The most interesting thing I've heard from operators this month has nothing to do with model benchmarks. CFOs and COOs getting their first real exposure to AI as a working tool told me almost the same thing: the value isn't that the model is clever. The value is that it grinds through forty pages of reports and thirty emails, then hands back a morning briefing they can act on. That feels less like chatbot novelty and more like a junior analyst finally showing up on time.
The Chatbot Became a Workforce. Now Someone Has to Manage It. — The main essay. OpenAI released GPT-5.5 ("Spud") on April 23. Frontier models are now good enough for most everyday office work. The value has moved to what Ethan Mollick calls the harness: context, permissions, memory, files, scheduled runs, tool access, and a place to put the output. Claude Code, Codex, OpenClaw, Claude Cowork, and ChatGPT workspace agents are all harnesses. Pick one task (daily email triage is the most universal), find the harness already approved in your stack, and run it for a week. Most breakdowns will not be model problems. They will be harness problems: missing permissions, wiring not done, OAuth never completed. That gap is the actual work of AI adoption today.
Also in this issue:
- Signals This Week — Microsoft offered voluntary buyouts to up to 7% of its US workforce. Meta is cutting roughly 8,000 jobs and will capture employee mouse, keystroke, and screen interactions to train models. Gartner: only 39% of tech leaders believe current AI efforts will improve financial performance. A new insider threat category emerged from the Vercel/Context.ai breach: an employee adopting AI to be more productive becomes the launchpad of the breach.
- The Wire — SpaceX takes an option to acquire Cursor for $60B. FT/Focaldata: 60% of top earners use AI daily versus 16% of lower earners. Law firms warn clients that AI-generated emails are pushing fees up, not down.
- What I'm Consuming — OffDeal: two bankers closed 8 deals worth $91M by consolidating 12+ workflows into a single Claude agent. McKinsey's AI Transformation Manifesto. Practical Claude Cowork setup walkthrough. Peter Steinberger's TED talk on building OpenClaw.
- Quanta Lab — Open weight models are closing the gap. Qwen 3.6 runs on consumer hardware. Kimi K2.6 added long-horizon agent swarms. DeepSeek V4 ships with a 1M-token context. The procurement question worth raising with your CIO this quarter: which workloads actually need a frontier vendor?
- After Hours — Frankenstein (2025), Guillermo del Toro. Low expectations going in, blew me away. The Creature's emotional arc locked me in. Five stars.