Back

Satya Nadella on Microsoft's AI Harness Race

"every company, having private evals may be the biggest IP"

Watch the recap video here

Latent.SpaceLatent.SpacesourceNo PriorsNo PriorssourceSatya NadellaSatya NadellaspeakerswyxswyxspeakerSarah GuoSarah GuospeakerElad GilElad GilspeakerWatch full clip

Recap

Satya Nadella is explaining Microsoft's Build 2026 AI strategy to Latent Space and No Priors. Enterprise AI value moves into the harness around agents: the operating layer that gives agents company data, tools, context, evals, permissions, interfaces, traces, and human review.

Coding agents make the problem obvious. Once a developer or manager has many agents producing work, chat becomes too thin. The company needs a work surface for assigning, inspecting, approving, and improving agent work. Nadella extends that pattern to the whole enterprise: private evals can test whether agents help with company-specific work, traces can show how good work happens, and the harness can improve the system over time.

Microsoft wants Foundry, GitHub Copilot, Microsoft 365, Microsoft IQ, and Azure to supply that harness. Customers may own their private evals, traces, and context, but Microsoft is building the platform where those assets become useful.

  • 06:24-10:37 - Coding Agents, New UI, And The Enterprise Harness: Coding agents create the first clear UI break. Many agent sessions need review, approval, context, and richer work surfaces, which gives Nadella the enterprise harness model.
  • 02:31-06:24 - Clean Models, Private Evals, And Workflow Hill-Climbing: Companies can turn general models into specialists by improving them against private evals and traces from their own workflows.
  • 10:37-16:05 - Platform Strategy, Company IP, And Microsoft IQ: Private evals may become company IP, but control depends on whether companies can use those evals and traces across models inside an open enough harness.
  • 16:05-22:04 - SaaS Business Models And Pricing Pressure: Agents pull apart the old SaaS bundle of data model, business logic, interface, and seat pricing.
  • 22:04-28:55 - Build Versus Buy And Future Engineering Roles: Cheaper software generation changes the build-versus-buy calculation, but agent-made software still needs ownership, security, maintenance, and review.
  • 28:55-35:03 - Ambition, Infrastructure, And Datacenter Permission: The agent platform depends on datacenters, power, water, construction, and local permission.
  • 00:00-02:31 - Ecosystem Platform: Nadella opens with Microsoft's platform frame: companies should build their own AI capability and consume models through a broader ecosystem.
  • 35:03-41:26 - Societal Impact, Education, And Credentials: The closing asks whether AI creates visible public gains in healthcare, education, credentials, and local economic opportunity.

Context

The source is a Latent Space / No Priors crossover interview with Satya Nadella, published by Latent Space on June 3, 2026, after Microsoft Build 2026. Sarah Guo, Elad Gil, and swyx ask Nadella about Microsoft's Build announcements, AI models, enterprise agents, SaaS pricing, engineering roles, infrastructure, and public benefits.

Microsoft Build is Microsoft's developer conference. The 2026 Build framing centered on agents that can use company context and intelligence through Microsoft products such as GitHub Copilot, Microsoft Foundry, Microsoft 365, and Microsoft IQ.

Microsoft Foundry is Microsoft's platform for building, deploying, evaluating, and governing AI apps and agents. Microsoft IQ is the company's framing for context layers that help agents understand workplace, business, web, and knowledge information. The MAI models are Microsoft's in-house model family announced around Build 2026.

Technical Need To Know

  • Harness: The system around an AI model: data, tools, context, evals, user interface, permissions, memory, and feedback loops.
  • Private eval: A company-owned test for whether an AI system performs well on that company's own work.
  • Trace: A record of how work was done by people, tools, and agents. Traces help improve agents inside real workflows.
  • Hill-climbing: Step-by-step improvement against a measurable target. Nadella uses it for improving AI around a company's own workflow.
  • Microsoft Foundry: Microsoft's platform for building, deploying, evaluating, and governing AI apps and agents.
  • Microsoft IQ: Microsoft's term for context layers that give agents access to workplace, business, web, and knowledge information.
  • Delegated authority: Permission for an agent to act on behalf of a person or company inside defined limits.
  • SaaS pricing: Software subscriptions often charge per user; agents complicate that because one user can trigger many model calls, tool actions, and background tasks.
  • Community permission: Local public acceptance of datacenter and infrastructure expansion.

Nuanced Take

Customers may own the private evals, traces, and workplace context that make enterprise agents useful, but Microsoft is building the platform where those assets are measured, routed, governed, and put to work. The control question is whether that harness stays portable across models and vendors, or becomes another Microsoft-shaped operating layer.

Back to allocation feed