I'm done using AI

Brett Codes · article · 2026-08-10 Reader 🖥️

Agentic Engineering Individual AI Abstention AI Backlash AI Environmental Footprint Capacity to Verify AI Output Developer Experience & Flow Emotional Impact of AI Work Professional Identity

After using LLM-powered coding harnesses in earnest for over a year, the author concludes that the personal (mental-health, craft, deskilling), societal, and environmental harms of AI outweigh any productivity benefit it offers him, and that no moderate or selective level of use is viable — so he is stopping entirely.

practical · mixed — A first-person memoir/narrative structure in service of a practical conclusion (you — or at least I — should stop using AI). The autobiographical chronology and anecdotes (the ED visit, the hamster-wheel realization, the game-programming return) do the argumentative work that explicit premises only partly carry; the piece reads as testimony more than as a formal case.

Structure

Part	Role
Autobiographical framing: 20 years programming, a hobby turned career, genuine love of the craft.	context
Chronological history of encountering AI coding tools — Copilot autocomplete, early hallucinating chatbots, a colleague's LLM-driven true-believer hire, Cursor, Zed, agentic harnesses (Claude Code) — culminating in building a full project without editing a line of code.	evidence establishing credibility ("I understand what these tools are capable of") before the critique
Personal-harm narratives: the AI-chat/medical ED anecdote, becoming 'a hamster on a wheel' reviewing rather than writing code, depression and apathy.	core evidence for the personal-harm premise
Reframing of software quality as a function of care ('how cared for it is'), not feature count or speed.	normative premise underlying the craft-loss claim
Existential-dread section: job security, meaning, whether the profession will still exist.	evidence for psychological harm and professional-identity threat
Brief environmental/societal cost section (data centers, water usage, stolen training sources).	evidence for the societal/environmental-harm premise, asserted rather than sourced
Account of trying to use AI only for trivial tasks, and abandoning that middle path.	rebuttal to the implied counter-argument that moderate use is a viable alternative
Return to hand-coding hobby projects (Playdate book, Usagi engine) during the low point.	evidence / natural-experiment comparison supporting the craft-restoration claim
Closing declaration and admission of uncertainty about professional consequences.	conclusion / call to action

Key terms

agentic programming Agentic Engineering
The current generation of LLM-powered coding harnesses (e.g., Claude Code) that can be given a task and autonomously write, edit, and build software with minimal human editing — distinct from the earlier autocomplete-based (Copilot) or inline-chat (Cursor) workflows.
cared for
A quality attribute of software determined by the degree of human attention, judgment, and craft invested in it — not by feature count, build speed, or test coverage.
net harm
AI's aggregate negative impact (personal, societal, environmental) exceeds its aggregate benefit, for the author specifically and, in his stated opinion, for society generally.
delusional
A state of impaired judgment produced by treating the AI chat as a trusted authority/friend for medical, career, and business decisions — leading to costly, unnecessary real-world actions (the midnight ED visit).

Propositions

explicitcorroboratedEarly AI coding tools were unreliable from the start: Copilot's autocomplete was 'distracting and basically useless,' and a colleague's LLM chat assistant hallucinated a nonexistent software library.conf: high
supports ←Practices and Challenges of Using GitHub Copilot: An Empirical Study · p4 — This study's systematically coded 2023 limitation categories (difficulty of integration, poor/limited code generation) are the empirically-grounded, contemporaneous counterpart to Brett's informal 2026 recollection that early Copilot autocomplete was 'distracting and basically useless.'
explicitcontestedUsing LLM-powered coding harnesses in earnest for over a year led to depression, apathy, existential dread, and worse software.conf: high
contradictsThe growing AI backlash · p1 — The backlash piece treats coding as one of the few domains where GenAI is net-positive/exempt from its 'net negative' verdict; Brett's own coding-specific experience was net negative for him (depression, apathy, worse software), directly disputing that exemption.
explicitRelying on AI chat as a substitute for professional judgment (medical, career, business advice) made the author 'delusional,' exemplified by an unnecessary, costly midnight ED visit prompted by AI medical advice.conf: high
explicitDelegating the bulk of coding work to AI and shifting into a reviewer/QA role made the author lazy, made him stop caring, made him a worse programmer, and made him depressed.conf: high
supportsKI, utvikling og følelser · p9 — First-person confirmation of the predicted downside risk — alienated, dulled, depressed, and (per the piece's closing uncertainty) potentially exiting the profession.
supportsPodcast: The AI Joy Gap: Why Some Developers Thrive While Others Struggle · p2 — Confirms the joy-gap claim that AI absorbs the creative work and leaves developers with the least enjoyable part — Brett explicitly became 'just a code reviewer and QA tester.'
explicitThe more he used AI for coding, the more distant/alienated he became from the software: unable to explain how it worked without asking the AI, and caring far less about code he didn't write himself.conf: high
supportsKI, utvikling og følelser · p4 — Matches Moe's alienation claim almost exactly — Brett reports no longer recognizing/understanding the code he ships once he stopped writing it himself.
explicitSoftware quality is fundamentally a function of how 'cared for' it is by the humans who build it, not its feature count or build speed; AI-built software is worse because it removes that care.conf: high
explicitAI carries severe, largely hidden environmental costs (data centers, water usage) and is built on stolen training sources.conf: high
supportsAI's environmental costs threaten water, land and climate · p1 — Qualitative, unsourced corroboration (data centers, water usage, hidden costs) of the report's core claim that AI's environmental footprint goes well beyond carbon and is largely invisible to users.
explicitAI is 'unquestionably a net harm compared to any benefits it may provide,' both for the author and for society.conf: high
explicitUsing AI only moderately or for trivial tasks is not a viable middle path — attempted and abandoned, both because it felt arbitrary ('all or nothing') and because any level of use still signals demand and thus complicity in AI's harms.conf: high
implicitAn individual's decision to abstain entirely is a meaningful and appropriate response to a technology's societal harms, even without evidence that individual abstention measurably reduces aggregate demand.conf: medium
explicitReturning to hand-coding hobby projects (a Playdate game-programming book, a Rust/Lua game engine) during his low point restored his enjoyment of programming and confirmed that AI use had been degrading both his craft and his mental state.conf: high
explicitTop-down organizational AI mandates, driven by leadership being 'sold' on AI at a conference rather than by developers' own technical assessment, can force adoption that a practitioner's direct experience contradicts.conf: high
implicitdevelopedA developer's professional identity and sense of purpose depend on authorship and understanding of the code they ship; reducing that to reviewing AI output constitutes an existential threat to identity, not merely a change in job tasks.conf: medium
extends ←How Coders Can Survive—and Thrive—in a ChatGPT World · p13 — This article's 2023 coping strategy for professional relevance (lean into fundamentals, problem-solving, and security judgment) is an early, optimistic answer to the identity-threat problem the later piece states much more starkly: that reducing a developer's role to reviewing AI output constitutes an existential threat to professional identity grounded in authorship and understanding.

Arguments

I should stop using AI entirely, including for trivial/moderate use.

Explicit premises:

Using AI for coding caused depression, apathy, existential dread, and worse software.
Relying on AI chat for personal decisions was actively harmful (the ED anecdote).
AI carries severe environmental costs and is built on stolen training data.
The cost of using LLM tooling is way too high and not worth it for myself or society.
Attempted moderate/trivial-only use did not hold; he kept reverting to using it for everything.

Implicit premises:

Any level of use signals demand and is complicit in perpetuating AI's harms.
Personal well-being and craft/meaning take priority over productivity gains.
The harms are inherent to how current LLM tools work, not fixable by better discipline, tooling, or moderated use.
An individual's abstention is a meaningful and sufficient personal response to a systemic, industry-scale problem.

Inferences:

If moderate use still counts as demand-signaling complicity, the same logic would implicate almost any professional tool with comparable supply-chain or environmental costs — a generalization the author doesn't draw or examine.

AI made me a worse programmer. strength: high