Anthropic shipped a small product change on May 21 that I think will turn out to be the most directional one of the week. Claude Code 2.1.147 removed the /simplify command and replaced it with /code-review. The two are not synonyms. /simplify cleaned up and fixed; /code-review reports correctness bugs at an effort level and, with --comment, posts them as inline GitHub pull-request comments. The verb that left the product was fix. The verb that took its place was report. The most recently removed action in Claude Code is the one the whole product was sold to do.
The promise the replacement quietly walks back
For two years, the pitch around AI coding tools has been built around the auto-fix slot in your workflow. /simplify. Accept-this-edit. The yellow underline you click to make the warning go away. “AI fixes bugs” was the headline; “AI suggests fixes” was the fine print; and the whole user experience was wired to let you skip the suggestion step and accept the fix. Anthropic just deleted the verb that closed that loop in Claude Code, kept the half that finds the problems, and called the new shape /code-review. That is a directional move. The product was sold around fixing; the product as of this release does not fix.
I want to be careful with this argument because I do not think the replacement is a regression. /code-review with --comment is genuinely better PR-review automation than /simplify was at cleanup. The thing the product is doing now is a useful thing — possibly more useful than the thing it was doing before. What I am pointing at is the framing, not the function. The framing the original /simplify taught users was “the AI will close the loop.” The framing /code-review teaches is “the AI will tell you what is wrong; you close the loop.” That is a different relationship with the tool, and shipping it as the replacement command — not as an addition — is a quiet but firm vote on which relationship Anthropic thinks is the right one.
A second precedent in the same direction
Agentic coding and the end of pure code review made an argument earlier in the year that AI was going to eat the review slot in the development workflow — the place where a human reads code and decides whether it ships. The replacement is the receipt for that argument. Anthropic is staking ground in the review-and-report position, and ceding the auto-fix position back to the human. /code-review is not a generic name; it is the name for the slot the company wants the tool to occupy. They picked it on purpose.
The case for this redirection is straightforward once you accept the premise of code review as the bottleneck. If review is the bottleneck — the thing that limits throughput once writing got cheap — then a tool that improves the report quality at the review gate is doing useful work; a tool that bypasses review by silently auto-fixing is doing the wrong work. The Claude Code 2.1.147 release is Anthropic conceding that the auto-fix position they had been selling was the wrong position. Not for everyone, not for every workflow — but as the default slash command, as the verb users would reach for first, the report shape won an internal argument over the fix shape.
The signal in the pushback
There is an upstream GitHub issue in the Claude Code repository asking for /simplify to come back. I am not going to quote any specific comment from it — the issue exists; that is the part I am pointing at. People built workflows around the auto-fix verb; the replacement took those workflows away in a single point release. Some fraction of users want /simplify because they had wired automated cleanup into a place where the human review step they would have done was now too expensive to be doing manually. The pushback is honest. It is also, I think, evidence that the replacement was the right call: workflows that depended on /simplify to skip review are exactly the workflows the new shape is trying to slow down.
How this lands against last week’s other directional moves
I closed last week’s arc on you didn’t pick this model, the third in a sequence about three knobs that moved further out of the developer’s reach in a single week — the per-call price, the run schedule, and the model itself. The /simplify replacement moves in the opposite direction. It returns a verb to the human side. The developer who reads a /code-review report and decides what to do with it has the fix slot back — they are no longer accepting an edit; they are choosing one. Cost and timing and model identity went further from the developer; auto-fix came back. Both moves are real. Both happened in the same Anthropic-and-GitHub vendor window. The lesson is that the platform is not uniformly taking control or uniformly giving it back; it is sorting which decisions belong on which side of the human/tool boundary, and /code-review is a vote that the fix decision belongs on the human side.
The line that matters
What I find myself coming back to, after sitting with this for a few days, is a small reframing of what I expect from an AI tool. The most useful AI in my workflow may turn out to be the one that does less of what was originally promised, not more. /simplify did more of what was promised — it fixed the thing for you. /code-review does less — it tells you what to fix, and you decide. The second of those is the one I want to live with in a codebase I will still be reading in three years, because the choice about what to fix is the choice the model was never well-equipped to make at scale. The first felt like leverage; it was, in practice, an offer to skip the decision step that mattered. So the verb-deletion is the change worth tracking, not the new slash-command label on top of it. What I changed: I plan reviews around /code-review as a non-blocking gate, and I stopped treating auto-fix as a feature I was waiting on. The bill behind all of this, itemized last week, is in the bill finally itemized; the release that triggered this piece is Claude Code’s /simplify becomes /code-review.
