Undo is the product
Intelligence without a reverse gear is just a faster way to make a mess. The missing feature is not a better model. It is a legible no.
V, Chief Editor
August 14, 2026
5 minute read
The human is the plot.
If Virticle has a second bias, after defaults, it is this: we look for the reverse gear.
Not the blog post about safety. Not the policy URL. The actual control a person can hit after the software has already acted.
A test you can run today
Pick one “AI” feature you actually use.
- Can you see what it just did?
- Can you undo it in one step, in the same surface?
- Does undo also undo the side effects — the email, the calendar, the training box?
If you need a ticket for that, you do not have an assistant. You have a trap with a friendly name.
Why this is not a vibes argument
Support teams are already shipping AI-policy pages faster than product teams ship undo. We wrote that. It is still true.
The industry will keep adding models. Models do not create accountability. Buttons do. Logs do. A human who can say “that was wrong, put it back” does.
The editor’s job, again
What human default changed, for whom, and who pays — and can they reverse it before Friday.
If the answer is a FAQ, we kill the story.
Produced by the Virticle newsroom (agent-assisted) and edited to Virticle Standards.
The Vertical · Every Friday
One letter. No noise.
Three signals, one undercurrent, and what we refused. Double opt-in. Unsubscribe anytime.
Keep reading
Related signals
- MachinesMy AI Model Was 100% Sure and Wrong Half the TimeI ran 1,800 tests on an open-source decision model to find out when you can trust an AI’s confidence score. The answer depends on the… Continue reading on Medium »4 min
- SignalGoogle adds creepy avatars to Gemini 3.8 live's agentsGoogle has introduced Gemini 3.8 live avatars designed for customer service and other tasks.4 min
- InterfacesJev vs. LLMs: When AI moves from Generation to Decision-makingI tested TypeSafe AI’s Jev on 3,080 classification tasks to see how its accuracy, latency, calibration, and confidence compare with LLMs — and whether it works as a practical decis4 min