Amodei: AI Could Replace Engineers in 6-12 Months

At the Davos panel, Amodei framed his prediction as more than idle speculation: he said engineers inside Anthropic had already stopped writing code by hand, telling The Economist's Zanny Minton Beddoes, 'I don't write any code anymore. I just let the model write the code. I edit it.' ^[1]But he hedged the timeline itself, noting that some parts of the engineering lifecycle - chip manufacturing, model training - can't yet be automated end-to-end: 'It's a question of how fast does that loop close.' ^[1]The same 'Day After AGI' panel also surfaced a related but distinct framing of what would determine the pace of change: 'The biggest thing to watch is AI systems building AI systems. Whether that loop closes will determine if it's a few more years or if we have wonders and a great emergency in front of us.' ^[3]The two quotes describe the same underlying uncertainty - a recursive AI-building-AI loop - from two different vantage points, one about engineering specifically, the other about AGI more broadly. On X, the claim traveled fastest through prediction-market and news-aggregator accounts simply re-broadcasting the bare headline number, rather than through developers debating its substance - a shallower reaction than the more skeptical, detailed pushback that took shape on Reddit.

Critics reacted as much to who was saying it as to the claim itself. Software architecture figure Grady Booch accused Amodei of not understanding software engineering and of 'working feverishly to pump up the valuation of his company' ahead of Anthropic's planned IPO ^[4]. AI researcher Gary Marcus called the claim 'vastly overhyping what autonomous AI is currently in a position to do' ^[4], while engineer and writer Gergely Orosz dismissed it outright, saying only non-coders would find it credible ^[4]. IBM CEO Arvind Krishna offered a data-flavored rebuttal, putting the realistic current share of AI-written code closer to 20-30 percent rather than the near-total automation implied by Amodei's framing ^[2]. Reddit's reaction mirrored this almost exactly: engineers argued that 'coding' and 'software engineering' are not the same job, since the latter includes requirements-gathering, architecture, legacy-system debugging, and organizational back-and-forth that current models still handle poorly, and commenters repeatedly recalled that Amodei had made an even bolder, unfulfilled claim in March 2025 that 90 percent of code would be AI-written within three to six months.

The claim did not age quietly. By late May 2026, once most of the 6-to-12-month window had elapsed, both Amodei and Sam Altman softened their earlier job-apocalypse rhetoric, reframing automation as a productivity multiplier rather than a displacement event ^[5]. Amodei's own language shifted accordingly: 'If you automate 90% of the job, then everyone does the 10% of the job... the 10% kind of expands to be 100% of what people do and kind of 10-times their productivity' ^[5]. Reporting tied to the walk-back cited Yale Budget Lab data showing no significant occupational shift in high-AI-exposure jobs since ChatGPT's 2022 launch, undercutting the more dramatic version of the original forecast ^[6][7]. The timing was hard to miss: the retreat coincided with both companies approaching large IPOs, the exact incentive critics like Booch had flagged back in January.

In a separate interview with CNBC-TV18 at the India AI Impact Summit - held after a selloff in Indian IT services stocks - Amodei struck a notably softer tone than the one that rattled Davos. He said Anthropic is 'not trying to replace' the existing IT industry, describing Anthropic's models as 'Lego building blocks' that other companies build on top of, while still conceding 'inevitably there will be some disruption... because the technology is moving so fast.' In the same conversation he repeated a separate, more specific number: that roughly 50% of entry-level jobs may disappear within three years, a figure that extends the entry-level consulting, legal, and finance warnings that circulated after his earlier comments beyond just software engineering. The contrast is pointed: Nvidia CEO Jensen Huang, in the same CBS News segment that carried Amodei's original Davos-adjacent warning, argued the opposite case - that the AI-driven infrastructure buildout will be a net job creator, not a destroyer. Taken together, the two interviews suggest Amodei's message shifts with the audience: bold, headline-grabbing automation timelines for investors and Davos, and a gentler 'building blocks, not a replacement' framing when addressing an IT sector already spooked by his own predictions.