Coding Agents: Nadella Calls for Verification and Separation

For Satya Nadella, producing low-cost code does not solve the challenge of controlling it. The President and CEO of Microsoft also advises companies to maintain their ability to change their AI model.
Nadella Wants Controllable Agents Without Technical Dependence
Companies should view their AI agents as an internal risk, believes Satya Nadella, who was interviewed by Gergely Orosz for The Pragmatic Engineer. While capabilities are advancing, reliability is not keeping pace, he asserts, echoing a point he made on 20VC last month. He calls for immutable logging, subject to audit, of the reasoning behind these agents.
In his view, the now low cost of code generation shifts the difficulty to its verification. The models used to examine this code may themselves suffer from alignment issues, so he favors cross-checking between distinct model families. He also wants to integrate traces from production into a loop that controls corrections, a mechanism he refers to as "my RSI."
This mastery also involves separating the software that hosts the agent, its memory, and the models used, to avoid lock-in with a provider. His criterion is to be able to substitute one model for any other while improving internal evaluations. A company unable to do this should, he believes, question its sovereignty.
Microsoft Sees Its Interfaces Evolving with Models
In terms of tools, Nadella acknowledges that Microsoft was surprised by Claude Code, the product from Anthropic that shifted work from the editor to the terminal. Orosz questioned him about the lack of an immediate response to its arrival in 2025, despite GitHub Copilot and VS Code. Nadella recalls that Copilot initially offered code completion; Cursor then introduced a robust agent mode, which Microsoft eventually reached.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
The executive attributes the success of this command-line interface to Anthropic's model. About six months after this shift, the integrated development environment, or IDE, had, according to him, lost its centrality. He deems it necessary to quickly adapt interfaces to the nonlinear leaps of models, without being attached to the existing ones. However, he envisions that the IDE will regain a place when decision models are integrated into software, comparing them to stored procedures for the era of intelligence.
Nadella also states that the most advanced models from last year are already running on a laptop, and those from the last quarter on a desktop. He predicts that routers like GitHub Copilot Fusion will distribute tasks between local execution and the cloud. He presents this perspective as the beginning of hybrid intelligence and describes local models as "unmeasured intelligence."
Microsoft Plans to Restructure Teams Before Hiring
Within Microsoft, the size of teams and their methods must evolve, announces Nadella, as the company employs 200,000 people. According to him, some components have not yet integrated the transformation; the role of deployed engineer is among those progressing the fastest. For recent graduates, he sets a prerequisite: reorganize teams and build their new "software factories," then recruit. However, he considers it premature to extend the conclusions drawn from the current slowdown over time.
Regarding voluntary departure offers, Nadella did not respond when Orosz asked if they preceded layoffs. He presents them as a way to grant more autonomy to employees, who may choose economically relevant roles for them in the long term. He encourages those who feel stuck to leave their team, highlights the existence of numerous internal positions, and refuses to promise that nothing will change.
Microsoft has also created PRAISE, a mentoring program for early-career engineers, in collaboration with Carnegie Mellon University. To engineers, Nadella recommends reading less about AI and practicing it more. He expresses a desire to apply this rebalancing himself starting now, prioritizing action over discourse.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.