The featured question
AGI will arrive before the end of 2030.
The labs say it is close. The sceptics say the word barely means anything. The deadline settles it either way.
729 readers have gone on record. Vote to see where they stand.
What the experts said
Sam Altman
“We are now confident we know how to build AGI as we have traditionally understood it.”Altman says yes.
Elon Musk
“If you define AGI as smarter than the smartest human, I think it is probably next year.”Musk says yes — and has moved the date twice.
Demis Hassabis
“AGI will start to emerge in the next five or ten years.”Hassabis says probably not — he puts it just after 2030.
Roman Yampolskiy
“We would have artificial general intelligence by 2027.”Yampolskiy says yes — 2027, earlier than the labs themselves.
The verdict is in
Deadline passed · every verdict →“Probably in 2025 we will have an AI that can effectively be a mid-level engineer that can write code.”
Readers split 35 to 65 on this one. Nobody was right, because nothing could settle it.
Coding assistants became standard practice during 2025, but no system was shown operating autonomously at the level of a mid-level engineer, and Meta published no internal measure either way. The claim was made about an internal capability nobody outside the company can verify, which is what makes it unsettleable.
“We may see the first AI agents join the workforce in 2025 and materially change the output of companies.”
Readers split 34 to 66 on this one. Nobody was right, because nothing could settle it.
Agent products did ship across the industry during 2025. Whether they materially changed company output cannot be measured against any agreed benchmark, and none was offered when the claim was made. Recorded as unclear rather than forced into a stamp it does not fit.
“AI will be writing 90 percent of code within three to six months.”
55% of readers said no. Readers called it.
Six months on there was no evidence anywhere near 90 percent of code was AI-written, and studies over the same period found AI tools slowing experienced developers down rather than replacing them.
“My guess is that we will have AI that is smarter than any one human, probably around the end of 2025.”
63% of readers said no. Readers called it.
No system at the end of 2025 was accepted as smarter than the ablest humans across the board. Gary Marcus publicly offered a million-dollar bet against this claim at the time and it was never taken up.
Support the record
AI on Record is kept by a small group of volunteers and is free to read — no paywall, no ads. Every hour on it is unpaid.
Support the record →The verdicts, by email
One email when a prediction is settled — what was claimed, what happened, how readers voted.