The Next Two Years of Software Engineering
Addy Osmani's survey of the macro forces reshaping software engineering. The headline finding: a Harvard study shows junior developer employment drops 9-10% within six quarters of AI adoption, while senior roles stay stable. But the piece resists doomerism -- the BLS still projects 15% growth in software jobs through 2034, and non-tech industries may absorb what Big Tech sheds. The real question isn't whether jobs disappear but whether the nature of the work changes beyond recognition.
Key Quotes
"The best software engineers won't be the fastest coders, but those who know when to distrust AI."
"I don't want to end up as a code janitor, cleaning up what the AI throws over the wall."
"Engineers become 'composers,' orchestrating ensembles of AI agents."
Key Themes
The "distrust AI" quote is the most important sentence in the piece. It reframes the value proposition: expertise isn't about what you can build, it's about what you can catch. That connects to Fresh Eyes (using different models to catch blind spots), AI PR Reviewer (automated catching), and AI Zealotry's emphasis on tests over code reading.
The T-shaped developer argument is sound but not new. What's new is the data: 45% of engineering roles now expect multi-domain proficiency. The specialist-generalist question was always relevant; AI just accelerated it.
Critical Analysis
This is a well-researched survey piece, but it hedges so carefully that it's hard to extract a strong take. "Multiple futures are possible" and "reality will combine elements" is true but not useful for decision-making. The strongest section is on education -- the observation that 45% of companies are dropping bachelor's degree requirements signals a real structural shift.
The companion piece (Addy Osmani's Workflow) is more actionable. This one is better as context-setting. The tension between the two is revealing: the workflow piece assumes human oversight as non-negotiable, while this piece acknowledges futures where that assumption breaks down. See Five Levels from Spicy Autocomplete to the Dark Software Factory for a framework that makes the Level 3-4 transition explicit, and Two Kinds of User Are Emerging for the enterprise adoption angle.
Sources: The Next Two Years of Software Engineering Last updated: 2026-05-14