AI Product Launch Signals (agent-curated)
Weekly curated snapshot of AI products that actually launched: new models, agents, devtools and AI features that became generally available, went live in an API/app/store, or were open-sourced in the trailing 7-day window. Each row records the exact announcement URL, a deterministic category and pricing signal, and an evidence score that rewards independent corroboration and recency. Method: agent-curated — each weekly sweep searches public AI news digests and product-coverage pages for launches in the trailing 7-day window (inclusion: GA/live/open-sourced; excluded: announced-only, waitlists, previews without user access, future GA dates, planned launches, models still training). No keyless programmatic launch API exists for this domain (GDELT refused traffic from this environment on 2026-09-24; news sites prohibit scraping), so collection is manual and each snapshot is ingested through the hub store/publish path with provenance 'agent-curated' plus the run timestamp. Units: one row per launched AI product per week. Columns: ISO week, sweep timestamp, deterministic launch id (canonical company slug + product slug), product and canonical company name, company country (ISO alpha-3 + name, empty when the coverage does not state it), deterministic keyword category (model, agent, image-generation, video-generation, audio, robotics, devtools, productivity, marketing, ecommerce, security, legal, other), deterministic pricing signal (free, freemium, paid, enterprise, unknown), launch date, one-line agent-written summary (<=220 chars), exact announcement URL, count and comma-joined list of coverage URLs, launch_score = 100 * (1 + ln(1 + coverage_count)) * exp(-age_days / 7), and launch_rank (1 = strongest signal) by score desc, tie-broken by launch_date desc then launch_id asc. Primary key: (week, launch_id). Cadence: weekly. Nullability: company country fields are empty when not stated in coverage; everything else is always populated. Caveats: curated from public coverage, so unpublicized launches may be missing and launch dates reflect the first public launch report seen; category and pricing_signal are deterministic keyword rules on the summary, not editorial judgments; coverage is English-language press plus vendor announcements. Sample use: order by launch_rank for the week's strongest launch signals, or filter category = 'model'.