LGTM was always a small lie. Now it is a fast one.
Four letters approve code nobody read. The machines only deleted the pause before typing them.

Before new code reaches customers, a colleague is supposed to read it and approve it. The approval is often four letters: LGTM, short for looks good to me.
In code review, a colleague reads your proposed code and says whether it can go in, and these four letters are the customary way to say yes. The Go project's own code-review guide uses LGTM exactly that way, as the reviewer's sign-off. The letters were established enough that GitHub held a US trademark on LGTM for software services, until it lapsed in 2024. They are also an old joke, because LGTM gets typed on code nobody really read.
A music video, "LGTM (Looks Good to Me)" from the channel Degenerative Pixels, went up on 5 October. Lately, says the description, engineers type those letters on code they "never had time to read". Credits: animation by Claude Opus 5.5, song made with Suno, both AI tools. 1.4 million views by 23 October.
One reply on Hacker News was just "LGTM!". The audience reviewed the song the way code gets reviewed.
LGTM was always a polite fiction between colleagues. Unread code is how mistakes reach users. Four letters promised no reading, only trust in a colleague, like the box claiming the terms were read.
Phone a garage, imitate the noise the car makes, and a mechanic says it sounds fine. Nobody has looked at the car, and both parties are content.
The song's narrator approves 40,000 lines in about three seconds, and the code goes live on Friday at 5. Four letters once sat beside some reading. Now the letters are the whole review. The description blames leadership, which "heard AI makes everyone ten times faster". Faster at producing, not at reading.
SmartBear, a maker of code-review tools, found that past 500 lines an hour, reviewers catch markedly fewer mistakes. A sum: 40,000 divided by 500 is 80 hours.

The ceiling on human attention has not moved. The time underneath it has. Machines did not invent the lie. Machines only removed the time it took to tell it. Only the pause is gone.
A polite fiction needs two people. LGTM was never a claim about the code. It was a claim about the author: that someone competent had thought about this, cared whether it worked, and would still be there on Monday to fix it. The reviewer was not auditing lines. The reviewer was countersigning a colleague. Cheap, and mostly fine, because the cost of being wrong landed on somebody who would feel it.
Every job has its small courtesies. The nod in the hallway. The "no worries" to an apology you did not quite hear. LGTM belongs to that family. It is not a lie, exactly. It is the sound a workplace makes when it is running smoothly, and like all such sounds, it keeps running long after the smoothness is gone. Nobody decides to stop reading. The phrase just gets easier to type than the alternative, and the alternative keeps getting longer.
The design from this story
We made a T-shirt out of this story and called it "LGTM": one huge LGTM across the front, drawn entirely in tiny lines of code, legible from across a room while nobody ever reads the lines up close. Faded Coal, Faded Bone, Faded Khaki, 499 kr. Approved at a glance, which is the only way anything gets approved.
Next story by email, in stdout
The field guides go out in the stdout newsletter. One email every two weeks. If there is nothing worth your time, we skip that week and send nothing.
Where this comes from
- primaryGo project wiki, CodeReview.mdLGTM - looks good to me: the reviewer approves of your changes as they currently are.
- primaryDegenerative Pixels, LGTM (Looks Good to Me), YouTube, 5 October 2026Published 5 Oct 2026 17:38 Pacific; description (never had time to read; leadership heard AI makes everyone ten times faster); credits (Suno v6, animation Claude Opus 5.5); 1,444,558 views on 23 Oct 2026. Song beats checked via a third-party transcript, not by watching.
- primaryHacker News thread on the video, 8 October 2026Comment by physPop: "LGTM!".
- primarySmartBear, Best practices for peer code reviewA significant drop in defect density at rates faster than 500 LOC per hour.