The question this session opens with is narrower than most Build sessions and more useful for it: you have built an agent with open-source tooling, so how does it reach production without being rewritten?
The answer they demonstrate is worth extracting from the product context, because the design principle underneath it applies regardless of platform.
Keeping the agent as ordinary code
The property they emphasise is that the tool boundary stays inspectable and the agent remains normal code (4:02).
This sounds unremarkable and is not. A great deal of agent framework design pulls the loop inside a runtime you configure rather than write — you declare tools, describe behaviour, and the framework decides what happens between steps. It works until something goes wrong, at which point the thing you need to inspect is exactly the part you did not write.
Their framing of the loop is deliberately plain: it asks the model for the next step, runs tools when needed, and uses each result to decide what follows (1:08). Kept as code, that loop is debuggable with ordinary tools. Wrapped in a runtime, it is debuggable only through whatever the runtime chooses to expose.
The token argument for local tools
The technical observation most likely to transfer concerns where tool definitions live.
Connecting to a remote tool server means retrieving its instructions — often large structured definitions — into context before anything is called. Defining the tool in the agent's own code avoids that round trip and the context it consumes (8:34).
The trade-off is real and worth stating in both directions. Remote tool servers give you a shared, versioned, independently maintained definition that many agents can use, which is exactly what you want across an organisation. Local definitions are cheaper per invocation and cost you that sharing.
Which means the correct answer is architectural rather than universal: tools used by one agent belong close to it, and tools used by many belong behind a shared interface. The failure mode is treating either as the default and paying the other's costs everywhere.
What packaging is actually solving
The problem statement — reaching production without a rewrite — is the honest description of where most agent projects lose time.
A prototype built with open-source tooling runs on a developer's machine with their credentials, their environment and their patience for restarting it. Production requires identity, isolation, observability, deployment and a way to update it, none of which the prototype needed. The usual outcome is that the prototype is treated as a specification and rebuilt on whatever the platform supports.
What is being offered here is that the packaging changes and the agent does not. Whether that holds in practice is the question a sixteen-minute session cannot answer. But it is the right thing to be attempting, and it is a more useful target than another orchestration abstraction.
演讲章节
关键要点
- 01
The design property they emphasise is that the tool boundary stays inspectable and the agent remains ordinary code rather than runtime configuration. 4:02
- 02
Their description of the loop is deliberately plain: ask the model for the next step, run tools when needed, use each result to decide what follows. 1:08
- 03
Remote tool servers cost context before anything is called, since their instructions must be retrieved; local definitions avoid that and give up sharing. 8:34
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.