2026: This is AGI

• AGI has arrived — defined functionally as “the ability to figure things out.”
  • Three ingredients: baseline knowledge (2022), reasoning (o1, 2024), iteration (agents, 2026).
  • Claude Code and coding agents close the loop on long-horizon autonomy.
• Agents now run reliably for ~30 minutes without human guidance.
  • METR data: agent capability doubles every 7 months.
  • Day-long tasks by 2028; year-long by 2034; century-long by 2037.
• Specialist agents are live across law (Harvey), medicine (OpenEvidence), security (XBOW), recruiting (Juicebox).
  • Also: Traversal (SRE), Day AI (GTM), Harmonic (math), Ricursive (chip design).
• Two scaling approaches: RL for focus; harness engineering for memory/context.
  • Manus, Claude Code, Factory Droids cited as strong harness examples.
• Shift from “talkers” (2023–24 chatbots) to “doers” (2026–27 agents).
• Builders must reprice from hours to outcomes and redesign UX for delegation.

Pat Grady and Sonya Huang (Sequoia Capital) · 2026-01-14 · Read on sequoiacap.com