Andrej Karpathy announced on 19 May 2026 that he has joined Anthropic, marking his first full-time research role since leaving OpenAI in February 2024. "I think the next few years at the frontier of LLMs will be especially formative," he wrote on X. "I am very excited to join the team here and get back to R&D." Karpathy joins under Nick Joseph on Anthropic's pre-training team, with a mandate to stand up a new effort that uses Claude itself to accelerate pre-training research — a model-assisted-model-training loop that has become a central capability bet across the frontier labs.
Eureka Labs — the AI-native education company Karpathy founded after OpenAI — continues, but he is stepping back from day-to-day operations. He stated he "remain[s] deeply passionate about education and plan[s] to resume my work on it in time." The move follows his 20 April 2026 Sequoia AI Ascent fireside, where he declared "vibe coding" obsolete and pushed his "agentic engineering" / Software 3.0 framing — noting his own ratio of human-written to agent-written code had flipped from 80/20 to 20/80 between November and December 2025. CNBC, TechCrunch, and Axios all framed the Anthropic move as a talent-war milestone in the Claude-vs-GPT race.
For Singapore enterprise tech, the signal is twofold. Anthropic opened its Singapore office in 2024 and Claude is already the default frontier model for many SG enterprises constrained by data-residency or government procurement — Karpathy joining Claude's pre-training core is a near-term indicator that Claude's reasoning and coding capability will keep closing on GPT. It also gives Singapore CTOs and engineering managers building "agentic engineering" teams a clear external playbook — Karpathy's Sequoia thesis — to point juniors and graduate hires toward.
