The most interesting number in this session is not about the database. It is that around 8 per cent of the lines of code modified in this release were changed by AI (9:05).
The speaker offers it as a useful measure, and the framing deserves attention. It is not a productivity claim. It is a disclosure about how a widely-used piece of infrastructure now gets built, from a team with no obvious incentive to volunteer it.
Eight per cent is also a more credible figure than the numbers usually quoted in this space. It is small enough to be believable and specific enough to be checkable against a repository, which is more than can be said for most adoption statistics presented on a conference stage.
The adoption curve they point to
The historical comparison offered is the more useful analytical content. When ChatGPT arrived, everyone learned what embeddings and vectors were, and adoption of embedding capability rose almost vertically (11:15).
That pattern is worth naming because it recurs. A capability exists in databases for years with modest uptake. An unrelated consumer product makes the underlying concept legible to a mass audience. Demand for the long-dormant feature appears suddenly, from people who were not asking for it because they did not know it was a thing they wanted.
For anyone building infrastructure, the lesson is uncomfortable: your feature's adoption may be gated on comprehension rather than on capability, and the event that unlocks it may have nothing to do with you.
What replaced the pipeline
The architectural argument is stated in one line and carries the session: rather than complex extraction and loading pipelines to move data somewhere it can be used, let the applications work against it where it lives (7:09).
The pipeline exists because analytical and transactional systems had incompatible requirements — one optimised for many small writes, the other for large scans — so data was copied between them on a schedule. Every organisation running that arrangement pays for the copy, the schedule, the lag and the divergence when it fails.
Removing the copy removes all four. Whether the underlying system can genuinely serve both workloads is the question every product making this claim has to answer, and a thirteen-minute session cannot. But the direction is right, and the reason it is newly plausible is agentic: an agent querying data does not fit either traditional profile, and building a third pipeline for it is not an appealing prospect.
The framing shift underneath
Their stated intent is not to ship finished experiences but to let organisations use application-building tools against the data directly (6:53).
That is a meaningful repositioning. A platform that ships packaged capability competes with what its customers can build. A platform that exposes primitives is betting the customers will now build things themselves, because generation cost has fallen enough to make it worthwhile.
It is the same bet running through this entire conference, arriving here in the least glamorous possible place — which is usually where you can tell whether a bet is real.
关键数据
- ~8%
- share of lines of code modified in this release that were changed by AI 9:05
演讲章节
关键要点
- 01
Around 8 per cent of the lines of code modified in this release were changed by AI, offered as a measure rather than a productivity claim. 9:05
- 02
Vector and embedding adoption rose almost vertically after an unrelated consumer product made the concept legible to a mass audience. 11:15
- 03
Their architectural argument is to let applications work against data where it lives rather than moving it through extraction and loading pipelines. 7:09
- 04
The stated intent is to expose primitives rather than ship packaged experiences, betting customers will now build for themselves. 6:53
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.