We Need An Ecosystem in AI, And Every Company Can Win A Place In It

2026-06-04 - 42 min - source - Read full transcript

Sarah Guo (host)Elad Gil (host)SwyxSatya Nadella

Key insights

Microsoft is positioning its AI strategy as an ecosystem play rather than a bet on one model or platform, so any company can become a first-class AI participant.

Nadella says the biggest takeaway from Build is to conceptualize this moment as an ecosystem rather than a single-model race. Microsoft's job, in his framing, is to define the recipe, stack, and tooling that let AI-native startups and traditional enterprises alike build and own their own intelligence on top of Microsoft's platform, drawing on his experience of four prior platform shifts at the company.

ai-ecosystem-strategy

A platform is defined by its ability to create more value for the ecosystem around it than it captures for itself.

Nadella states this explicitly as his personal test for a successful platform layer, contrasting it with a strategy that just captures value inside a single model or app. He ties Microsoft's developer-conference purpose to this: without the ability for others to build and extend on top of the platform, there is no reason to hold a developer conference at all.

ai-ecosystem-strategy

Private evals, not the underlying model, are becoming the most defensible IP a company can build.

Nadella describes an acid test: if a company has a private eval and can swap the underlying model from provider A to provider B while still improving against that eval, it is in control of its own stack; if it can't, it isn't. He predicts every company's private eval will become one of its biggest sources of proprietary value, more durable than any specific model choice.

agentic-harnesses-and-evals

The agent 'harness' - the combined loop of models, data, and tools - matters more for real-world performance than raw model capability alone.

Microsoft is standardizing an open, GitHub-based harness with progressive tool disclosure and heavy context preparation across products like GitHub Copilot, Security Copilot, and the M-dash security tool. Nadella cites M-dash finding vulnerabilities that a narrower tool (referred to as Mythos) missed as evidence that harness quality can outperform a model used in isolation.

agentic-harnesses-and-evals

SaaS is being unbundled by agents, but the durable parts (data models and business logic) should survive even as the UI layer is reinvented.

Nadella argues the entity schemas underneath enterprise software (a general ledger, or the semantic model beneath a Power BI dashboard) are genuinely robust and shouldn't be reinvented from scratch. What changes in the agent era is the UI and workflow layer on top, meaning the challenge for SaaS vendors is to unbundle and rebundle their offerings, not assume customers will rebuild everything internally.

ai-business-models

AI pricing is converging on hybrid models, not a single dominant scheme, because outcome-based pricing breaks down once customers actually get the outcome.

Nadella expects per-user subscriptions to persist because buyers need budget certainty, layered with usage-based consumption pricing. He says customers who initially love outcome-based pricing tend to resent it once they see how much of the upside they're sharing away, and points to GitHub Copilot's recent shift from pure per-user pricing to per-user plus a consumption meter as evidence hybrid models are winning in practice.

ai-business-models

Company-specific traces of humans and agents working together could become a bookable asset, turning previously uncapturable tacit knowledge into IP.

Nadella suggests training a 'company veteran agent' on the traces between human judgment and agent execution captures institutional knowledge that was never possible to put on a balance sheet because it lived only in people's heads. He frames this as a candidate for future accounting standards around token-based or agent-based expertise.

future-of-work

Generalist leverage, not narrow specialization, is the biggest career upside in the agent era, illustrated by LinkedIn's 'full-stack builder' restructuring.

Nadella cites coding agents turning knowledge workers into people who can produce a working app 'in the same sentence' as a document or spreadsheet. He points to LinkedIn merging design, product management, and front-end engineering into a single full-stack builder role (while keeping specialist edge) as a preview of how most roles will gain leverage, alongside a smaller, more critical need for deep infrastructure specialists (e.g., building reinforcement-learning environments).

future-of-work

True organizational ambition means reconceiving what the team's job fundamentally is, not just making the existing job easier.

Nadella's example: an Azure networking team that manages hundreds of fiber operators stopped thinking of its job as doing network operations and rebuilt it as an agentic system (nicknamed Miles) that performs the operations, then began asking for more tokens instead of more headcount. He contrasts this 'meta work' reframing with merely automating existing tasks, and calls it the model for how enterprise value gets created going forward.

future-of-work

The AI industry will only earn permission to keep scaling infrastructure if benefits are provably tangible at the community level.

Nadella says the public will be skeptical of any tech company that just says 'trust us, the future will be glorious,' and that this time the stakes (energy prices, jobs, tax base, health outcomes) are too large a share of the economy for that framing to work. He argues communities are right to be skeptical and demand evidence, and that without demonstrated local benefit, the industry won't have the social license to continue the current data-center buildout pace.

ai-societal-impact

Education is the AI societal-benefit category with the least visible impact so far, and Nadella attributes this to outdated credentialing and incentive structures rather than pedagogy alone.

Compared to wealth creation and health care (where he cites companies like Open Evidence), Nadella says education hasn't shown comparable AI-driven impact yet. He points to Alpha School's founders rethinking pedagogy and floats that the next major startup success story could be a company that builds a new university or curriculum model suited to how people now access information and build skills.

ai-societal-impact

Media referenced

Essay on organizational ambition in the age of AI - article - Elad Gil mentions his partner Mike Rynal (who started his career at Microsoft) wrote an essay arguing this is an age where organizations can and need to be much more ambitious, given the pace at which users and companies are adopting new technology.

Companies

Microsoft - Nadella's company; the episode is a crossover recorded at Microsoft Build covering Azure, MAI models, GitHub Copilot, M365, and Work IQ.
Azure - Cited as the example of scale: Microsoft built more Azure capacity in the last 15 months than in the first 15 years, and the Azure networking team rebuilt its job as an agentic system (Miles) rather than manual operations.
GitHub - GitHub Copilot recently moved from pure per-user pricing to per-user plus a consumption meter; the GitHub harness is used across Microsoft's other agentic products.
OpenAI - Nadella traces his AI conviction back to the OpenAI partnership and the scaling-laws paper; MAI models were trained in part using traces from models like GPT-5.5.
LinkedIn - Cited as an example of structurally rebuilding engineering roles into a 'full-stack builder' discipline that merges design, product, and front-end.
NVIDIA - Referenced via Jensen Huang and the CUDA platform as an example of a platform layer that others (like Microsoft's DX) built on top of.
Power BI - Cited as an example of durable business logic (its underlying semantic model) that should survive the agentic unbundling of SaaS.
Alpha School - Nadella mentions meeting its founders to learn how they are rethinking pedagogy for an AI-native education model.
Open Evidence - Cited by Elad Gil as an example of AI already creating visible healthcare benefit.
Anthropic - Referenced (as 'Claude', partially garbled in the auto-captions as 'claws') in the context of Microsoft's autopilot demos being built with models beyond its own MAI lineup.

Techniques and frameworks

Ecosystem platform strategy - Nadella's framing that a platform's job is to create more value for participants than it captures for itself, letting any company - AI-native or traditional - become a first-class builder on top of it.
Private evals as IP - The idea that a company's proprietary evaluation set, not its model choice, is becoming its most defensible asset: if you can swap the underlying model and still hill-climb on your private eval, you're in control of your stack.
Agent harness (models + data + tools loop) - Microsoft standardizing an open, GitHub-based harness with progressive tool disclosure and rich context prep across Copilot, Security Copilot, M-dash, and science-discovery products.
Hill-climbing scaffold - Nadella's description of how MAI models are built: start from a clean pre-training lineage, then let companies build their own specialist evals and reinforcement-learning traces around the base model.
Full-stack builder role - LinkedIn's restructuring of design, product, and front-end engineering into one combined role with an area of specialist edge, cited as a model for how generalist leverage increases in the agent era.

Summary

This is a crossover episode recorded live at Microsoft Build, with No Priors hosts Sarah Guo and Elad Gil joined by Swyx (Latent Space) interviewing Microsoft CEO Satya Nadella. The conversation centers on Nadella's framing of the current AI moment as an "ecosystem play" rather than a race to own a single model or platform: Microsoft's stated job is to build the recipe, stack, and tooling that let any company, whether AI-native or a traditional enterprise, become a first-class participant that owns its own intelligence layer. He repeatedly returns to a test for what makes a platform real: it must create more value for the ecosystem around it than it captures for itself, the same dynamic he says made Windows, and later NVIDIA's CUDA, durable platforms.

A large section covers the technical and business anatomy of Microsoft's approach, including the MAI model lineage, the concept of an open "harness" (models, data, and tools working together with rich context and progressive tool disclosure) that Microsoft is standardizing across GitHub Copilot, Security Copilot, and other products, and the emerging idea that a company's private evaluation set, not its choice of underlying model, is becoming its most defensible IP. Nadella extends this into a broader theory of enterprise software: SaaS is being unbundled by agents, but the durable core (data schemas, business logic like a Power BI semantic model) should persist even as the UI and workflow layer around it gets rebuilt, meaning the winners will be vendors flexible enough to rebundle rather than either side assuming total replacement.

On pricing and business models, Nadella predicts a hybrid future rather than one dominant scheme: per-user subscriptions persist because buyers need budget certainty, layered with usage-based consumption, while outcome-based pricing tends to collapse once customers actually experience how much upside they're sharing away. He cites GitHub Copilot's own recent shift from pure per-user pricing to per-user-plus-consumption as a live example.

The conversation turns philosophical in its back half, with Nadella arguing that true organizational ambition means reconceiving what a team's job fundamentally is, illustrated by an Azure networking team that rebuilt its own operations as an agentic system and began asking for tokens instead of headcount. He frames the biggest career upside in this era as generalist leverage, citing LinkedIn's restructuring into "full-stack builder" roles, while carving out continued demand for deep infrastructure specialists. He also floats that company-specific traces between human judgment and agent execution could become a genuinely bookable balance-sheet asset, capturing institutional knowledge that was previously invisible to accounting.

The episode closes on societal stakes: Nadella argues the AI industry will only earn permission to keep scaling data-center infrastructure if the benefits (energy costs, jobs, tax base, health outcomes) are made tangible and provable at the community level, not simply promised. Asked where AI has underdelivered on visible societal benefit so far, he singles out education, arguing the gap is less about pedagogy and more about outdated credentialing and incentive structures, and suggests the next major startup success story could be a company that rebuilds the university or curriculum model itself.

Notable Quotes

"A platform is defined by fundamentally its ability to create more value about the platform versus what's captured in the platform." - Satya Nadella

"You have an eval that's private. You're using model A. Can you switch it to model B and climb up? If you can, then you're in control. If you can't, you're not in control." - Satya Nadella

"Our job is not to do Azure networking. Our job is to build the agentic system that does Azure networking." - Satya Nadella

"The world is going to be way skeptical of tech and tech companies that say, 'Trust us, we've got it, the future is going to be glorious.' You kind of have to deliver tangible benefits." - Satya Nadella

"True ambition is about making the impossible possible." - Satya Nadella