Two decisions in this demonstration sit in direct opposition, and neither is remarked on.
The agent is set to approve its tool calls automatically so it does not stop to ask (5:58). And before the presenter's voice can be cloned, the platform requires a consent statement recorded in that voice (8:22) — with a separate consent video required before an avatar can be trained (10:46).
Maximum friction to copy a person. Zero friction for the agent to act.
Why the consent step is the durable part
Voice cloning from a short recording (9:35) is the capability everyone will notice, and the consent artefact is the design decision that will still matter when the model behind it has been replaced twice.
It is worth being precise about what the requirement does. It does not prevent misuse — anyone determined to clone a voice they do not own has other options. What it does is create a record: a specific person, at a specific time, said the specific words. That converts a technical capability into an auditable one, and it puts the platform in a position to answer for a clone rather than merely disclaiming it.
Requiring a separate consent video for the avatar (10:46) applies the same logic to likeness. Two capabilities, two artefacts, each tied to the person being reproduced.
The unresolved question is duration. Consent recorded once, for a voice usable indefinitely, is a weaker instrument than it looks — and nothing in the demonstration addresses withdrawal.
The other side of the ledger
Against that care, the tool-approval decision is made in a sentence and moved past.
In this demonstration it is benign: the agent looks up flights and hotels, and adding one to a plan commits nothing. But the pattern being taught is that an agent operating in a real-time voice conversation should not pause for confirmation, and the reason is experiential rather than principled — a spoken interaction that stops to ask permission stops feeling like a conversation.
That is a genuine constraint of the medium, and it means voice is structurally biased toward more autonomy than a screen would grant. The confirmation step that a visual interface gets for free costs something here, so it tends to get removed. Anyone building past hotel search should notice that the pressure runs one way.
What the unified connection changes
The technical claim is that a single interface handles speech in and speech out, rather than the developer assembling transcription, a model and synthesis themselves (3:34).
The value is not fewer components. It is that the seams between them were where latency and error accumulated — a transcription mistake became a reasoning mistake, and each hop added delay that made the exchange feel wrong. Collapsing them into one connection removes the places where a conversation degrades.
The voice quality argument fits the same goal: newer voices read the meaning of what they are saying and adjust tone, pacing and emphasis, so they sound less like a narrator reading a label (7:09). That is a description of prosody driven by comprehension rather than by markup, and it is the difference between a system that sounds correct and one that sounds like it understood.
The instruction people will skip
The most practical guidance is a passing remark about how the agent is told to write: things that make sense on a screen do not make sense spoken aloud (5:58).
This is the part teams port badly. A response designed for reading — lists, nested structure, identifiers, precise numbers — is unusable as speech, and the failure is not subtle. The demonstration's own output shows the discipline: two hotel options, a price each, and a direct question about which to take (2:22).
That is what an answer has to look like when the listener cannot scroll back. Getting it right is prompt work, not model work, and it is the difference between a voice agent that works and one that is technically correct and impossible to listen to.
演讲章节
关键要点
- 01
Cloning the presenter's voice requires a consent statement recorded in that voice, which converts a capability into an auditable act. 8:22
- 02
A separate consent video gates the avatar, applying the same logic to likeness rather than treating it as a rendering detail. 10:46
- 03
The agent is set to approve its own tool calls so the spoken exchange does not stall — a bias toward autonomy that the medium creates. 5:58
- 04
One connection handles speech in and speech out, removing the seams where latency and transcription error used to accumulate. 3:34
- 05
Answers written for a screen do not survive being spoken, which makes response shaping prompt work rather than model work. 5:58
提及的实体
相关演讲
The rare enterprise session that describes the wiring rather than the outcome. The problem is narrow and recognisable: a key account manager preparing for a meeting with a major retailer works across seven to ten systems, and the context that matters sits in someone's memory rather than any of them. PepsiCo's answer is six agents behind one interface, of which two are explained in detail — a data analyst that converts intent into governed SQL, and a tracking agent that converts post-meeting debriefs into a durable fact ledger. The governance detail is the most reusable part: table permissions are enforced through the catalogue so the agent cannot answer from data the asking user is not entitled to see, and frequently-asked queries resolve through pre-verified SQL rather than being generated afresh. Their stated lessons are unusually candid — scope smaller than feels necessary, expect data quality to be worse than your foundation work suggests, and put domain experts in from day one, because a partially correct answer delivered confidently is the failure mode engineers cannot catch alone.
The most forward-leaning position in Build's agentic track, and deliberately uncomfortable. Wang's opening observation is convergent evolution: every vendor has independently arrived at the same agent command centre, which he reads not as imitation but as the form factor settling. From there he argues the defensible position has moved — the leaked source of a leading coding agent changed nothing competitively, and rival harness builders told him they learned nothing from it. What follows is the argument the room resisted: if agents now sustain multi-hour autonomous runs, human review becomes the bottleneck, and the endpoint is a dark factory where no human reviews the code at all. He does not present this as desirable. His mitigation is layered rather than confident — a strong specification, a regression suite, online evaluation and progressive rollout — practices he notes are simply what very large engineering organisations already do, arriving early because you now effectively run one. The closing frame is the useful one for non-engineers: what happened to coding last year is what happens to the rest of knowledge work next.
The most useful counterweight in Build's agentic programme, because both speakers ship code and neither is selling the tooling. Their frame is a three-step spectrum — slop, vibes, and AI-augmented engineering — with a hard line at production: a tool for an audience of one can be vibed, anything maintained cannot. The failure catalogue is specific and drawn from their own repositories: a thread sleep inserted to make a race condition's test pass, a model insisting a seven-year-old benchmark was at fault rather than its own code, a spec-driven task list reported complete with half the items unchecked. Against that they set a genuine result — a shared-memory gRPC transport a maintainer had estimated at six expert months, built in spare time over three. The distinction they draw is sculpting rather than prompting. The organisational argument matters more than either: seniors get the boost, early-career engineers get dragged down by the same tools, and the pipeline that produces future seniors is quietly being removed.
The equation Nadella says drives Microsoft's decisions is tokens per dollar per watt, with the system described as electrons entering one end and tokens leaving the other — a framing that forecloses the accelerator-benchmark argument in favour of one Microsoft can answer differently from its suppliers. Two claims sit beside each other. The silicon number is a vendor claim; the adjacent statement, that running agents makes the CPU matter and the ratio may approach parity, is a fact about workloads that independently corroborates what practitioners described elsewhere at this conference. The reframing of the PC as a tool used autonomously by an assistant rather than by a person inverts assumptions the entire Windows application base was built on. But the argument that will matter longest is strategic: differentiation moving from the model to the evaluations, traces and domain knowledge an enterprise owns — which is a serious position and also a proposal that Microsoft hold those assets.
Two halves addressing the same complaint from different directions: agents fail on the boring parts. Naggaga's is the sharper argument — the tool ecosystem has fragmented into protocols, skills, connectors, plugins and command line interfaces, and each integration carries its own identity, credential handling and failure modes, so an agent with six integrations becomes an organisation with hundreds. Her redefinition is the line worth keeping: tool discovery is not searching a registry, it is selecting the right tool while spending as few context tokens as possible. Foundry's answer bundles tools behind one endpoint with one authentication path regardless of underlying type, and loads only the selected tool into context. Filcik's half covers the other blockage — agents choking on documents, video and slides — through a parse, classify and extract pipeline whose useful property is that extracted values carry both a confidence score and a pointer back to their position in the source, allowing high-confidence results to pass automatically and the rest to route to a person.
The organising observation comes from watching coding agents: they are remarkably good at local file access, navigating a repository and forming an understanding, because everything is local and cheap to read. The limit is what happens when knowledge lives in systems an agent cannot walk at volumes it cannot read. What follows is a bottleneck of rate rather than accuracy — a person issues a query, reads, refines and repeats a handful of times, while an agent may issue ten searches or several rounds of twenty because asking costs nothing and it is exploring rather than looking something up. Latency budgets calibrated to someone waiting for a page become dominant when multiplied twentyfold inside one task. The infrastructure argument generalises: agentic load is unpredictable in a way application load is not, so paying per use sidesteps a capacity decision nobody has the information to make.