The recommendation buried at the end of this nine-minute segment is the most immediately usable advice from Build: use the larger model for planning, and a cheaper automatic selection for implementation (8:11).
The reasoning is economic and specific. Having analysed the cost of their own conversations (7:32), the team found the expensive model earns its price during planning — where a wrong decision propagates through everything after it — and stops earning it during implementation, where the work is largely mechanical once the plan is settled.
Most guidance about model selection is per-task. This is per-phase within one task, which is a finer distinction and produces a much larger saving.
The problem they admit they have not solved
The candid moment is worth more than the feature.
Context switching between agents, they say, is an unsolved problem — visible in user testing and in their own experience, and it burns you out (5:19).
This is the cost nobody prices when demonstrating parallel agents. Running six agents concurrently sounds like six times the throughput. What it produces is six streams of work in different states, each requiring you to reconstruct where it was, what it decided and whether it is on track before you can usefully intervene. The reconstruction is the expensive part, and it happens every time you switch.
Human working memory does not multiply. An interface that dispatches many agents easily has solved the cheap half of the problem and made the expensive half worse — which is precisely what the VS Code team is describing about their own users, and their own working days.
Why the admission matters
An agents window that collects concurrent sessions in one place is a reasonable response, and the team is clear that it is a partial one.
The unsolved part is not display. It is that supervising parallel work requires holding parallel context, and no arrangement of panels changes what a person can hold. The eventual answer probably looks less like a better dashboard and more like agents that summarise their own state well enough that reconstruction is unnecessary — which is a model capability rather than an interface feature.
Until then, the honest position is the one they take: here is a place to see them all, and switching between them still costs you something no tool currently removes.
演讲章节
关键要点
- 01
Their recommendation from analysing their own conversation costs: the larger model for planning, a cheaper automatic selection for implementation. 8:11
- 02
The saving comes from selecting per phase within a task rather than per task, since the expensive model stops earning its price once the plan is settled. 7:32
- 03
They name context switching between agents as an unsolved problem, visible in user testing and in their own working days. 5:19
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.