AI Cannot Replace Experts in Complex Software Failures

A senior engineer argues that AI tools lack the system context needed to diagnose complex outages, citing Microsoft's recent disruption.

Updated on Nov. 18, 2026 in Artificial Intelligence

AI Cannot Replace Experts in Complex Software Failures

Live Poll

Do you trust AI to handle important work without experienced people involved?

A senior software engineer posted on Reddit that AI cannot replace experienced human programmers when complex software failures occur. He pointed to a Microsoft outage in which engineers took approximately 15 hours to determine what needed to be rolled back.

Why it matters

The argument highlights a gap between AI's strength at writing code and tests and its limits when failures span databases, infrastructure, applications and networks at once. The engineer warns that companies may only recognize these limits after damage has already occurred.

The only quantified detail is that Microsoft took approximately 15 hours to determine what needed to be rolled back during its outage, a figure drawn from the engineer's account rather than an official incident report. The root cause and the number of affected systems were not disclosed.

The players

Microsoft

Microsoft is a multinational technology corporation headquartered in Redmond, Washington, known for Windows, Azure cloud services and its major investment in OpenAI.

The details

The engineer said that reversing a faulty change requires first identifying the specific component responsible, a task complicated when database, infrastructure, application and network elements are entangled. He noted the Microsoft failure could have originated at any of thousands of checkpoints, and that AI tools lack the broader system context needed to pinpoint the source. He has worked on large projects at multiple US, European and Korean companies.

Timeline

  1. The engineer's Reddit post followed Microsoft's recent outage; the exact date of the post was not specified in the source.

Big Picture

The argument arrives as GitHub Copilot, launched in 2021 by Microsoft's GitHub unit, has made AI-written code routine across the industry. The engineer's claim tests whether those gains extend beyond writing code to diagnosing production failures.

Developers and teams relying on AI tools for code may still need experienced engineers available during production incidents, since diagnosing outages requires system-wide context. Companies deciding how much to automate should account for how long recovery can take when failures span multiple components.

The takeaway

Organizations adopting AI for software work should keep experienced engineers in the loop for incident response, where system-wide context matters most. Treat AI-generated code and AI-assisted troubleshooting as tools that still need human judgment when failures are complex.

Further reading

For more on where these tools are heading, explore our Artificial Intelligence coverage.

Source note: This article includes information reported by NDTV Profit.

Live Poll

Do you trust AI to handle important work without experienced people involved?