The framing that opens this session is worth separating from the product it introduces: agents are no longer static routers shuffling requests between a fixed set of tools (1:04).
They acquire skills. They generate memory. They accomplish things nobody programmed them to do. The speakers call them general-purpose systems (1:22), and the operational consequences of that sentence are what the remaining forty minutes are about.
What changes when the system stops being fixed
A router can be tested exhaustively, because its behaviour is bounded by its configuration — enumerate the routes, verify each, ship.
A system that acquires capability during operation cannot be tested that way. What it will do next week depends on what it accumulated this week. Traditional acceptance testing establishes that it worked once, under conditions that no longer apply.
The response described is continuous: every action and every cost signal feeds an evaluation service that keeps learning (3:45). This is the correct shape of answer, and it is also a considerable operational commitment. You are not deploying a system and monitoring it; you are running an evaluation apparatus permanently alongside it, at a cost proportional to the thing being evaluated.
Proactive rather than reactive
The design shift most likely to matter arrives quietly. Rather than waiting for a request, agents can be configured against events that wake them for particular tasks (21:37).
This is a larger change than the demonstration suggests. A request-driven agent has natural boundaries — a person asked, so someone is accountable and the work is scoped by the asking. An event-driven agent operating continuously has neither. Its cost is unbounded by anyone's attention, its actions are not attributable to a specific request, and the question of who authorised any given piece of work becomes genuinely hard to answer.
Every operational discipline around this — budgets, audit, incident response — currently assumes a requester. None of it survives contact with a system that wakes itself.
The measurement that deserves scrutiny
Late in the session comes a claim about waiting: an evaluator producing eleven per cent better performance in exchange for the time spent letting it finish (37:18).
Framed that way it is an easy trade, and it deserves more examination than it receives. Eleven per cent of what, measured how, across how many runs? The number invites exactly the comparison that the observability sessions at this same conference demonstrate you cannot make from a handful of samples.
What survives is the underlying idea, which is sound: latency and quality are exchangeable in agent systems, and the exchange rate is measurable. Publishing a specific number without its denominator is how a measurable thing becomes a marketing claim.
The honest core
Strip out the tooling and one argument remains, and it is correct.
Agents that acquire capability cannot be validated once. They require permanent instrumentation, continuous evaluation, and an operational model where the question is never "did it work" but "what is its current success rate." That is a substantially heavier commitment than deploying software, and organisations planning agent programmes on the assumption that testing ends at launch are budgeting for something else.
关键数据
- 11%
- claimed evaluator performance gain in exchange for waiting, stated without a denominator 37:18
演讲章节
关键要点
- 01
Their framing is that agents are no longer static routers between fixed tools — they acquire skills, generate memory and do things nobody programmed. 1:04
- 02
Because capability accumulates during operation, acceptance testing establishes only that something worked under conditions that no longer hold. 1:22
- 03
Every action and cost signal feeds a continuously learning evaluation service — the right answer, and a permanent operational commitment. 3:45
- 04
Agents can be woken by configured events rather than requests, which removes the boundaries a requester provides on cost, attribution and authorisation. 21:37
- 05
A claimed eleven per cent performance gain from waiting for an evaluator is offered without a denominator, inviting exactly the comparison the observability sessions show cannot be made from few samples. 37:18
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 decisions in this demonstration sit in direct opposition and neither is remarked on: the agent approves its own tool calls so it does not stop to ask, while cloning the presenter's voice requires a consent statement recorded in that voice and cloning their likeness requires a separate consent video. Maximum friction to copy a person, zero friction for the agent to act. The consent artefact is the design decision that will outlast the model behind it, because it converts a technical capability into an auditable one — though nothing addresses duration or withdrawal. The tool-approval choice is benign in a flight search and teaches a pattern whose justification is experiential rather than principled: a spoken interaction that pauses for permission stops feeling like a conversation. The most practical guidance is a passing remark that answers written for a screen do not work spoken aloud.
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.