HTMX's CEO has mandated "No AI Fridays", one day a week with AI off, and a bet the industry can test

'No AI Fridays': a mandate from HTMX's self-described CEO that turns AI fatigue into a scoreable bet

No AI Fridays comes with arithmetic attached: one day a week with AI assistants off costs a fifth of the working week and, by the manifesto's own logic, exactly one fifth of whatever productivity gain AI actually delivers. The single-page site 1, whose author self-identifies as HTMX's CEO, reached Hacker News' front page this week 2, and it hands the industry a scoreable wager: if the productivity claims hold, giving back a fifth of the week is cheap insurance.

What the mandate actually asks for

Four moves, per the site 1, all aimed at judgment rather than typing speed:

• AI assistants stay off for the day
• code is written by hand
• the documentation gets read
• decisions get reviewed

Read as a bundle, they are not one ban but four separate audits: the off switch tests dependence, handwriting restores the felt cost of every decision, reading the docs keeps retrieval muscle warm, and review asks which choices drifted while nobody was watching.

The page speaks in the first person of a boss: "For now, as the CEO of HTMX, I've mandated No AI Fridays" 1. Companies that adopt the practice are invited to message @lazilyevaluated to join a public list; as of our fetch, that list holds only the author's own entry, which makes the list's growth, or its silence, the experiment's readout.

The arithmetic inside the bet: a fifth of the week, a fifth of the dividend

The wager, paraphrased: if AI's productivity gains are as large as claimed, one day a week spent minimizing AI's downsides should be an easy trade 1. The math is kinder to the bet than the culture-war framing suggests. Call the weekly gain g. If an assistant lifts a developer's output to (1 + g) times baseline, an all-assistant week yields 5(1 + g); swapping one day back to unassisted baseline costs exactly g, one fifth of the week's 5g dividend. The team gives up a fifth of its gain, not a fifth of its output, and as long as g is positive, a week with one day swapped back still beats an all-unassisted week. The site adds a compounding claim: a token-free day can lead to significant long-run token savings 1. If usage is evenly spread, one day in five is up to 20% fewer tokens, a number a finance team can audit on its own.

The research it leans on, and why no study is named here

The evidence paragraph makes four linked claims: that LLM use can cause you to accumulate "cognitive debt," cuts engagement, weakens critical thinking, and slows skill formation 1. Each phrase is a hyperlink, but the target studies did not survive our text-only fetch, so this piece names none of them; verification comes before citation. Notice what the four claims share: all of them describe the human, none describe shipped output. That split resizes the wager. The cost side, productivity lost on Fridays, shows up within a week and reads straight off a sprint board. The benefit side, skill formation and critical thinking, accrues on a timescale no Friday can register. On our reading, the bet is built to be falsifiable on cost and nearly unfalsifiable on benefit. The sharpest pushback on the thread skips the studies and disputes the premise. One developer says they would not call their thinking offloaded to AI: in their view AI is great for pushing text, but they haven't found its system design skills, where the thinking happens, to be that good 2. Another proposes a fix cheaper than abstinence: have the AI present decisions rather than make them 2.

A "mandate" whose builder pitched it to a boss

The artifact undercuts its own framing. The page speaks as a CEO issuing a mandate; the developer who submitted it to Hacker News, and who says they built the site, describes the motive in the opposite direction: "I made the site to convince my boss to let me not use AI on Fridays" 2. A mandate does not need sign-off from a boss. And in the thread's deeper replies the builder answers a skeptic himself: "Have you read the page? I'm the COE of HTMX. I'm the boss." 2 A commenter's edit lands the same reading: the original poster is pitching the practice, not practicing it 2. The FAQ's register runs to pub banter; it jokes about "a couple of shots of Claude" 1. The site carries no date stamp, and htmx.org's front page, as of our fetch, makes no mention of the policy 3. The honest reading: an earnest core argument wearing a mandate costume. That does not sink the bet; it relocates it. The test is not whether HTMX runs the policy, which nothing we fetched confirms, but whether any company on that list runs it long enough to report what happened.

The carve-outs draw the real line: feedback yes, delegation no

The FAQ's exceptions are the most careful part of the page. Feedback tools such as greptile and prelint are endorsed alongside handcrafted code, with a rationale that is pure register: "Your LLM is not learning shit from their feedback, but you can" 1. Read against the ban, the dividing line is not human versus machine. It is tools that surface feedback a person metabolizes versus tools that act on your behalf. The option to scale from one day up to seven AI-free days a week makes the same point: calibration, not abstinence 1.

What one no-AI day would cost and teach

For a team that builds with AI daily, the cost is bounded and computable: a fifth of the working week, a fifth of the claimed dividend, and up to a fifth of token spend on evenly spread usage. What it teaches is the audit the site names: a weekly retrospective on which choices the assistant made and whether the direction it steered still matches the team's preferences 1. The sharpest objection on the thread is not about cognition at all: a weekly AI-free day is a 20% time grant, and only privileged workers will be allowed one 2. Note the pricing: the skeptic calls one fifth of the week the cost, and the manifesto calls it the premium. That disagreement is the whole bet, and the scoreboard is public.

No AI Fridays, in seven sourced lines

1. The setup, per the undated manifesto: assistants off for one day a week, code by hand, documentation read, decisions reviewed 1.
2. The site self-identifies its author as HTMX's CEO and says that author has mandated the practice; no independent source confirms either 13.
3. The bet, in the site's words: "spending one day a week to minimize its downsides shouldn't be a difficult trade-off" 1.
4. On the Hacker News thread, fetched for this story, the site's builder says it was made to persuade their own boss to allow AI-free Fridays 2.
5. The FAQ permits up to seven AI-free days a week and jokes about "a couple of shots of Claude" 1.
6. As of our fetch, the public participant list names only the author's own company 1.
7. The site carries no publication date; when the mandate began is unconfirmed 1.