In a new interview, Altman separated fast-moving model capability from slower business adoption, extending a retreat from earlier job-loss forecasts.
By Ryan Merket · Published
Primary source: Fireside Alpha on X
Why it matters #
Altman's revision shifts AI's near-term bottleneck from model capability to deployment, giving incumbents more time and forcing startups to solve organizational adoption as well as technology.
Sam Altman (@sama) says he overestimated how quickly advances in artificial intelligence would translate into business disruption, arguing that economic and social adoption will move more slowly than model capability.
https://x.com/firesidealpha/status/2091506987137896560 The OpenAI co-founder and CEO made the admission in an interview with David Senra released on August 23rd, 2026. Altman, who co-founded Loopt before serving as Y Combinator's president from 2014 to 2019, has spent much of his career betting that technical shifts create openings for new companies. His revised forecast puts considerably more weight on the institutions, habits and buying cycles that stand between a capable model and an altered economy.
"I thought when we got to GPT-4, which was back in 2023, that very quickly after that there was going to be much more disruption," Altman said in a clip posted by Fireside Alpha. He expected software businesses to become vulnerable sooner, he said, before concluding that "the economy just has so much inertia."
GPT-4 launched on March 14th, 2023. OpenAI described it at the time as a multimodal model that performed at human levels on several professional and academic benchmarks, while acknowledging that it remained less capable than people in many real-world situations.
Altman's revised view separates two timelines that were often bundled together during the generative AI boom: how quickly models improve and how quickly organizations rebuild work around them. Models can gain new capabilities on a laboratory or product-release schedule. Businesses still have to change software, data access, budgets, compliance processes, employee responsibilities and customer expectations before those capabilities produce measurable economic disruption.
Capability moved faster than adoption
Altman said the lag could make the transition "smoother and slower," framing economic inertia as a stabilizing force rather than evidence that AI development has stalled. He still expects AI to rank among humanity's most consequential technologies. His concession concerns the speed of diffusion through businesses and society.
The comments extend a shift Altman had already begun making publicly. At a Commonwealth Bank of Australia event in May, he said enterprise adoption remained early despite rapid technical progress and acknowledged that he had expected more entry-level white-collar jobs to disappear by then. He said he was pleased that forecast had proved too aggressive, according to the bank's account of the discussion.
In July, Altman also said frontier labs might eventually need to pace AI development long enough for society to adapt to new capability levels. That argument focused on safety and institutional preparedness. His latest comments apply the same timing gap to commercial adoption: powerful systems can exist well before companies are willing or able to reorganize around them.
OpenAI's long-horizon pitch
That distinction carries financial consequences for OpenAI. In March, OpenAI said it had closed $122 billion in committed capital at an $852 billion post-money valuation, figures supplied by OpenAI. The financing rests on a long-term bet that model capability will eventually become core infrastructure across consumer products, software development and enterprise work.
Slower adoption does not necessarily break that thesis, though it stretches the period between infrastructure spending and broad economic payoff. It also shifts pressure toward deployment: integrations, security controls, workflow redesign and products that can deliver reliable results inside existing organizations.
For startups, Altman's revision narrows the near-term opening. A model advance alone is less likely to erase established software vendors overnight when customers remain tied to existing data, processes and contracts. New entrants still have room to compete, but they must solve adoption friction alongside the technical problem. Altman is effectively moving the bottleneck downstream. The models may arrive quickly. Rebuilding companies around them will take longer, and the software businesses once presumed to be immediately "up for grabs" have been given more time to adapt.