This is a Davos-style interview focused on AI capability progress, timelines to AGI, economic displacement, and geopolitical/safety risks. Dario Amodei and Demis Hassabis both argue the technology is advancing faster than many expect, but they differ on how soon the self-improvement loop closes and how much time society has to adapt.
Watch on YouTube ›Get the market thesis, key claims, assets, contradictions, and follow-up questions from any financial video — then unlock a version personalized to your portfolio, watchlist, and favorite speakers.
The conversation centers on whether AI is approaching a phase where it can materially improve itself, and what that implies for jobs, geopolitics, and safety. Dario Amodei argues that the key near-term breakthrough is code: models already write much of the code, and he thinks AI may do most or all of software engineering end to end within roughly 6 to 12 months. Demis Hassabis is more cautious on timing, but still says he maintains a similar overall AGI timeline, while noting that some domains like coding and math are easier to automate than natural science because outputs are verifiable. Both speakers treat the “day after AGI” framing as premature; they want first to understand how quickly the loop of AI building better AI can actually close. A major thread is the self-reinforcing research loop. …
Immediate setup is dominated by accelerating AI capability in coding and the first signs of hiring pressure at the junior end. The tactical risk is that deployment and competition outrun safety policy, while the main upside catalyst is evidence of AI materially producing more of the software stack.
Over the next few quarters, expect the narrative to pivot from AI assistance to partial replacement in software and some office work, with hiring patterns as the clearest confirmation. The setup strengthens if AI systems begin improving the next generation of models and weakens if self-improvement stalls on hard-to-verify tasks.
The long-run implication is that AI may become a foundational scientific and industrial infrastructure, with chips and compute treated as strategic assets. If that regime takes hold, the lasting question becomes how wealth, labor, and meaning are reorganized in a post-scarcity environment.
Dario believes AGI-like models capable of doing the full work of software engineers could arrive in about 6 to 12 months.
He says coding models are already letting engineers stop writing code themselves and that the remaining step to full end-to-end software engineering may be only months away.
How quickly AI systems can build other AI systems will determine whether the world has only a few years before a major breakthrough or faces immediate urgency.
The speaker says this development is the most important thing to watch and that its evolution will determine the timeline and urgency of what comes next.
Dario thinks the full self-improving loop is not yet solved and may be limited by hard-to-verify domains and embodied AI.
He says the loop may require AGI itself in some areas and could be slowed by messy domains, physical AI, and robotics hardware in the loop.
Do you still think a Nobel-level model by 2026 or 2027 is the right timeline?
Dario says he still thinks it will happen not too far off, though exact timing is hard. He explains the main driver as a feedback loop where better coding and AI-research models speed the next generation, but notes chips, manufacturing, and training time add uncertainty.
How has your prediction changed over the last year about having systems with full human cognitive abilities by the end of the decade?
Demis says he is still on the same timeline. He thinks coding and math are easier to automate because results are verifiable, while natural science is harder because outputs may need experimental validation and the system may still lack some ingredients for higher-level scientific creativity.
What surprised you most about Google DeepMind's progress this year?
Demis says he expected DeepMind to return to the top because of its deep research pool, but it took work to restore intensity and startup-like focus. He points to progress in Gemini 3 and in the Gemini app, including market share gains, while saying there is still a lot more to do.
Unlock the full claims, asset map, scores, related transcripts, follow-up questions, and AI chat — shaped around your portfolio, watchlist, favorite speakers, and risks.
