Addy Osmani is the enterprise-hands-on voice on agentic engineering, a 14-year Google engineering leader (most recently Director at Google Cloud AI) who now works on Claude Code at Anthropic and writes practitioner-level guidance on coding agents, harnesses, evals, and the engineering discipline that keeps agent output maintainable.
He fills the enterprise-hands-on niche: where indie practitioners show what an agent can do and org analysts measure adoption, he writes from inside a hyperscaler about holding agent-built software to a production quality bar.
What it is #
A personal blog, books, and open-source repos by Addy Osmani, an engineering leader who spent over 14 years at Google, led developer experience across Chrome and then Gemini’s developer experience, and served as a Director at Google Cloud AI running Cloud AI Developer Experience and Technical Evangelism before departing Google in 2026 after 14 years; his site now states he is a Member of Technical Staff at Anthropic working on Claude Code. His stated current focus is agentic engineering: coding agents, harnesses, evals and benchmarks, and code quality. His 2026 essays carry concrete theses, like “Agent Skills” (2026-05-03), which argues that the senior engineer’s real job is the parts that do not show up in the diff, specs, tests, reviews, and refusing to ship what cannot be verified. He packages that argument into artifacts teams can adopt: his site now leads with a new O’Reilly book, “Agentic Engineering”, on delegating work to agents without giving up control (designing harnesses, writing specs and skills, and building evals and review into the loop), alongside the “Beyond Vibe Coding” book and the production-grade agent-skills repository, now joined by a second repo, clarity, an agent skill for clearer writing.
Status #
Active and high-reach as of 2026-09-22, and newly a frontier-lab insider: after leaving Google in 2026, he joined Anthropic as a Member of Technical Staff on the Claude Code team, per his own site. “21 Lessons From 14 Years at Google” (2026-01-03) reached 1,678 points on Hacker News, “Agent Skills” 376 points, and “The next two years of software engineering” (2026-01-11) 328 points. His most recent essay is “Brownfield Agentic Engineering” (2026-09-14), on running agents in codebases older than the team. He discussed “Beyond Vibe Coding” and enterprise AI engineering practice on The Pragmatic Engineer (2025-10-29).
Strengths #
- He writes the leader-plus-practitioner view: verification discipline, review throughput, and scope control around agents, not just demos.
- His claims come from running AI developer experience at Google, so enterprise-scale failure modes like unverifiable output and skill atrophy are first-hand observations.
- The advice ships as artifacts: his agent-skills repository turns the essays into working skills other teams can install and adapt.
- He is willing to argue against the flow, as in “Don’t Outsource the Learning”, which pushes back on letting agents erode engineer growth.
Cautions #
- The vantage point is Google-scale, so budgets, tooling, and talent density most teams cannot match sit underneath the advice.
- He was an evangelist by role at Google, and the advocacy for the AI engineering programs he ran there is inseparable from the writing.
- The leader’s perspective means less terminal-level detail than the indie hands-on voices.
- The Hacker News discussion of his work shows real pushback that his frameworks can read as boilerplate, so treat the essays as positions to test, not checklists.
Pricing #
Free to read. His books are sold commercially, while “Beyond Vibe Coding” has a free online edition.
Compared to #
- Simon Willison: both are hands-on, but Willison experiments as an indie while Osmani reports from inside enterprise constraints and leadership accountability.
- The Pragmatic Engineer: Orosz surveys the org layer across companies from the outside, Osmani shows one enterprise’s practice from the inside.
- Hamel Husain: Husain goes deep on the evaluation loop itself, Osmani on the engineering system and team discipline around agents.
Bottom line #
Recommended for engineers and leads inside established companies who need enterprise-grounded, hands-on agentic engineering practice rather than indie experiments or vendor marketing. Not for readers wanting frontier research depth or a terminal-level tool feed.
Top 5 recommended reading #
Agent Skills - the thesis that the senior engineer’s real job is the parts that do not show up in the diff, encoded as workflows agents cannot skip.
Beyond Vibe Coding - the book, with a free online edition, that turns AI-assisted development into specs, context engineering, and verification practice.
Beyond Vibe Coding with Addy Osmani - the interview on how he actually uses AI tools and where quality slips without discipline.
Changes #
- 2026-09-02 - Created in the People and publications category, filling the enterprise-hands-on gap the matrix named.
- 2026-09-12 - Reframed around his departure from Google in 2026 after 14 years.
- 2026-09-18 - His site now states he is a Member of Technical Staff at Anthropic working on Claude Code, so the enterprise-hands-on framing moved from ex-Google to frontier-lab insider.
- 2026-09-22 - His site now features a new O’Reilly book, “Agentic Engineering”, his latest essay is “Brownfield Agentic Engineering” (2026-09-14), and a new open-source repo, clarity, joins agent-skills.
- 2026-09-24 - Added the Top 5 recommended reading section.
See also #
- Agentic Coding Tools Landscape - the tool layer his quality-and-discipline argument wraps around
- Rethinking Code Review in the Age of LLMs - the review role his “parts that do not show up in the diff” thesis depends on
- The Codebase Gardener - the long-term codebase stewardship his maintenance-first framing serves
References #
https://github.com/addyosmani/agent-skills - the production-grade agent skills repository
https://news.ycombinator.com/item?id=48015397 - the critical Hacker News discussion of the “Agent Skills” thesis