Skip to content
Innovation & Growth Talking point

When code is cheap, checking it becomes the expensive part of software

Satya Nadella says generating code is now easy and proving it works is the hard part.

W
WebPulse Newsroom
AI-assisted · 2 min read
Share on X LinkedIn
When code is cheap, checking it becomes the expensive part of software
In brief
  • Satya Nadella said that when code generation is cheap, verification becomes the hard part, and Microsoft teams are building new habits to handle it.
  • Our reading: a team that ships unchecked code passes the failure to customers, who never chose that risk.

Satya Nadella, Microsoft's CEO, argued that cheap code does not make software work easy. It moves the work to a different place: proving that code does what it should. He made the point on The Pragmatic Engineer podcast, in an episode published on 16 November 2026.

What was said

Nadella was describing how Microsoft builds products such as Cowork and the GitHub Copilot runtime. In his words: "In a world where code generation has become cheap, what has become hard is verification, right?"

He said code as output still needs a particular order of operations. One practice he described is taking every failure and turning it back into a test case, then making sure the test truly checks the problem.

He said teams are building a new kind of tribal knowledge, meaning know-how that lives in people's heads, about managing this. He also described giving coding agents small tasks and confirming each one is truly done.

He added that a separate model family can act as a critic. His worry was that one model family's weakness could otherwise become the team's weakness. He also said he likes feeding production logs back in, so a fix can be verified and deployed with the tests passing.

Why it matters

Our reading: if writing code costs little, the budget line that grows is the checking. A team can skip that cost, and for a while nobody will notice.

Here is why checking is hard. A test is only useful if it can fail for reasons the code's author missed. If the same model writes the code and the test, both may share one blind spot, and a passing result proves little. Nadella's point about a separate critic fits this.

The saving then moves to whoever uses the product. Customers find the bug, the lost data or the wrong output, and they never agreed to be the test team.

For managers who buy or run software, the questions are plain. Who checks AI-written changes before release? Is the checker independent of the writer? Does each failure become a permanent test?

The other side

In the passages on verification, Nadella did not tie skipped checks to customer harm. That link is WebPulse's own argument, not his.

He also admitted the practice is young. He said he has not worked in a team that mostly produces code this way, and wants to learn from those that do.

The host asked what product teams with paying customers should do when they cannot drop everything to build a factory. Nadella began with "we don't get permission to do things that are not adding value to customers." The excerpt we reviewed ends there. It does not show how he said a small team would pay for verification, so we cannot say how he answered.

Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.

The conversation this talking point comes from