Anthropic’s Michael Sellitto told an IMF-World Bank panel in Bangkok on 16 October 2026 that Claude led 26% of the work on Anthropic’s next models in August, up from zero at the start of the year. He added that “sometime next year it could be 80–90%”, according to Andrew Curran’s transcription of the remarks on X.
Sellitto set it beside a code curve, per the same transcription: “Last year, about zero percent of our code was written by our model Claude.” By the end of that year, he said, “we were at 80–90% of the code in the company being written by the model.”
Business Pill · PRETRAINING
A one-minute explainer of pretraining: how an AI model learns before anyone teaches it. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, Sellitto put a date on a curve Anthropic had already started to measure: the 26% share of model-building work Claude led in August matches Anthropic’s own published index, and his projection, as Andrew Curran transcribed it, is that “sometime next year it could be 80–90%”.
The panel on AI and cyber risks at the IMF-World Bank Annual Meetings 2026 in Bangkok was moderated by IMF Deputy Managing Director Bo Li, according to a report by The Standard in Thailand. Alongside Sellitto were FINMA chair Marlene Amstad, Central Bank of Brazil deputy governor Paulo Picchetti and BIS deputy general manager Andrea Maechler, the report says.
Curran, who posted the quotes at 17:30 UTC, wrote: “This is the first time Anthropic has given a projection for what percentage of frontier model training will be done by Claude in 2027.”
In Curran’s transcription, Sellitto asked: “What percentage of training the next-generation models is Claude leading today?” He answered: “At the beginning of the year it was zero percent. In April it was 3%. In August, 26% of the process of building the next generation of models is being led by Claude.”
He then said: “And we can project out – and we had a blog post about this not too long ago – that sometime next year it could be 80–90%.”
Anthropic’s Michael Sellitto said on 16 October 2026 that Claude led 26% of the work on Anthropic’s next models in August, up from zero at the start of the year and 3% in April, and that sometime next year it could be 80–90%, according to Andrew Curran’s transcription.
What Was Posted on X
Quoting from Anthropic’s Michael Sellitto at the IMF Artificial Intelligence and Escalating Cyber Risks panel today. This is the first time Anthropic has given a projection for what percentage of frontier model training will be done by Claude in 2027. In March this number was 1%. By next year they are projecting 80-90%.
‘Last year, about zero percent of our code was written by our model Claude. By August of that year, about 50% of the code inside the company was written by the model. That’s because we released a new model in August of last year, and the engineers said, โActually, this is the first time that Claude is writing code at my level.โ Anthropic only hires the best software engineers in the world, and it was at that point that Claude finally got to their level. In August we were at 50%. By the end of last year we were at 80โ90% of the code in the company being written by the model.
What percentage of training the next-generation models is Claude leading today? This is following a similar step change. At the beginning of the year it was zero percent. In April it was 3%. In August, 26% of the process of building the next generation of models is being led by Claude. And we can project out – and we had a blog post about this not too long ago – that sometime next year it could be 80โ90%.’
Andrew Curran posted his transcription of Anthropic’s Michael Sellitto at the IMF panel on 16 October 2026. It is the source of every Sellitto quotation in this piece; his own introduction calls it the first time Anthropic has given such a projection.
Anthropic has singlehandedly compressed RSI timelines by about a year.
Andrew Curran, who transcribed Michael Sellitto’s IMF remarks, posted on 16 October that Anthropic has “singlehandedly compressed RSI timelines by about a year”. That is his own reading, not an Anthropic statement.
SITUATION DETECTED: Anthropicโs Head of Global Affairs Michael Sellitto has told the IMF that Anthropic projects Claude could lead 80-90% of frontier model training by next year, up from 3% in April.
MTS posted on 16 October that Anthropic’s head of global affairs, Michael Sellitto, told the IMF that Anthropic projects Claude could lead 80-90% of frontier model training by next year, up from 3% in April. That is the poster’s summary of the remarks.
The Code Curve He Compared It To
Sellitto described the code share first. “By August of that year, about 50% of the code inside the company was written by the model,” he said, per Curran. He tied that to a model release: “That’s because we released a new model in August of last year, and the engineers said, ‘Actually, this is the first time that Claude is writing code at my level.’”
Turning to model building, he described it as “following a similar step change.”
What Anthropic Has Published
The 26% matches Anthropic’s own published figure. Its page “Measurements for understanding the pace of AI development inside frontier labs”, released in September, states: “As of August 2026, Claude is not operating fully autonomously for any measured subset of AI R&D work.”
The same page says Claude “leads” “26% of Anthropic’s AI R&D work”. It adds that the share of work at or above the level it calls “AI collaborates” “is above 90%”.
As we read it, the code curve is the template he used: per Curran’s transcription, Sellitto said Claude went from about zero percent of Anthropic’s code to about 50% by August and 80–90% by the end of that year.
As we read it, the August data point is the one that can be checked: Anthropic’s September page says Claude “leads” “26% of Anthropic’s AI R&D work” and is not fully autonomous on any measured subset.
As we read it, the projection rests on the remarks alone: we found no recording, and the Anthropic page we read gives no figure for next year.
Michael Sellitto, Anthropic, per Andrew Curran’s transcription (16 October 2026)
“In August, 26% of the process of building the next generation of models is being led by Claude.”
Three Implications
AI LABS Sellitto said the share of model-building work Claude leads went from zero at the start of the year to 3% in April and 26% in August, per Curran’s transcription.
REGULATORS AND THE IMF AUDIENCE The remarks came at an IMF-World Bank panel on AI and cyber risks moderated by IMF Deputy Managing Director Bo Li, with FINMA, the Central Bank of Brazil and the BIS, per The Standard.
ANYONE READING THE PROJECTION Anthropic’s page states Claude is not operating fully autonomously for any measured subset of AI R&D work as of August 2026; the 80–90% figure is a projection, not a measurement.
The framework argues: “Model capability now compounds on a six-to-twelve-month doubling cycle”.
As we read it, Sellitto’s remarks describe one input to that compounding: the model doing more of the work of building its successor. The framework predates the panel and does not discuss Anthropic’s internal figures.
Per Andrew Curran’s transcription of Michael Sellitto (16 October 2026): zero at the beginning of the year, 3% in April and 26% in August. Anthropic’s September page also gives 26% for August 2026.
What Is Not Established
We could not find a recording or an IMF transcript of the panel, so every quotation here comes from Curran’s transcription and we could not check it against the audio.
Curran’s introduction says the number “In March” was 1%; the quotes he transcribed give zero at the start of the year and 3% in April. We report the transcribed quotes.
Sellitto mentioned a blog post on the projection; the Anthropic page we read gives the 26% August figure but no projection for next year. We did not contact Anthropic.
Business Engineer Framework
The Product Overhang Doctrine
A Business Engineer framework on why model capability compounds faster than the products built on it.
Anthropic’s Michael Sellitto said at an IMF-World Bank panel on 16 October 2026 that Claude led 26% of the work on Anthropic’s next models in August, up from zero at the start of the year. Sometime next year, he said, it could be 80–90%, according to Andrew Curran’s transcription.
The 26% matches the figure in Anthropic’s own published index; the projection for next year comes only from the remarks as transcribed.
A note on sourcing. Every quotation from Michael Sellitto comes from Andrew Curran’s transcription of the IMF-World Bank panel, posted on X on 16 October 2026; we found no recording or official transcript to check it against. The venue and panellists come from The Standard’s 16 October report, and the August figure is checked against Anthropic’s own September 2026 measurement page, which we read in full. We did not contact Anthropic. The references to our earlier reporting link to our own pieces.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur, HackerNoon and Search Engine People. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
