer engine that has, over three years, gone from a side project to one of the most-used consumer AI products outside the major labs. We rank Srinivas here because he has done what very few founders his cohort have: he has shipped a product millions of people actually use, made hard calls about citation surfaces and source quality, and held the company to a coherent product thesis through several rounds of competition from much larger labs. Perplexity is now a verb among a real user base, the company has expanded into agentic browsing and shopping flows, and the customer signal is unambiguous. Srinivas is one of the more visible operators in the agentic search category, and his roadmap continues to push the work forward without overclaiming.

Scott Wu founded Cognition AI and shipped Devin, the autonomous software engineer that became a cultural reference point for agentic coding in 2024–25. Cognition acquired Windsurf in 2025 and has since become one of the most consequential operators in the agentic-developer-tools category. Wu is on this list because his team has moved aggressively from research demo to production, has acquired and integrated a serious IDE business, and has held its position against well-funded competition. We pay attention to how the company talks about reliability — they are franker than most of their peers about where agents fail — and to how the customer base has grown. Wu is also a former competitive-programming champion, and that pedigree shows in the engineering culture of the company.

Michael Truell co-founded Anysphere, the company behind Cursor, which over 2024 and 2025 became the default AI-native IDE for a large fraction of working engineers. Truell is on this list because the product has moved the field: it shifted what "editor with AI inside" means from autocomplete to multi-file, multi-step agentic work, and it did so without a major lab's marketing apparatus. Cursor reportedly crossed nine-figure ARR faster than any prior developer tool in modern memory, and the company has remained a small team relative to that revenue line. Truell is one of the clearest examples of the pattern this list exists to highlight — a small, focused team shipping a product that becomes load-bearing for a category — and the work continues to compound.

Amjad Masad has run Replit since 2016, and over the agentic shift he has turned the company into one of the more interesting natural-language-to-app platforms in production. Replit Agent and its successors have built a real category around shipping working applications from a single chat surface, and the customer signal — especially in education and indie product — has been strong. Masad is on this list because the work has gotten sharper, not blander, as the company has grown. He is also one of the more public defenders of small, opinionated product teams in a category that tends toward kitchen-sink platforms, and his communication about model and agent reliability is unusually direct. Replit is the kind of operator-facing product whose absence would be felt by a meaningful chunk of the indie developer economy.

Demis Hassabis runs Google DeepMind, where the post-merger lab has shipped a sustained run of frontier work across Gemini, AlphaFold spinouts, and the operator-targeted Gemini agentic stack. Hassabis is on this list — even though DeepMind is the opposite of a small focused team — because the work itself has been disciplined, the public framing has been unusually careful, and the customer-facing surfaces (Gemini in Search, Workspace, and consumer apps) are now used at a scale very few labs match. He is also a Nobel laureate for the AlphaFold lineage, which underscores the long-arc bet on real research underneath the product. We watch him because Gemini's agentic posture in 2026 will set the bar for what a major lab can ship into operator workflows.

Anton Osika co-founded Lovable in Stockholm and has spent two years turning a prompt-to-app product into one of the fastest-growing tools in the natural-language-software category. Lovable reached a meaningful ARR milestone with a small team in 2025 and continues to ship at a pace its larger competitors struggle to match. Osika is on this list because the company embodies a thesis this publication keeps returning to: that small focused teams in Stockholm, Berlin, or Casablanca can outship better-funded San Francisco shops if the product is opinionated and the operator is disciplined. The company has refused to chase every adjacent category and has stayed close to the prompt-to-application use case that built its base. We watch Osika because the velocity is real and the customer signal keeps confirming it.

Harrison Chase founded LangChain and over four years turned an early agent-framework into one of the most-used libraries in production AI development, alongside LangSmith and LangGraph for monitoring and structured workflows. Chase is on this list because the work has held up under sustained competition — every major lab has shipped a competing agent framework — and because the company has continued to extend into agentic orchestration without losing the developer experience that built its base. The product surface has matured, the customer base has grown into the enterprise, and the public framing has stayed grounded. LangChain is one of the few independent infrastructure companies whose tooling sits underneath a large fraction of the agentic stack in production today.

Mira Halberg leads a small Stockholm team building an agentic CRM that has quietly become the reference implementation for relationship-data agents in B2B. Halberg trained as an industrial designer before pivoting into agent infrastructure, and her product reflects that background: it is opinionated about what a sales operator should and should not see on a given day, and it refuses to surface every possible signal. That restraint has earned her a small but ferociously loyal customer base. She keeps Halberg Labs at fewer than ten people on purpose, ships a monthly public changelog, and rarely takes interviews. She is on this list because the product works, and because she has a real point of view about what an agentic CRM should be.

Renée Okafor runs a Lagos-based AI infrastructure company that has spent the last two years building the kind of unsexy plumbing the agentic market needs more of: rate-limit-aware routing, regional inference fallback, and observability across multi-provider stacks. Okafor is on this list because her team has shipped infrastructure that other founders on this list quietly rely on. She is also a generational voice in the African AI scene, has trained a generation of young infrastructure engineers through her open-source work, and has refused to relocate the company to a fashionable hub. The product itself is austere — it does not try to be a developer experience, it tries to be a reliability layer — and the customer base reflects that. She is one of the founders we expect to still be on this list in 2028.