Addy Osmani

Who they are

Addy Osmani is an AI Engineering & DevRel Leader at Google — joined in 2009, co-created Lighthouse and Yeoman, and runs the 'Elevate' Substack with 600,000 readers on AI agent engineering.

Person

Addy joined Google in April 2009 — spending roughly 15 years on Chrome before moving to Google Cloud AI. On Chrome, he led developer experience, shipping tools that became industry defaults: Lighthouse (the web performance auditing tool), Workbox (service-worker caching libraries), TodoMVC (the framework comparison tool), and Critical (critical-path CSS extraction). Before Google, he worked on Fortune 500 websites as a developer — and as a teenager around age 16, he built XWebs, a self-described 'megabrowser' written in hundreds of thousands of lines of C++, which tells you everything about how he approaches a problem. His public output is prolific: a blog at addyosmani.com, a Substack newsletter called 'Elevate' (addyo.substack.com) with 600,000 readers on AI agent engineering and developer experience, and books including 'Learning JavaScript Design Patterns,' 'Image Optimization,' 'Leading Effective Engineering Teams,' 'Beyond Vibe Coding,' and 'Stoic Mind.' He's spoken at 175+ events worldwide — most recently Google Cloud Next 2026 and Google I/O 2026 — and his AI Engineer World's Fair talk was titled 'Don't build agents you can't answer for,' a pointed frame for someone now building agent infrastructure from inside Google. His current side projects include agent-skills (production-grade engineering skills for AI coding agents on GitHub) and OpenClaw/VisionClaw (open-source projects enabling agents to perceive and act). The through-line is tools that scale developer leverage — from browser performance tooling to engineering leadership writing to agent design patterns.

Company

Google Cloud's most recent strategic moment is Google Cloud Next 2026 (April 2026), where it declared the end of the 'AI pilot era' and launched the Gemini Enterprise Agent Platform, the Agentic Data Cloud, and eighth-generation TPU chips. In the same month, Google completed its $32 billion acquisition of Wiz, integrating that security platform into the agentic AI stack — and committed a $750 million partner fund (the largest single partner investment from any hyperscaler) to accelerate agentic AI development with firms like Accenture, Deloitte, KPMG, PwC, and NTT DATA. Sundar Pichai announced planned capital expenditure of $175 billion to $185 billion for 2026, with over half of machine learning compute investment going to Google Cloud. The strategic posture is vertical integration — from TPU silicon to Gemini application layer to Wiz security — with the explicit ambition of becoming the 'operating system' for AI.

Market

Google Cloud holds approximately 14% of the cloud infrastructure market as of Q1 2026, ranking third behind AWS (roughly 28%) and Microsoft Azure (roughly 21%), but grew cloud revenue to $20 billion in Q1 2026 with 63% year-over-year growth — faster than either rival that quarter. Its main AI competitors span cloud (AWS, Azure) and model providers (OpenAI, Anthropic, Meta), and it's taken the unusual step of hosting competing models like Anthropic's Claude on its own infrastructure. EU regulatory dynamics — the AI Act enforcement and the Cloud and AI Development Act — are creating compliance pressure across all hyperscalers, with sovereign cloud and data residency products becoming a differentiating battleground.

Network

Addy's visible network spans publishing (Tim O'Reilly, who hosted him in a live conversation on Google Cloud AI; Gergely Orosz of the Pragmatic Engineer newsletter, who featured him in a deep-dive on 'Beyond Vibe Coding') and engineering community platforms (LeadDev, where he is a featured contributor and speaker). No direct peer or team edges are available from the data.

Tim O'Reilly· Founder & CEO, O'Reilly Media
Gergely Orosz· Author, The Pragmatic Engineer newsletter

How they likely show up

Long tenure at Google (joined April 2009, still there in 2026) → thinks in multi-year product arcs; won't be impressed by short-term metrics or quarterly framing.
Shipped multiple open-source tools (Lighthouse, Workbox, Yeoman, TodoMVC, Critical) while employed full-time → high agency, builds things to solve his own problems and ships them publicly rather than keeping them internal.
Runs a Substack with 600,000 readers and has given 175+ talks worldwide → comfortable being the visible, public face of a technical position; engages with ideas in public before they're fully settled.
AI Engineer World's Fair talk titled 'Don't build agents you can't answer for' → brings a skeptic's lens to AI hype; will engage more with accountability and failure modes than with capability benchmarks.
Wrote 'Beyond Vibe Coding' and regularly posts about software quality alongside AI tooling → draws a hard line between AI-accelerated engineering and slop; likely loses patience with sloppy AI demos.
Career arc from browser performance tooling (Lighthouse, Core Web Vitals) to agent engineering → sees developer experience and AI capability as the same problem at different layers of abstraction.

Conversation tips

→ Reference a specific post or talk by name — he produces a lot and will immediately know whether you read it or just saw the title.
→ Frame AI questions around accountability and quality, not just speed — his public thesis is that AI coding tools raise the floor but can collapse the ceiling if engineers disengage.
→ Ask about the Chrome-to-Cloud-AI transition — moving from browser tooling to agentic AI is a significant shift and he's written about it ('Hello, Gemini'); he'll have a considered view on what transferred and what didn't.
→ Bring a concrete engineering problem rather than a general topic — he writes about high-leverage activities specifically because he's skeptical of meetings that don't move something forward.
→ Don't pitch Stoicism back at him unless you've actually read something — he wrote 'Stoic Mind,' so surface-level name-drops will land flat.

Toolbox

Openers

Open on agent-skills — he published a GitHub project of production-grade engineering skills for AI coding agents, which is a very specific bet that agents need curated, opinionated skill libraries rather than general capability. Ask what gap it fills that existing agent frameworks don't.
Reference the AI Engineer World's Fair talk title directly: 'Don't build agents you can't answer for' — it's a pointed frame, and asking what 'answering for' an agent actually looks like in practice will get you into the substance fast.
Bring up 'Beyond Vibe Coding' — he wrote a book explicitly pushing back on low-accountability AI-assisted development, from inside the team that ships Gemini. That tension is worth naming.

Discovery questions

You shipped Lighthouse from inside Chrome to set a performance standard the whole industry adopted — what would the equivalent accountability layer look like for AI agents, and does Google Cloud have the leverage to impose it?
Your Substack argues for high-leverage engineering activities — how does that calculus change when an AI agent can do the low-leverage work, but the engineer still has to own the outcome?
You moved from Chrome DevEx to directing Google Cloud AI — what from 15 years of browser tooling turned out to be directly applicable to agent engineering, and what had to be unlearned?

Avoid

Don't lead with generic AI productivity claims or 'AI will 10x developers' framing — he has written a book ('Beyond Vibe Coding') specifically arguing against low-accountability AI coding, and he'll disengage if the conversation stays at that level of abstraction.

Make it yours