Subject
Morning brief — Tuesday, April 28
Eight things from this morning's read:
• Anthropic shipped a new agent SDK with first-class scheduled execution. Looks aimed squarely at the prosumer category. (TechCrunch)
• A YC-backed startup called Riff raised $12M to build agents that drive logged-in browsers. Same playbook as browser-use, more consumer-facing.
• OpenAI quietly rolled out a $5/month tier for ChatGPT Free users — explicitly framed as a price floor for agents.
• The Information confirmed Lindy is at $20M ARR, with the bulk from individual professionals on the $49 plan.
• A long Hacker News thread on "why my agent broke at 3 AM" — top comment is a strong argument for between-step pause as a default.
• A new benchmark called AgentBench-2 dropped. Sonnet 4.6 leads, Haiku 4.5 punches above its weight on cost-adjusted scores.
• Notion shipped its own agent framework. Tightly coupled to their data model — won't matter for general-purpose work, but could lock in some prosumer use cases.
• A thoughtful Stratechery piece on why prosumer AI tooling is a winner-take-most market.