The Pragmatic Engineer, #Gergely Orosz, recently published some thoughts (link here) into how AI tools are being used in real-world software development. By talking to and studying engineers at AI companies, Big Tech, startups, and seasoned developers, he captures a “tale of two realities” picture that sits between the extremes of executive hype and developer skepticism.
The Two Realities: Hype vs. Hesitation
Executive Hype: The Future Is (Almost) Here
Many tech leaders have made sweeping predictions about AI’s transformative potential:
- Anthropic's CEO predicts all code will be AI-generated within a year.
- Microsoft's Satya Nadella claims AI already writes up to 30% of their code.
- Google's Chief Scientist believes AI will soon perform at a junior developer level.
These statements suggest that AI coding tools are rapidly reshaping software development. But how does this bold optimism align with what’s actually happening on the ground?
Developer Skepticism: Not So Fast
While executives tout a revolution, many developers Gergely cites remain unconvinced:
- Half of all developers don’t even use AI tools weekly, indicating that many have tried them and moved on.
- At one biotech startup, 90% of AI-generated code review comments were unhelpful.
- GitHub Copilot Agent failed repeatedly in a real .NET codebase, creating “schadenfreude” among developers (here).
This discrepancy raises an important question: if AI is so capable, why isn’t it more universally embraced?
Changing Minds, Rediscovering Joy
Some of the most respected voices in software engineering have shifted from skepticism to enthusiasm:
- Armin Ronacher (Flask): Now acts as an "engineering lead to a virtual programming intern."
- Peter Steinberger (PSPDFKit): Calls this "the most exciting time since learning to code."
- Simon Willison (Django): Says AI coding agents now “actually work”—no longer just toys.
- Kent Beck (XP): “I’m having more fun programming than ever—in 52 years.”
- Martin Fowler: Compares LLMs to the leap from assembly to high-level languages.
Their evolving views underscore that AI isn’t just about speed—it’s about transforming how we think and feel about coding.
Adoption, Impact, and Unanswered Questions
Adoption Metrics
The data paints a mixed picture of usage:
- 50% of developers use AI tools at least weekly
- At top companies, that rises to 62%
- Median time savings are about 4 hours/week—a solid 10% productivity gain
Yet this still means half of developers aren’t using AI tools regularly.
Open Questions
Several big questions remain:
- Why are executives more optimistic than engineers?
- Why have so many developers tried AI tools and stopped?
- How much real-world time savings do these tools provide?
- Why do tools often underperform on an organizational scale?
- Why isn’t there more buzz if the impact is truly massive?
These gaps suggest that while AI tools have matured, their effectiveness depends heavily on integration, trust, and workflows.
Bottom Line: A New Reality Is Emerging
The reality of AI in software engineering is neither overhyped magic nor overblown disappointment. It’s something more grounded—and promising. For those who’ve embraced these tools, the productivity gains are real. And thanks to recent breakthroughs—especially AI agents that use command lines and can receive feedback—capabilities are advancing fast.
As Kent Beck puts it:
“Things that we didn't do because we assumed they were expensive or hard just got ridiculously cheap. So, we just have to be trying stuff!”
If you haven’t started experimenting with tools like Claude Code, Cursor, or Windsurf, now’s probably the time.
AI development tools are poised to become as ubiquitous as IDEs and Git. The future is pragmatic—and it's already here.