Dario Amodei on Coding Going Away First: What He Actually Said
By Matt Cullerton, founder of Mavric and Apso
Anthropic CEO Dario Amodei told podcaster Nikhil Kamath in February 2026 that "coding is going away first," and that the broader job of software engineering will follow more slowly. A TikTok from @investbesrclips cuts that exchange under the banner "Coding is dying / Learn THIS instead." The quotes are real, but in the same answer Amodei says end-to-end software engineering "is going to happen as well," so the clip offers less of a safe alternative than its title promises.
Where is the clip from?
The speaker is Dario Amodei, co-founder and CEO of Anthropic. The interviewer is Nikhil Kamath, the Zerodha co-founder who hosts the People by WTF podcast. The footage comes from the episode "The AI Tsunami is Here & Society Isn't Ready," published on February 24, 2026 and running about 69 minutes. A published transcript contains the clip's lines word for word and places the recording in Bangalore.
The TikTok was posted on October 7, 2026, roughly seven and a half months later, with no date, episode title or link.
Did Amodei say coding is dying?
He said coding goes first. The clip opens on that line, "I think coding is going away first," and then rewinds to the question that produced it.
In the full conversation, Kamath asks which industries get disrupted and which have runway left, framed around a young person in India choosing a book, a college or a skill. Amodei first answers in general terms about human-centred work. Kamath then asks him to separate coding from engineering. The coding-versus-engineering split was Kamath's framing, offered as a follow-up.
What did Amodei say comes after coding?
His full answer is that coding goes first "and then the broader task of software engineering will take longer." In the same breath he adds, "But I think that is, doing that end-to-end, I think that is going to happen as well."
He names design, understanding users, sensing demand and managing teams of AI models as things that "may still be present." He hedges with "may" and closes with a ceiling: "at some point, you get to 99%, and then it becomes harder." The speaker is describing a refuge that shrinks, which is different from the "learn THIS instead" the banner sells.
For context on his timelines, Amodei told The Economist at the World Economic Forum in January 2026 that AI models could do "most, maybe all" of what software engineers do within six to twelve months, according to Entrepreneur.
Can AI make developers 20 times more productive?
Amodei's math in the clip is explicit. If a human does 5% of a task and AI does 95%, the human's output multiplies by 20, which is 1 divided by 0.05. That is an upper bound. It holds only if the human's 5% never becomes the bottleneck and the AI's 95% needs no rework.
The one randomized controlled trial on this question found the opposite in its setting. METR had 16 experienced open-source developers work 246 real issues from their own repositories, with AI tools randomly allowed or forbidden per issue. Developers forecast a 24% speedup and reported a 20% speedup afterward. Measured, they took 19% longer with AI.
METR says it does "not claim that our developers or repositories represent a majority or plurality of software development work," and the study used early-2025 tools. It does not refute Amodei. It does show a 39-point gap between felt and measured speedup, which is a reason to read 20x as a model and not an observation.
Is AI better than radiologists?
The clip ends with "similar to the radiologist example I gave," but never shows it. Earlier in the episode Amodei argued that AI has become better than radiologists at reading scans, that the number of radiologists has not fallen, and that what remains is walking patients through results.
Radiologists dispute this. Ben White, a practising radiologist, called the claim "completely baseless and without a reality correlate," arguing that the technical core of the job has not changed much and that patient-facing work is a smaller share than Amodei implies. The American College of Radiology's June 10, 2026 bulletin piece "Radiologists Will Adapt" cites projections that imaging demand will grow at least as fast as the radiology workforce, which describes a shortage.
Amodei offers no study for the scan-reading claim in the episode, so treat it as his assertion.
Bottom line
Amodei did say coding goes away first, and he also said end-to-end software engineering will follow, with the remaining human share getting smaller near 99%. His 20x figure is arithmetic under ideal conditions, and the one controlled trial measured a 19% slowdown with early-2025 tools. The radiologist comparison he leans on is contested by radiologists and by the ACR's workforce data.
Sources
- The AI Tsunami is Here & Society Isn't Ready - the full People by WTF episode, February 24, 2026
- Episode transcript - the clip's lines verbatim
- CXOToday episode writeup - publication date and guests
- Anthropic's India expansion announcement - Amodei's title and the Bengaluru office
- METR randomized trial - measured vs self-reported AI speedup
- Ben White, "Dario Dreams of Electric Radiology" - a radiologist's rebuttal
- ACR Bulletin, "Radiologists Will Adapt" - radiology workforce projections
- Entrepreneur on Amodei at Davos - his six-to-twelve-month timeline
Published October 7, 2026. Based on a TikTok by @investbesrclips. Full video notes (markdown)