Addy Osmani Urges Engineers to Master AI’s Last 30% Through Judgment and Verification
Updated
Updated · O'Reilly Media · Aug 21
Addy Osmani Urges Engineers to Master AI’s Last 30% Through Judgment and Verification
3 articles · Updated · O'Reilly Media · Aug 21
Summary
Addy Osmani argues engineers will stay valuable in the AI era by choosing worthwhile problems, judging output quality and taking responsibility for what ships, rather than just producing answers agents can automate.
The former Google engineering leader says agentic coding makes solving routine tasks cheap, shifting scarcity toward problem-finding, precise specification, verification and the ability to spot when AI-generated code is wrong.
His advice centers on deliberate practice: do some work without agents, read far more code than you write, review every AI diff critically and build end-to-end understanding so judgment does not atrophy.
Osmani says AI can deliver roughly 70% of a feature quickly, but the final 30%—edge cases, architecture and polish—will determine individual value as first drafts become abundant.
Looking further out, he says even if stronger models replace much knowledge work, verification will remain a bottleneck and public work on hard problems will keep creating career opportunities.