Yoann Saunier opens with a number and immediately declines to treat it as a problem: 90 per cent of AI-generated prototypes never become production applications, and he says that is for good reasons (1:29).
That is the right starting position, and it is rarer than it should be. Most sessions on this topic treat the gap as friction to be removed. This one treats it as a filter that is mostly working, and asks what should pass through it.
What actually fails
The failure he describes is specific and does not concern code quality (1:44). Security and data leaks from prototypes pushed public a little too early.
The mechanism is worth stating because it is structural rather than careless. Generating a working application is now fast. Everything that makes an application safe to expose — authentication, access control, secret handling, tenancy boundaries — is not part of what gets generated, because it is not part of what was asked for. The prototype works, in the sense that it does what the prompt described. It also has no idea who is allowed to use it.
An engineer building the same thing by hand would encounter those questions as unavoidable steps. Generation skips them, and their absence is invisible in a demonstration.
Which is why guardrails beat instructions
The alternative he describes is the useful part: the agent had guardrails, a workflow designed for agents, templates to start from, and platform primitives like authentication already present (27:21).
That last clause is the load-bearing one. Identity and access control are part of the application's foundation rather than something added later (20:40), which means the generated application inherits them instead of needing to be told about them.
The distinction matters more than it first appears. Telling an agent to handle authentication properly relies on it knowing what proper means in your environment and remembering across a long generation. Starting from a foundation where authentication already exists removes the requirement entirely — the agent cannot forget something it is not responsible for.
This generalises past security. Any property you need every generated application to have is more reliably supplied by the starting point than by the prompt. Prompts are advice; templates are structure, and only one of them survives a model that is having an off day.
What the ninety per cent should be
Read alongside his opening, the argument resolves cleanly.
The gap between prototype and production is not a tooling failure to be engineered away. It is the appropriate resistance of an environment that will not accept software lacking properties nobody asked the generator to provide.
The right response is not to lower the resistance. It is to move the properties into the starting conditions, so that what gets generated already has them — at which point some of the ninety per cent passes on merit rather than by exception.
关键数据
- 90%
- share of AI-generated prototypes that never become production applications 1:29
演讲章节
关键要点
- 01
He opens with 90 per cent of generated prototypes never reaching production and declines to treat it as a problem, calling the resistance justified. 1:29
- 02
The failure is security and data exposure from prototypes made public early, not code quality. 1:44
- 03
Identity and access control belong in the application foundation rather than being added later, so generated applications inherit rather than request them. 20:40
- 04
Guardrails, an agent-shaped workflow and templates with primitives already present outperform instructing the agent to handle these concerns. 27:21
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.