Good morning, Two Minute Warriors.

The AI bubble is showing signs of puncture. Companies are pulling back on AI spend, Meta is walking back features, and AGI looks to be facing government delays. Another fun time in the AI hype cycle. Also, a prompt idea for a helpful summary.

⏱ The 10-Second Version

  • Companies are pulling back on AI spend after costs outran projections. The Economist piece is the clearest signal yet that ROI pressure is real, not just CFO noise.
  • Meta launched an Instagram feature that let anyone generate AI images using a public account’s content without consent, then killed it within days after widespread backlash.
  • A scenario-planning document called AI 2040: Plan A is making the rounds on Hacker News. The short version: it’s a recommendation, not a prediction, for how the US and China could delay superintelligence to 2040 and manage the transition.

The Big Thing

AI costs are now big enough to have their own line-item crisis

What happened

The Economist reports that companies across sectors are actively scrambling to curtail AI spend after costs scaled faster than anticipated ROI. Usage-based pricing that looked cheap in pilots is turning into material budget problems at production scale.
Read the full story at The Economist →

Why you care

If you’re in any room where AI tooling is being approved or renewed, the CFO now has an Economist article to cite. The question isn’t whether AI is useful; it’s whether your org can show the math. Prepare the number, not the narrative.

What to say in your next meeting

“The AI cost conversation has moved from IT to the CFO’s desk. Anyone signing off on new tools should have a rough ROI estimate ready, not just a use case.”

Speed Round

Meta launched an Instagram feature letting users generate AI images from any public account’s content without permission, and pulled it within days after a wave of creator backlash.
Read the full story at TechCrunch →

AI 2040: Plan A, the new scenario document from the team behind AI 2027, is getting serious traction on Hacker News. Worth the 15-minute read.
Read the scenario →  |  See the Hacker News thread →

OpenAI is hiring a dedicated PM to build ChatGPT experiences for families, caregivers, and older adults. The household market is apparently the next frontier.
Read the full story at TechCrunch →

Palantir’s Alex Karp is publicly voicing what many CEOs are privately saying about AI underdelivering on promises. The WSJ piece is a useful temperature check on boardroom sentiment.
Read the full story at the WSJ →

Prompt of the Day

The ROI One-Pager Builder

“I need to justify AI tooling spend to a skeptical CFO or budget committee. Here’s context on the tool and how we use it: [DESCRIBE TOOL + USE CASE]. Draft a one-page ROI summary with: (1) estimated time saved per week per user, (2) cost of the tool vs. cost of the time saved, (3) one risk or caveat we should acknowledge, (4) a recommended decision. Be direct, no marketing language.”

Why it works: It forces the analysis into the frame a finance person actually uses, which is the only frame that matters right now.

Time saved: 60 minutes of spreadsheet wrestling becomes a 3-minute review and edit.

Inside Claude/ChatGPT: Either can generate a solid brief from a tool’s pricing page and your notes.

Forward this to the one person on your team who still approves AI tools without running the numbers.

Don