[{"claim":"Agent execution is the new infrastructure category: agents need complete stateful computing environments with persistent state, OS access, portability, isolation, policy enforcement, scheduling, and observability — not human-developer abstractions.","confidence":null,"evidence_type":"independent_confirmation","labels":["Prediction","Agent Infra","Watch"],"score":null,"signal_date":"2026-07-16","signal_id":"sig-20260716-thesis-a16z-runta","source_url":"NEEDS_URL","theme":["agent infrastructure as control layer","agents need computers not abstractions"],"verification_status":"needs_verification"},{"claim":"Investor thesis movement converges from broad AI exposure and early entry toward concentrated ownership of frontier winners, durable execution infrastructure (agents), and public-market-grade economics — monopoly/oligopoly proof over model access.","confidence":null,"evidence_type":"synthesis","labels":["Prediction","Convergence","Late Stage","Watch"],"score":null,"signal_date":"2026-07-20","signal_id":"sig-20260720-thesis-combined-movement","source_url":"","theme":["concentrated late-stage ownership","agent execution as control layer","monopoly/oligopoly proof"],"verification_status":"needs_verification"},{"claim":"Simultaneous capital deployment across transformer-specific inference (Etched), open chiplet architectures (TYLSemi), CPU agent execution (Runta), and scientific compute (CuspAI) converges at the market layer: the workload-routed heterogeneous stack means ownership of the intelligent scheduler exceeds ownership of any single accelerator.","confidence":null,"evidence_type":"independent_confirmation","labels":["Convergence","AI Infrastructure","AI FinOps","Emerging"],"score":null,"signal_date":"2026-07-21","signal_id":"sig-20260721-conv-heterogeneous-execution","source_url":"NEEDS_URL","theme":["heterogeneous execution stack","scheduler as control point","capital markets bet across compute architectures"],"verification_status":"needs_verification"},{"claim":"Five convergence events (agent governance, vertical industrial platforms, open-weight bargaining, deployment ownership, heterogeneous execution) restructure the AI market from model-centric to governance/execution-centric: the winning position owns governance rules, deployment workflows, scheduling intelligence, and industrial feedback loops — operational control points, not model access.","confidence":null,"evidence_type":"synthesis","labels":["Convergence","Synthesis","Watch"],"score":null,"signal_date":"2026-07-21","signal_id":"sig-20260721-conv-synthesis","source_url":"","theme":["AI market restructuring to governance/execution-centric","operational control points over model access","five convergences combine into single thesis"],"verification_status":"needs_verification"}]