Automatisert KI-koding vil endre alt ("Automated AI coding will change everything")

Henning Meese · article · 2026-01-09 source ↗ Reader 🖥️

AI Bubble Debate AI Labor Disruption Code Was Never the Bottleneck Enterprise AI Adoption & Readiness

AI's eventual takeover of most coding work is near-certain and the industry is already betting on it ('all in'), but the specific timeline keeps sliding later than predicted — illustrated by LessWrong's own 'superhuman coder' threshold estimate being pushed back 3.5 to 5 years in a single revision — so human developers are not about to become unemployed en masse in the near term, even though what role they will play once that threshold does arrive remains genuinely open.

theoretical · argumentative — A short Norwegian tech-outlet (Computerworld/cw.no) opinion piece, extracted and written up in English with quotations kept verbatim in Norwegian. Its central evidence is a specific external forecast (LessWrong's AI timelines/takeoff model) rather than primary research; the author's own argument sits on top of that forecast, reasoning about near-term developer employment given the forecast's own recent revision.

Structure

Part	Role
Framing: AI already generates billions of lines of code worldwide; if a well-formed prompt can build almost anything, running a large human development department starts to look pointless in the long run	context
An explicitly bracketed aside: deploying systems nobody but the AI that built them understands raises obvious security/readiness (beredskap) questions, which the piece names but declines to pursue	context
LessWrong's 'superhuman coder' threshold and its definition (30x the agents, 5% of the budget, best-human-level autonomy, 30x faster than the org's best engineer), and the revision of its estimated arrival from Jan 2027-Sep 2028 to within 2032	evidence
Industry strategic uncertainty: an 'all in' posture so far, but experience that the big AI gains keep arriving later than first assumed, which should cool near-term optimism	evidence
Conclusion: no reason to expect mass developer unemployment imminently; developers are still needed where they are, though their eventual role (possibly strategy/control over AI-driven efficiency) is uncertain	conclusion

Key terms

overmenneskelig koder (superhuman coder)
LessWrong's defined capability threshold: an AI system able to run 30 times as many agents as an organization has human developers, using only 5% of that organization's budget, operating autonomously at the level of one of the best human developers, and completing tasks 30 times faster than the organization's most skilled engineer.

Propositions

explicitAI already generates billions of lines of code worldwide; if a well-formulated prompt can in principle build almost anything, there is little point running a large human development department, making a major long-run reduction in software-development headcount seem close to inevitable.conf: medium
supports46 thoughts on the near future · p1 — Both anticipate substantial further AI capability growth (this piece via LessWrong's 'superhuman coder' threshold, the other via remaining algorithmic orders-of-magnitude), converging on the same general forward trajectory from different source material.
supportsWhat's an Orchestrator—and Why Does Software Need One? · p16 — This piece's speculative, future-dated 'superhuman coder' headcount-reduction claim and the other extraction's already-asserted-as-current claim ('one orchestrator can now do the work of 20-30 developers') point the same direction; the latter offers a concrete, present-tense data point for the former's speculative forecast.
explicitPutting into production a huge volume of systems that essentially nobody — other than the AI that built them — actually understands raises obvious security and emergency-preparedness (beredskap) questions, which the author explicitly flags but declines to pursue as this piece's focus.conf: high
explicitLessWrong's December 2025 revised AI-timelines report pushed back its estimated arrival of a 'superhuman coder' AI system from a prior window of January 2027 to September 2028, out to sometime within 2032 — a delay of roughly 3.5 to 5 years in a single revision.conf: high
explicitDespite near-certainty that AI can code, software companies and organizations have not found it trivial to settle on a strategy; the prevailing posture so far has been 'all in,' but experience so far shows the big gains — both for AI generally and for coding specifically — keep arriving later than first assumed, which will likely have a cooling effect on short-term optimism.conf: medium
explicitBecause the anticipated big AI gains keep arriving later than expected, there is no reason to expect a large wave of unemployed developers in the very near term; for now, human developers are still needed where they already are.conf: medium
explicitIf and when AI reaches the 'superhuman coder' threshold, what role human developers will have is genuinely open; the author speculates they might be needed in strategy and control/oversight roles to manage the increased efficiency AI produces — calling this the best-case outcome, while explicitly stating it is far from certain.conf: low

Arguments

Large-scale developer unemployment is not imminent, even though AI will likely eventually make most coding an AI-performed job. strength: medium

Explicit premises:

The anticipated big productivity gains from AI — in coding specifically and AI generally — have repeatedly arrived later than first assumed.
LessWrong's own 'superhuman coder' threshold estimate was itself just revised 3.5-5 years later in a single update.
Software organizations still need the developers they currently employ, where they currently are.

Implicit premises:

A single forecasting community's (LessWrong's) revised estimate is a reasonable proxy for the industry's actual trajectory, even though that same community's prior estimate just proved substantially wrong in one direction.
The pattern of this one revision (pushed later) is representative, rather than forecasts being equally likely to be revised earlier in response to a future capability jump.

Inferences:

If AI capability-threshold forecasts keep being revised later on roughly this cadence, the 'not long now' framing could remain perpetually true without the threshold ever quite arriving on schedule — worth tracking as a recurring pattern in AI capability forecasting generally, not just in this one report.

Questions addressed

✅ Will AI automation of coding cause a large wave of developer unemployment in the near term?
No — not in the near term, because the anticipated capability thresholds keep arriving later than predicted.
⭕ What happens to deployed systems that nobody but the AI that built them understands, from a security/readiness standpoint?
⭕ What role will human developers play once (or if) AI reaches a 'superhuman coder' threshold?
Speculated as possibly strategy/control/oversight roles, but explicitly left uncertain.

Gaps

incompleteThe piece names what it calls the obvious security and emergency-preparedness (beredskap) risk of deploying systems that nobody but the AI that built them understands, then explicitly declines to pursue it ('det er ikke poenget her') — leaving the single most concrete, near-term organizational risk the article itself raises completely unexamined.
Read next: Material specifically on accountability, on-call ownership, and incident response for AI-authored systems that no human on the team fully understands — e.g. extract-whats-an-orchestrator-why-software-needs-one.json's treatment of accountability-through-understanding-without-reading.
incompleteThe article frames developer headcount reduction as following directly from AI reaching a raw coding-speed/autonomy threshold ('superhuman coder'), without engaging the corpus's recurring counter-finding that code generation speed was never the actual bottleneck in software delivery — so even an AI that codes 30x faster than the best human engineer might not proportionally reduce headcount if requirements-gathering, review, security, and organizational absorption capacity remain the real constraints.
Read next: extract-new-sdlc-with-vibe-coding.json and other 'code was never the bottleneck' material in the corpus.