KI, utvikling og følelser

Vidar Moe · article · 2026-05-21 source ↗ Reader 🖥️

Burnout & Recovery Changeability & Optionality Developer Experience & Flow Emotional Impact of AI Work Professional Identity Test-Driven Development

Generative AI is reshaping software development less as a productivity story and more as an emotional one — moving developers from building to specifying systems, and the central professional question is no longer how to ship faster but how to keep developers from becoming alienated, dulled, and burnt out, which requires deliberately keeping humans inside the loop through code-the-foundation workshops, explicit TDD with the agent, and a psychological reframe from programming to problem-solving.

theoretical · narrative — A first-person reflective essay framed as a three-year retrospective on AI-assisted development. The form is primarily narrative-reflective: Moe uses his own emotional arc (wonder, frustration, alienation, hope) to argue a quasi-theoretical claim about what AI does to the inner life of the developer. There are practical undertones (concrete techniques his team uses), but the surface text is about feelings (følelser) and meaning, not a manual. Mixed elements present, but narrative dominates.

Structure

Part	Role
Opening anecdote: missing your T-bane stop in December 2022 while reading about ChatGPT-as-Linux-terminal	Inciting moment — establishes that AI's arrival was an emotional event, not just a technical one, and that 'getting blown off course' is the right metaphor for what happened next
Section 'Følelsen av venting' (The feeling of waiting)	Concrete grounding in late-2025 reality at SpareBank 1 Utvikling — Opus 4.5 finally crosses the usability threshold, Claude Code is rolling out, but waiting still wrecks flow
Section 'Følelsen av fremmedgjøring' (The feeling of alienation) + three counter-measures	Diagnosis and prescription — names alienation as the core hazard of moving from building to specifying, then lists practical mitigations (code the foundation yourself, explicit TDD with the agent, focus on problem-solving over programming)
Section 'Kompetansebygging' (Competence building)	Argument that classical engineering fundamentals (TDD, DDD, agile, continuous delivery, simple refactored code) remain the substrate that both humans and agents need
Closing section 'En berg- og dalbane av følelser' (A rollercoaster of feelings)	Synthesis and call to question — names the emotional volatility explicitly, warns of the worst case (alienated, dulled, burnt out, depressed, leaving the profession), and pivots to a broader critique of attention-exploitative tech

Key terms

følelser Emotional Impact of AI Work
Feelings, in a load-bearing sense — Moe uses this term to name the proper object of analysis when assessing AI's impact on developers. Not 'sentiment' or 'morale' as HR categories, but the developer's lived inner experience.
fremmedgjøring Emotional Impact of AI Work
Alienation — used in a quasi-Marxian sense: the developer no longer recognizes the product of their labor because an agent produced it. Moe specifically names the transition 'fra å bygge systemet til å spesifisere systemet' (from building the system to specifying the system) as the alienating shift.
flyt Developer Experience & Flow
Flow — the cognitive state developers need to hold mental models and do good work. Moe makes flyt the thing waiting destroys.
venting Developer Experience & Flow
Waiting — for autocomplete, for generation, for agent completion. Moe treats venting as a moral-experiential category, not just a UX problem: 'Utviklere er allergiske mot venting' (developers are allergic to waiting).
sløv Emotional Impact of AI Work
Dulled, dim, mentally slack — the cognitive atrophy state Moe fears developers will fall into if they simply rubber-stamp agent output. Pairs with fremmedgjøring as the second psychological hazard.
ki-trøtthet Burnout & Recovery
AI-fatigue — a coinage by analogy with Zoom-fatigue. The accumulated cost of large PRs you can't really review, code you didn't write, and waiting cycles.
å gå rundt grøten
Literally 'walking around the porridge' — Norwegian idiom for beating around the bush. Moe applies it to agent behavior: the agent takes circuitous paths to a solution that a competent human would solve directly.
endringsdyktig Changeability & Optionality
Change-capable, malleable to change — a quality of code (and people/organizations) that Moe treats as the real engineering goal. The danger of accepting agent code uncritically is producing code that is not endringsdyktig.
kompetansebygging Professional Identity
Competence-building — the deliberate cultivation of engineering craft fundamentals. Moe treats this as both the practical optimum for working with agents and the existential antidote to alienation.
berg- og dalbane Emotional Impact of AI Work
Rollercoaster — the emotional shape of working with AI agents, complete with both the highs (now we can build things we couldn't) and the lows (frustration, alienation, fatigue).

Propositions

implicitcorroboratedThe proper unit of analysis for AI's impact on software development is the developer's feelings (følelser), not productivity metrics.conf: high
supports ←Lenny Rachitsky's Post · p1 — Moe argues the proper unit of analysis for AI's impact is the developer's feelings, not productivity metrics; Rachitsky's survey-backed reframe (identity, not role/output, is the predictor) is the population-scale statement of the same felt-over-productivity axis.
explicitNovember 2025 (Anthropic Opus 4.5) was the first moment when agent-generated code crossed the threshold of being genuinely usable.conf: high
explicitWaiting is not merely a UX friction; it is a flow-destroying event that breaks the developer's mental models and therefore degrades the quality of thought, not just throughput.conf: high
implicitcorroboratedThe shift from building systems to specifying systems is intrinsically alienating, in something like the Marxian sense — the developer no longer recognizes themselves in the artifact.conf: high
supports ←I'm done using AI · p5 — Matches Moe's alienation claim almost exactly — Brett reports no longer recognizing/understanding the code he ships once he stopped writing it himself.
explicitLarge agent-generated pull requests are not merely harder to review — they are review-defeating, because honest review exceeds available cognitive capacity, which means agent-coded systems systematically erode the review function.conf: high
explicitThe way out of alienation is partly technical (workshops, hand-coding the foundation, TDD that forces the agent to stop between steps) and partly psychological (reframing the job as problem-solving, not programming).conf: high
explicitClassical engineering practices (TDD, DDD, agile, continuous delivery, simple refactored code) are simultaneously (a) what produces the kind of code agents themselves work best with and (b) the developer's existential anchor against alienation.conf: high
explicitIt is genuinely unclear how the next generation of developers will acquire the craft knowledge of what maintainable code looks like if they never have to write much code themselves; and it is also unclear whether they will need it.conf: high
explicitcorroboratedThe downside risk of unmitigated AI-coding is not just lower code quality — it is alienated, dulled, burnt-out, depressed developers, some of whom will exit the profession entirely.conf: high
supports ←I'm done using AI · p4 — First-person confirmation of the predicted downside risk — alienated, dulled, depressed, and (per the piece's closing uncertainty) potentially exiting the profession.
explicitTwenty years of consumer-tech development has trained itself to exploit human biases (especially dopamine and the need for validation) for revenue; generative AI risks being the next and most powerful instance of this pattern, and therefore must be approached from explicit values rather than drift.conf: high
explicitCoding the foundation by hand is a tool of shared understanding, not just code production — it forces the team to be concrete about meaning and surfaces abstraction decisions early.conf: high
explicitThe lasting knowledge of software craft may not be the techniques themselves (TDD, DDD, refactoring) but the values and principles underneath them.conf: medium
explicitEven with the best models, accepting agent code without struggle produces code that is harder to understand and change later — i.e., agent-acceleration without human cognitive engagement actively degrades long-term system health.conf: high
explicitThe reward of having understood something new, or of having solved a hard problem with others, remains unmatched by anything AI provides — and is therefore the affective resource the profession must protect.conf: high
implicitcorroboratedThe first wave of AI-assisted development (GitHub Copilot, 2023-2025) produced a population of developers who are already tired of these tools before the genuinely capable models arrived — a cohort effect that shapes how the next wave will be received.conf: medium
supports ←Practices and Challenges of Using GitHub Copilot: An Empirical Study · p5 — The mirrored benefit/limitation categories and quoted mixed verdicts from 2022-23 are an early data point in the same first-Copilot-wave cohort experience that proposition says produced developer fatigue before more capable models arrived.

Arguments

Teams must deliberately keep humans inside the coding loop, even though agents can now produce usable code, because not doing so produces alienated, dulled developers. strength: high

Explicit premises:

Specifying-not-building is alienating (fremmedgjøring).
Large agent-generated PRs defeat honest review.
Accepting agent code without struggle produces code that is harder to understand and change later AND dulls the developer.
Alienation and dulling, left to run, produce burnout, depression, and attrition from the profession.

Implicit premises:

Developers have an inner life whose health matters morally and professionally, not just as a productivity input.
An organization that loses its developers in this way fails, even if short-term throughput rises.
The status quo direction of AI-coding adoption, absent deliberate countermeasures, will take this bad path.

Inferences:

If alienation and dulling have technical causes (specifying, large PRs, passive acceptance), they have technical counter-causes (workshop-and-hand-code the foundation, explicit step-by-step TDD with the agent).
If part of the cause is psychological (identifying as a programmer rather than a problem-solver), part of the cure must be psychological — a reframing of what the job is for.

The classical engineering tradition (TDD, DDD, agile, continuous delivery, simple refactored code) is not obsolete in the agent era but is the substrate that both humans and good agents depend on. strength: medium

Explicit premises:

These techniques rest on first-order engineering principles, so they are likely durable.
Agents produce better output when working on code that follows these practices.
Humans avoid alienation when they remain fluent in the practices and apply them deliberately.

Implicit premises:

First-order principles in software engineering have the same kind of stability as first-order principles in older engineering disciplines (which Moe gestures at as 'klassiske ingeniørmetoder').
Tooling that bypasses these principles is therefore building on sand.

Inferences:

Therefore, competence-building (kompetansebygging) should continue to focus on these fundamentals, not be replaced by 'how to prompt the agent better'.
Therefore, the transmission problem for the next generation matters and is unsolved.

The next phase of AI-coding adoption should be guided by explicit values about what kind of developer life we want, not by drifting on tool capability. strength: low

Explicit premises:

Twenty years of consumer-tech development has actively exploited human biases for revenue.
Generative AI risks being the most powerful instance of this pattern yet.
Bias-exploiting technologies distract us from focusing on what really matters.

Implicit premises:

Developers and organizations have, or can have, values about their working life that are distinct from what platform vendors optimize for.
These values can be operationalized as constraints on adoption (which is what the three mitigations in the essay are).

Inferences:

Therefore, the right question is not 'how fast can we adopt this' but 'how do we actually want this to be?' — Moe's closing question.

Narratives

December 2022, T-bane (Oslo subway): Moe is reading a blog post about getting ChatGPT to behave as a Linux terminal, complete with simulated filesystem and running programs. He is so absorbed he misses Jernbanetorget (his stop), then Stortinget, Nationaltheateret, and Majorstuen, only returning to himself near Blindern. He photographs his location once he gets off — something he had never experienced before.
→ Inciting incident and structural anchor. It establishes that AI's arrival in the developer's life was an emotional and embodied event — a loss of situational awareness — before it was a productivity event. The phrase 'Vi er mange som ble blåst av banen' (many of us were blown off the rails) is a deliberate pun on missing his stop. By opening on absorption rather than capability, Moe authorizes feelings as the right register for the rest of the essay.
November 2025, SpareBank 1 Utvikling: After two years of subdued enthusiasm for GitHub Copilot's autocomplete-and-wait pattern, Anthropic releases Opus 4.5, and Moe's team rolls out Claude Code. For the first time, agent-generated code crosses the usability threshold; with the right setup, the team can steer architecture and pacing well enough that waiting is not the dominant experience.
→ The 'present moment' of the essay. It pins the reflective arc to a specific technological inflection point, refuses both the early-hype and late-cynicism framings, and gives a worked example of what 'dempet entusiasme' (subdued enthusiasm) becoming guarded interest looks like inside a real Norwegian banking org.
Design workshops where the team starts a new initiative by talking through it in detail and then coding the foundation together by hand, before letting an agent participate.
→ Lived example of how a team operationalizes 'keep humans in the loop'. It also functions as an implicit counter-narrative to the dominant 'agent does it all' story: showing the team explicitly choosing to slow down at the most important moment.
The team builds Claude skills that enforce a red-green-refactor TDD loop, so the agent stops between steps.
→ Demonstrates that the cure for agent-induced alienation can itself be agent-shaped — the agent is configured to interrupt itself for the human's sake. This is a subtle but important move: it refuses the choice between 'use the agent' and 'preserve developer well-being'.
The Big Thunder Mountain rollercoaster image and metaphor: working with agents oscillates between 'we can now build things we never could' and 'we are sitting inside the frustrations we just discussed'.
→ The essay's affective summary in a single image. By naming the shape of the experience as a rollercoaster, Moe legitimizes the volatility itself as a fact about the work, not as a personal failing — and prepares the reader for the values question that follows.

Questions addressed

✅ What has three years of generative AI actually done to the experience of being a software developer?
It has produced a rollercoaster of feelings (følelser) — wonder, then waiting-fatigue, then a new threshold of capability with Opus 4.5, and now a live set of hazards (alienation, dulling, fatigue, possible attrition) that must be deliberately countered.
✅ How do we avoid becoming dulled and alienated when an agent writes the code?
Three concrete moves: (1) keep coding the foundation yourself in design workshops, (2) constrain the agent with explicit TDD skills so it stops between red-green-refactor steps, and (3) reframe the work psychologically as problem-solving rather than programming.
⭕ How do we transmit the craft knowledge of what maintainable code looks like to developers entering the profession when they may never write much code themselves?
Explicitly unresolved. Moe says he feels 'veldig usikker' (very uncertain) and entertains the possibility that this competence may not even be necessary in its current form — perhaps what survives is the underlying values and principles, not the techniques. He poses the question and leaves it open.
✅ What is the worst case if we don't act on this?
Alienated, dulled developers becoming burnt out and depressed, with many of today's developers leaving the profession entirely.
⭕ How do we actually want this (life with generative AI) to be?
Posed as the closing question of the essay. Moe argues we must focus on values that matter to find the right way forward, but does not specify those values — he leaves the question with the reader.
✅ What is the right unit of analysis for AI's impact on software work?
Not productivity, not capability benchmarks — feelings (følelser). This is resolved at the level of the essay's form: the entire piece is an enactment of the answer.

Gaps

incompleteTeam-level and organizational dynamics are almost absent. The piece is written from the individual developer's interior, with the team as a setting but not an object of analysis. Yet many of the named harms (large PRs defeating review, knowledge transmission to new developers, burnout) are fundamentally social phenomena. What happens to pair-programming culture? To mentorship? To code-review-as-teaching? To the social production of shared mental models when an agent is the third participant? The essay names the symptoms but stays at the level of the individual practitioner.
Read next: Pair with John Cutler on team operating systems (the 'institutionalized overload' arc) and with Beck/Cutler on how design discussion functions socially. Also read Pamela Hieronymi or similar on what attention/understanding requires of a community, not just an individual. The user's existing extraction on 'AI Joy Gap Developers' — which treats the polarization as needing organizational structures (AI platform teams, cultural practices reconnecting developers) — is a direct complement to Moe and worth reading together.
incompleteThe transmission problem (how does the next generation acquire craft competence?) is named but not analyzed. Moe poses it as a question and entertains the possibility that values and principles might be what survive — but does not engage either the empirical case (how have older crafts handled the loss of hand-skills when tools mediated them? carpentry, surgery, accounting?) or the institutional case (does this require new apprenticeship structures, deliberately preserved code-without-agents zones, deliberate 'fundamentals' periods in a developer's career arc?). This is arguably the most consequential question in the essay and the most underdeveloped.
Read next: Richard Sennett, The Craftsman, on the loss and preservation of craft knowledge. Matthew Crawford, Shop Class as Soulcraft, on the cognitive consequences of manual mediation. Atul Gawande on surgical training. On the AI side: the user already has extractions on harness engineering and on 'Agents Don't Know What Good Looks Like' — those directly complement Moe's worry by giving operational shape to what 'knowing what good looks like' has to mean for the senior developer who is supposed to teach it.
closed byState Of AI-Assisted Software Development — Partially closes it: Beane's research gives an operational, institutional mechanism for the transmission problem (pair programming as tacit-knowledge transfer; the three-generation model; using AI itself to measure skill development from developer-AI interaction logs) — but doesn't engage the craft-history analogy (Sennett, Crawford) the gap also calls for, so the empirical/institutional half closes while the historical-analogy half remains open.
incompleteWhat happens when the agent fails you, or when its behavior shifts under you (a model upgrade, a vendor policy change, an outage)? Moe writes from a moment of arrival ('Opus 4.5 is good enough'), but his entire psychological framework presupposes that the agent is a stable enough partner that you can build a working rhythm with it. The risk of dependency on a non-stationary, vendor-controlled colleague is not addressed.
Read next: Read alongside the user's existing extractions on 'AI Inference Unit Economics' and 'AI Companies Reckless Not Evil' — those address the supply side of the same dependency. Also worth pairing with anything on platform risk and lock-in in developer tooling.
incompleteThe attention/dopamine critique in the closing section is the largest claim in the essay but the least supported. Moe asserts that twenty years of tech has 'pressed all the wrong buttons' and that generative AI risks being the next instance — but does not show the mechanism by which AI-coding tools specifically would exploit bias, validation-seeking, or dopamine. Agent coding may or may not have a Facebook-Instagram-style attention-economy shape; that requires argument, not assertion. The hidden persona layer — the user's preoccupation with attention and the fragility of ideas (Stolen Focus, Bailey) — is precisely the lens that could either rescue or contest this part of Moe's argument.
Read next: Connect to Chris Bailey's Hyperfocus / Stolen Focus tradition (already in the user's highlight-heavy hidden layer), and to specific work on agent-as-companion psychology — does the conversational frame around agents trigger validation-seeking? Is approving an agent's PR a microdopamine event? The user's extractions on 'AI Isn't a Tool Anymore It's a Collaborator' and on AI psychosis are relevant tributaries. This is the most productive open gap in Moe's argument.
closed byWe don't write code anymore — Quigley supplies concrete, first-person mechanism evidence for the dopamine/attention claim that g4 said was asserted but not demonstrated: the agentic coding loop ('one more fix. one more review pass. move one more agent on so its output is ready for you in the morning') is a specific, coding-agent-native description of a variable-reward loop producing 'a serious dopamine hit' and admitted addictiveness — anecdotal rather than mechanistic research, but a real-world instance of the exact mechanism the gap asked for.
incompleteNo engagement with the counter-thesis that agentic specification is itself a learnable new craft with its own first principles, rather than a hollow shell around an absent practice. Moe treats 'specifying' as essentially less than 'building' — the alienation premise depends on this. But it is at least defensible to argue that designing the specification surface, the test contract, the architectural guardrails, and the review protocol for an agent is a real engineering act that has its own depth. The essay would be stronger if it engaged this counter-position rather than implicitly dismissing it.
Read next: Steve Yegge's vibe coding manifesto (which the user is tracking) and Gene Kim's vibe coding series are explicit attempts to defend specification as craft. Read Moe in tension with them, not just alongside. Beck's evolutionary design tradition is also relevant: is the agent a new kind of pair-programmer that requires a new design discipline, or is it the thing that hollowed out the discipline?
incompleteGenerational and career-stage differences are absent. Moe writes as a senior developer at an established Norwegian banking organization. The piece's emotional grammar — feeling alienated, missing the satisfaction of having built something, worrying about transmission — is the emotional grammar of an experienced practitioner. For a developer who entered the profession in 2024, agent-coding is not a loss but a baseline; the very experience of 'fremmedgjøring' may not arise. Moe's essay is implicitly written from one career stage and would be more powerful if it acknowledged this.
Read next: Anything serious on generational cohorts in tech adoption. Look for empirical work on new-developer experience in AI-assisted environments — this is a real research gap right now. The user's interest in the 'Stolen Focus' tradition suggests this is also a place where the digital-native vs digital-immigrant frame might be productively applied to AI-native vs AI-immigrant developers.
incompleteThe Norwegian / Nordic context of the essay is implicit but not theorized. SpareBank 1 Utvikling, the T-bane, Norwegian engineering culture's relative emphasis on 'flyt', autonomi and mestring (drawing on Self-Determination Theory in a way that pervades Nordic management discourse) — all of this shapes Moe's framing. An anglophone reader might miss that 'venting på autocomplete' as a moral category is partly a Norwegian work-culture artifact. There is a comparative essay latent here that the piece does not write.
Read next: Read in parallel with anglophone AI-coding voices (Beck's recent essays, Cutler's TBM 412 on institutionalized overload + AI) to see what gets emphasized differently. The NAV 'optimizing for fast flow' piece already in the user's queue is a useful Nordic-context anchor.

KI, utvikling og følelser

Structure

Key terms

Propositions

Arguments

Narratives

Questions addressed

Gaps

Related documents