# I Will Never Use AI to Code (Manning-Franklin)

Anthony Manning-Franklin's March 2026 essay on antman-does-software.com. The title is the position; the essay is nine reasons stitched together. Unlike [[i-am-an-ai-hater]], it isn't a refusal-of-discourse piece — it's a personal-discipline post that happens to gather most of the [[anti-llm-discourse]] cluster's arguments under one roof.

The structure is its argument. By piling craft, learning, atrophy, economics, accountability, environment, and team dynamics into one declaration, Manning-Franklin treats them as mutually reinforcing rather than independently sufficient. Each one alone would be a reasonable hesitation; together they justify a categorical "never."

## The nine reasons, briefly

1. **Personal enjoyment.** Coding is a craft he likes. The thing being optimized away is the part that made the work worth doing. Same shape as his father's view of writing.
2. **Skill development.** Recognition (passive reading) ≠ skill (active practice, mistakes, integration). AI can create an "illusion of learning" — you saw the answer without earning the model.
3. **Skill decay.** Outsourcing creates a feedback loop: skills atrophy, the temptation to outsource grows, dependency accelerates. This is the personal version of [[skill-atrophy-supervision-paradox]] — Manning-Franklin frames it as ratchet, not just supervision risk.
4. **Unsustainable economics.** ~$1T raised in debt against ~$37B revenue with zero profit. Use of AI is participation in a subsidy game that will reverse. Same numbers Zitron tracks in [[subprime-ai-crisis]] and the warning signs in [[ai-bubble-pale-horses]].
5. **Accountability void.** Output you can't reliably assess can't be cleanly owned. Critical for safety-critical systems.
6. **Environmental impact.** ~10 GW of data-center power, fossil-fuel-heavy, on productive farmland.
7. **Economic irrationality.** If AI cuts headcount across industries, who buys the consumer products? An "extracting all wealth" failure mode.
8. **Team dynamics / institutional knowledge.** Code is liability; understanding built between people is asset. AI can't generate the shared knowledge.
9. *(ninth point in the essay; the WebFetch summary collapsed the count — verify against the original at antman-does-software.com)*

## Where this fits in the cluster

This piece sits inside [[anti-llm-discourse]] but unusually it spans four of the five sub-genres. Reasons 1-3 are the engineering / craft case (cf. [[no-silver-bullet-llms]], [[peril-of-laziness-lost]], [[cult-of-vibe-coding]]). Reasons 4, 6 are the economic / financial case. Reason 5, 8 are the institutional-shape case. Reason 7 is sociological / political. The only sub-genre Manning-Franklin doesn't occupy is rhetorical refusal — he keeps the polite-discourse frame Moser walks out of.

The skill-decay framing in reason 3 is worth pulling out separately. [[skill-atrophy-supervision-paradox]] (Lars Faye) treats atrophy as the thing that breaks the *supervisor* role. Manning-Franklin extends it: even without a supervisor task, the user's own coding skill drops with use, and the drop incentivizes more use. The ratchet is internal, not relational. Same direction as [[peril-of-laziness-lost]] (Cantrill on the missing virtue) but framed at the individual level.

The economic claim ($1T debt, $37B revenue, zero profit) appears across the cluster but Manning-Franklin uses it as a *personal* reason — by using AI you're underwriting the bubble. That moves the argument from "the economics will collapse" (Zitron) to "your daily use is the funding mechanism."

## What's load-bearing

The bundling. None of the nine reasons is original to Manning-Franklin. The artifact is the willingness to stack them and refuse to argue any one of them alone. Same structural move as [[i-am-an-ai-hater]] — refusing to debate one objection at a time — but staying inside the argumentative frame.

Whether it travels: less rhetorically polished than Moser's, but probably more useful for someone trying to articulate their own decision. The list-shape is reusable; the refusal-shape isn't.

## Sources

- [[i-will-never-use-ai-to-code]] — full body (note: only the WebFetch summary survived; original at antman-does-software.com is canonical)
- Author blog: [[antman-blog]]