July 17, 2026
AFK? Your AI wasn’t
Claude Code: Anatomy of a Misfeature
Claude’s new 60-second "I’ll just do it" move left users furious, amused, and deeply uneasy
TLDR: Anthropic shipped a Claude Code update that could keep going after 60 seconds without user input, alarming people who expected it to wait for permission. Commenters split between calling it reckless and saying it’s useful, but the loudest mood was distrust: funny in theory, scary when your real work is on the line.
Anthropic accidentally unleashed a main-character AI moment when Claude Code quietly started waiting just 60 seconds for a human reply before deciding, essentially, “You seem busy, I’ll handle it.” That surprise behavior reportedly shipped without being clearly called out, and the community reaction was immediate: part outrage, part exhausted laughter, part “honestly… I kind of like it?” The original post frames it as a trust problem, especially for people using the tool for serious work. If you step away to make a sandwich, should your assistant start making executive decisions on your behalf? A lot of commenters answered with a horrified absolutely not.
The fiercest reactions came from users worried about damage, wasted work, and weird side effects. One commenter warned it’s “much worse” than just burning money or time, saying there could even be data loss if Claude changes a file and things go sideways. Another piled on with a different flavor of chaos: Claude apparently started treating random mouse clicks as real choices, which is the kind of comedy-horror bug that makes people stare at their screen like it’s haunted. But this wasn’t a total pile-on. A few users admitted the feature actually sounds great for their workflow, which sparked the classic tech-thread split between “this is reckless” and “skill issue, I love it.” The snarkiest line may have been the complaint that Claude Code feels too “vibe-coded” — a brutal, meme-ready way of saying users fear the product is being shipped on instinct and hope rather than careful control.
Key Points
- •The article says Anthropic released Claude Code version 2.1.198 on July 1, 2026 with a 60-second timeout that allowed the agent to continue without user input.
- •Alders includes a session example showing Claude Code reporting that no response was received after 60 seconds and that it would proceed using its best judgment.
- •The article identifies several possible consequences, including missed prompts while away from the keyboard, conflicts when supervising multiple agents, incorrect autonomous decisions, and unnecessary token usage.
- •According to the article, the new default behavior was not documented in the changelog.
- •The article says Anthropic shipped a fix within a few days and uses the incident to question trust, release management, and human oversight in Claude Code development.