AI-assisted analysis. See our editorial policy.
Human editorial review not recorded
The claim in this conversation most likely to matter within a few years is not about computation at all. It is about data.
The argument runs through a comparison. The measurements behind modern structural biology began accumulating in the 1970s, and it took roughly fifty years of painstaking work — many doctorates, considerable public funding — to assemble the database that made the protein-structure breakthrough possible (25:16). A quantum computer could produce valuable training sets in settings where collecting them experimentally is impractical, and this is described as the first place quantum contributes to AI: here is a dataset worth using (25:44).
Why that inverts the usual framing
Quantum computing is normally sold as an accelerator — the same problems, faster, or problems too large for classical machines. That framing invites an obvious rebuttal about how long the hardware will take to arrive.
Framing it as a data generator changes the shape of the argument entirely. It does not require a fully error-corrected machine, because a dataset does not need to be produced in real time or reliably on the first attempt. It needs to be produced once, correctly, and it then has permanent value. Their note that datasets relevant to materials science had appeared the week before (25:44) fits a production model rather than a computation model.
It also identifies a real bottleneck. The limiting factor in applying machine learning to physical science is very often that no adequate dataset exists and generating one requires decades of laboratory work. If simulation can substitute for some of that, the constraint moves from time to compute — and compute is a constraint that money and engineering address, while fifty years of measurement is not.
What is no longer the hard part
The technical status report is refreshingly specific: coherence times have improved roughly tenfold, from around twenty microseconds to more than two hundred and fifty, which is described as broadly sufficient for useful quantum computing. What holds things back now, in their account, is more mundane system-level engineering (11:46).
That is a meaningful transition to have reached and an easy one to misread. Moving from a physics problem to an engineering problem does not mean the remaining work is small — it means the remaining work is the kind that responds to effort and organisation rather than to insight.
They also note that the field's approach has broadened: work on neutral atoms progressed to the point where it looked capable of reaching certain applications earlier, so it was added rather than dismissed (38:50). A research programme that adds a competing approach to its own portfolio is behaving well; the alternative is defending the path already invested in.
The security timeline moving the wrong way
The most actionable statement is about cryptography. Better algorithms mean fewer resources are needed for both simulation and cryptanalysis, so the timelines are moving inward — and waiting until the late 2030s is not the right posture (31:36).
The scale of that movement is what makes it serious. In 2019 the field's estimate for breaking widely used public-key encryption was a system of at least twenty million qubits, and a result about a year before this conversation reduced that requirement substantially (32:02).
The number that shrank was not hardware progress. It was an algorithmic result — which is the harder kind of change to plan around, because hardware roadmaps can be tracked and extrapolated while a better algorithm arrives without warning and moves the date by years in a single step.
Anyone whose planning assumes a decade of margin should notice that the margin was reduced by mathematics rather than by engineering, and that nothing prevents that happening again.
Where the useful work already is
The through-line worth taking away is that the interesting results are arriving before the machine is finished.
They describe an algorithm demonstration as the first showing of something genuinely useful (19:51), used with academic collaborators to compute a molecular quantity and answer open questions in chemistry (22:07). They point at battery chemistry as a design problem simulation could address (14:28). They note the Nobel Prize awarded for the foundational superconductivity work, two of whose recipients worked on the team (3:11).
The pattern is that a moonshot organised around a distant goal is producing usable science along the way — which is both a genuine observation and exactly what one would say if the distant goal remained distant. Both readings are consistent with everything presented here, and the data-generation argument is the one that would still hold if the timeline slips.
Key numbers
Talk chapters
Key takeaways
- 01
The first contribution of quantum to AI is expected to be valuable training sets in domains where experimental data collection is impractical. 25:44
- 02
The dataset behind the protein-structure breakthrough took roughly fifty years of manual measurement, funding and doctoral work to assemble. 25:16
- 03
Coherence times improved roughly tenfold to over 250 microseconds, and the remaining obstacles are described as system-level engineering rather than physics. 11:46
- 04
Cryptographic timelines are moving inward because better algorithms cut the resource requirement, not because hardware arrived early. 31:36
- 05
The team added a competing qubit approach to its own portfolio when that path looked capable of reaching some applications sooner. 38:50
Entities mentioned
Organizations
Related talks
Four of Google's model, product and search leads on what changes once agents run for hours rather than seconds, and the most quotable argument comes from Dean: the constraint is moving out of the model and into the tools around it. By Amdahl's law, an agent spending half its time in tools built for human-speed interaction cannot gain more than a doubling however fast the model becomes — which reframes a great deal of current infrastructure work as latency debt. Their internal response is concrete: rewriting Python tooling into Go, framed as a fully specified translation task rather than an open prompt, produced order-of-magnitude speedups overnight. Reid supplies the counterweight from Search, where acceptable latency turns out to scale with how much work is being taken off the user rather than being a fixed budget. Woodward's detail is the quietest and perhaps the most telling: teams that have stopped writing product documents for humans and now write context files for models to act on directly.
The anecdote that opens the panel does the work: a robot asked to pick up the extinct animal selected a dinosaur toy, with nothing in its training data connecting the phrase to the object. That transfer from language models into machines with hands is the premise of the current wave. What the practitioners then describe is where it stops. Physical intelligence is about exerting force and using a body to do it, which is knowledge about consequences — the one thing a corpus of internet images contains almost nothing about. The humanoid question gets an honest treatment: not that human shape is optimal, but that the world is already built for it, plus a development-loop argument about collecting data and deploying on the same hardware. The most useful passage is scepticism about the field's favourite shortcut: generated video looks realistic and does not hold up for dexterous manipulation, because looking right and being physically consistent are different properties.
The observation that organises this session is not about capability but about attention. Engineers increasingly file a ticket rather than opening an editor, and the code comes back — which changes what the surrounding tools are for. If the agent works while you do something else, the conversation between you is no longer a workspace; it is the mechanism by which the agent surfaces a question it cannot resolve alone. Interfaces built for continuous conversation optimise for flow, and interfaces built for interruption should optimise for the opposite. A runtime constraint follows immediately: an agent that starts a long-running job cannot block until it finishes, which turns out to be a workflow-engine problem rather than a model one. The panel's closing formulation — that deciding what to build is the hard skill and always was — reads as reassurance and functions as a warning, since that judgement is downstream of exactly the work now being delegated.
The most honest moment here is an aside about how the presenters have tracked their own projects: plans in documents, plans in spreadsheets, plans in bug comments, and once a plan written on a receipt. That describes the actual category being addressed — not software nobody has built, but the small internal tool every team improvises badly because building it properly was never worth the effort. The demo turns on a single question: the generated app is strictly single player, so what happens when you want to share it with the team? That boundary is where improvised tools historically died, because it is where accounts, shared storage and access rules begin. Here it is crossed in one step, with the access rules generated and deployed automatically — which is convenient, and is also the moment the application acquires obligations nobody reviewed.
The clearest statement at I/O of how an agent API differs from a chat API, and the reasoning behind each departure is stated rather than assumed. Three changes matter. Conversation state moves to the server: a call returns an identifier, and passing it back continues the thread, retiring the client-side history array. The data model abandons alternating user and model turns for discrete steps, on the argument that a trace containing reasoning, tool calls, environment responses and compaction was never really a conversation and modelling it as one distorted it. And agents receive their own persistent remote environment rather than acting on the caller's machine — addressable by identifier, and shareable, so a research agent's output files become an application builder's input without passing through the context window. Schmid is explicit that scaffolded environment files are deliberately not model input, which is what keeps large artefacts out of the context budget. Schaeff's first half covers the real-time voice path, where the notable property is speech-to-speech across ninety languages with transcription of both directions.
Pichai's framing of Google as the buffer between people and the raw internet is offered as continuity — search did it, browsers did it, agents do it more — and it is also the most contested claim in the industry, because a buffer decides what passes through. He reaches immediately for the counterweight, the connection people feel to creators they follow, which is precisely the tension the company is currently managing without resolving. Two answers are sharper than the format usually produces. On competition he describes participants running on different pre-training and release cadences rather than at different speeds in one race, which is a more honest account than the leaderboard framing and comes from someone with an interest in leaderboards. On security he acknowledges models improving at cyber work, which is the one domain where better capability does not obviously net out positive, since an attacker needs one vulnerability and a defender needs all of them.