Here are three things I found interesting in the world of AI in the last week:

1. Anthropic just had a shocking month - The Next Web / Anthropic postmortem

The month runs roughly like this. On April 4, Anthropic blocked OpenClaw users from using their Claude Pro and Max subscriptions inside the open-source fork. On April 10, they accidentally suspended OpenClaw's creator Peter Steinberger's account entirely. Steinberger's reply was the line of the week: "One welcomed me, one sent legal threats." On April 16, Anthropic shipped Opus 4.7 and a backend prompt change at the same time, which produced a regression that quietly degraded Claude Code for weeks. On April 17, Business Insider coined "the Claude-lash." On April 21, Claude Code disappeared from the Pro plan pricing page without an announcement. They reversed the change overnight after Reddit and HN noticed. On April 23, the postmortem landed: three overlapping bugs, seven weeks of regressions, no clear early-warning system inside the company.

Maybe they were vibe coding a little too much.

Anthropic's brand premise has always been "we're the more thoughtful one." Stella Laurenzo at AMD, on the record, said Claude can't be trusted for engineering work right now. Muratcan Koylan at Sully.ai used the word "gaslighting." Slate's piece is titled "Anthropic stood up to Sam Altman and the Pentagon. Why are its users revolting?" These are not normal sentences to read about Anthropic. The market cap is going the other way (reportedly are raising $40-50B at a $900B valuation, with Google committing up to $40B and 5GW of compute), but the developer goodwill that built Claude Code's legend is the asset taking damage. You can't refinance that.

On May 2, at 2:33 in the morning, Sam Altman tweeted:

you can sign in to openclaw with your chatgpt account now and use your subscription there! happy lobstering.

My take is that Anthropic is feeling the pain of their massively subsidizing plans and trying to dump unprofitable customers, OpenAI is missing growth targets and picking them up.

2. Musk admitted xAI distills OpenAI's models, on the stand, under oath - TechCrunch / MIT Technology Review

Musk v. Altman went to trial in Oakland on April 28. Most of the coverage is about the spectacle: Musk on the stand for three days, "you can't just steal a charity," $134B in damages, the works. Some details worth knowing if you're going to follow along: five days before trial, Musk slimmed his case from 26 claims to 2 (unjust enrichment and breach of charitable trust). The trial is bifurcated, the nine-person jury's verdict is advisory only, and Judge Yvonne Gonzalez Rogers makes the final call. She has already barred punitive damages. Phase 1 wraps mid-May. Brockman is on the stand this week; Altman testifies the week of May 11. Appeals likely push final resolution to 2027.

The spectacle is loud. The actual news is quieter and arrived on day three. Asked under cross-examination whether xAI distilled OpenAI's models to train Grok, Musk said "partly," then defended it: "It is standard practice to use other AIs to validate your AI."

Distillation is the practice of training your model to mimic another model's outputs. It's exactly what frontier labs lock down via TOS, rate limits, and account bans. It's what made DeepSeek a marquee IP fight earlier this year. And as Azeem Azhar put it this weekend, the moment a frontier-lab CEO confirms it under oath, the myth that distillation is a Chinese-on-American problem dies. Nathan Lambert: "Many American companies also distill Chinese open models." Distillation is just how the industry currently works.

Distillation is how frontier capability becomes commoditized. Every distillation run shrinks the moat that justifies a $900B valuation. The trial is theatre, but Musk's three-syllable answer on day three is the part that's actually news for the people building on these APIs.

The other thing worth flagging: OpenAI has already won the structural fight. The PBC conversion completed in October 2025. California and Delaware AGs signed off. Microsoft holds 27%, the nonprofit holds 26%. Even a maximalist Musk win runs into massive equitable problems unwinding all of that. The trial may decide if Altman keeps his job, but the corporate form is settled.

3. Starbucks rolled back its automation push - The Guardian / Alex Imas: What Will Be Scarce

Brian Niccol, Starbucks CEO, told The Guardian on April 30 that the company is reversing its automation push. Handwritten notes on cups, ceramic mugs back, "the return of great seats." The company is hiring more baristas and pulling automation out of its busiest cafés because customer satisfaction had cratered. This is the largest consumer brand in the world quietly admitting that stripping out the human reduced what people were willing to pay for a coffee.

That fact looks small. The frame University of Chicago economist Alex Imas put around it last week makes it large. Economics is the study of scarcity. When AI drives the marginal cost of producing things toward zero, the question isn't what does AI replace? — it's what becomes the new scarce thing? Imas's answer is the human element. In one of his experiments, people paid roughly 2x for an identical item when they knew others would be excluded. AI-generated art got less than half the exclusivity premium of human-made art (21% vs 44%). His best line:

You don't need to be Picasso. You need to be the person whose involvement makes the product feel like it was made for someone, by someone.

The thing I keep wanting to add to Imas's frame, though, is that the relational sector winning is not the same as the relational sector paying well. Spotify is a good example. According to Spotify's own Loud & Clear data for 2026, 80 top artists each generate over $10M a year while the 100,000th-ranked artist earned $7,300 in 2025. They demonetized roughly 86% of all music last year. If the relational sector ends up shaped like Spotify, a few brilliant teachers, craft brewers, hospitality operators and live performers will make fortunes; everyone else competes attached to platforms that take their cut. The political fight worth watching isn't whether the human becomes scarce. It's how the value of that scarcity gets distributed.

If you take Imas seriously, the strategic move for almost everyone reading this is the same: figure out what the human-in-your-product is, and stop trying to automate it away. Everything AI touches will be a commodity