OpenAI's CEO Says His Own 2023 Forecast Was Wrong

The Admission

Sam Altman admitted on air that his own 2023 forecast was wrong. On August 23, 2026, he appeared on David Senra's Founders podcast and said: "I think I was wrong about a few things, but in terms of the speed, one of them is the economy just has so much inertia." He named the specific miscalculation directly: after GPT-4 shipped in 2023, he expected "much more disruption, software businesses up for grabs right away, than it turned out to be."

He broadened the admission beyond himself, saying "we've all been too ambitious on timelines," and argued that even a technology as capable as AI diffuses through the economy far more slowly than technologists assume, because, in his words, "people keep doing the same things they're doing, they keep buying from the same company, they keep wanting to use their tools in the same way."

What Actually Changed, And What Didn't

Altman's capability forecast did not move. He said he still expects AGI-level capability by the end of 2028, and described a future where "most of humanity's intellectual capacity" could reside inside data centers. What moved is his forecast for adoption speed, not for what the models themselves would eventually be able to do.

He offered his own diagnosis for the gap: the technological pieces for transformative AI use already exist, but the product experience that would make adoption obvious and immediate has not arrived, what he called AI's missing "iPhone moment", and he called that "mostly a product failure," not a capability shortfall.

The Decisions That Leaned On His Old Timeline

Real decisions leaned on a timeline that has now been walked back. Between 2023 and 2025, organizations made budget plans that assumed near-term headcount reductions from AI tools, signed vendor contracts under an act-now-or-fall-behind framing, and built internal reorganizations around the expectation that specific software roles would shortly disappear, all on the premise that disruption was imminent.

Those decisions were not irrational given the information available at the time. Altman was the CEO of the company that built the models being discussed, about as close to inside knowledge as a forecast source gets. That is exactly what makes his admission instructive rather than obvious: the best-positioned forecaster was still wrong about speed.

Why A Vendor's Timeline Is Not A Base Rate

A vendor's roadmap reflects what it sells, not a neutral forecast. A company selling a technology has a structural incentive to forecast fast adoption, because urgency drives purchasing decisions, valuation, and deal-closing, independent of what its own leadership privately expects about actual diffusion speed. That does not make the forecast dishonest; it makes the forecaster, by Altman's own account here, a participant in the same overconfidence he is now naming.

The corrective habit is to separate two different questions that get bundled into one roadmap slide: how good will the technology get, where the lab genuinely has the closest view, and how fast will your own organization's routines actually change, where the better guide is the base rate of organizational change generally, which moves on multi-year cycles even for far less disruptive technologies than this one.

What To Do With This On Your Next Budget Cycle

Test the timeline, not just the capability, before you budget against it. When a vendor's roadmap deck cites a disruption timeline as the justification for an urgent purchase or a reorganization, ask what has to be true about organizational inertia, not just about model capability, for that timeline to hold, and weight the vendor's historical accuracy on that specific kind of claim, not only their accuracy on capability, before building a budget around it.

Altman's walkback does not mean AI capability is disappointing; his own AGI timeline is unchanged. It means the gap between "the technology can do this" and "your organization will actually do this soon" was wider than even its builder assumed, and that gap is where a realistic budget should live.

Read next: OpenAI's Own Report: Reward Hacking, Not Malice  |  Why OpenAI's New Chip Won't Replace Nvidia