Addy Osmani is an Irish Software Engineer and leader currently working as a director at Google Cloud AI, focused on helping developers and businesses succeed with Gemini, Vertex AI, and the Agent Development Kit (ADK). After nearly 14 years leading developer experiences in Chrome, he now bridges Google DeepMind, engineering, product, and developer relations teams. A developer for 25+ years at Google, he is passionate about AI-assisted engineering and developer tools. Follow for motivation, productivity and engineering.

Addy Osmani

The real engineering skill isn't getting AI to 100% - it's recognizing when you've hit the point of diminishing returns at 90%. That's your cue to take the handoff.

You're not "giving up" by reading the code and finishing it yourself. 

You're doing what senior engineers do best: knowing when to switch tools, when to own the architecture and quality, and when speed matters less than understanding what you're shipping.

Let the AI sprint. You close with precision. It's okay if that last mile takes a little longer.

#ai #programming #softwareengineering

Director, Google Cloud AI. Best-selling Author. Speaker. AI, DX, UX. I want to see you win.

Chrome DevTools now supports *individual* network request throttling!

Folks have been asking for this for years! DevTools now allows developers to simulate slow network conditions for specific requests rather than the entire page. This helps in testing how a web application performs and handles issues when specific resources (like images, scripts, or API calls) are slow to load

Open the panel from ⋮ > More Tools > Request Conditions. Toggle the Enable blocking and throttling checkbox to enable request blocking. You can click the + button to add a new text pattern. 

#webdevelopers #programming #softwareengineering

Announcing big changes to Gemini in Chrome! Agentic browsing arrives.

Today marks a massive shift in how we experience the web. Starting today, Gemini in Google Chrome is evolving from an assistant into an agent that can navigate, reason, and execute complex tasks on your behalf.  

This includes a new design, Nano Banana support, deeper integration with Google products and agentic features like auto-browsing.

Here is what's changing:

👉 1. The Agentic "auto browse" 🤖

Gemini can now handle multi-step workflows across the web. Whether it's finding a walking tour on Expedia that fits your weather forecast or scrubbing your Redfin favorites for pet-friendly apartments and inviting your roommate to collaborate - Gemini does the clicking so you don't have to.

👉 2. "Nano Banana" image integration 🍌

Real-time image transformation is now native to Chrome. Powered by Nano Banana, you can now re-image furniture in a room or edit photos in-line without the tedious download-edit-upload cycle.

👉 3. Uninterrupted workflow with the new side panel 💻

Multitasking has a new home. The redesigned side panel stays with you as you browse, giving Gemini native access to your tabs. It can synthesize data from multiple sources, draft emails in Gmail based on a syllabus you’re reading, and even fill out complex registration forms from a PDF.

👉 4. Agentic commerce (Etsy, Expedia, & more) 🛍️

Planning a party? Show Gemini a photo and it will hunt down the supplies on Etsy, stay under your budget, and populate your cart. It’s moving from "searching" to "getting things done."

Rolling out first for Google AI Pro and Ultra subscribers in the US. 

#ai #programming #softwareengineering

Announcing Web Quality Skills: let AI agents write higher-quality code! 🚀

I'm excited to share Web Quality Skills - a comprehensive collection of Agent Skills designed to teach your AI coding assistants (Claude Code, Cursor, OpenCode, Antigravity) not just what to build, but how to build it right.

Repo: https://lnkd.in/gfGDWXTW
Skills: https://lnkd.in/g-Rrfn38

While most prompts focus on functionality, these skills encode the collective wisdom of web performance engineering into your AI.

What's inside?

✅ Guidance for many Lighthouse Audits (Performance, A11y, SEO)
✅ Core Web Vitals optimization patterns (LCP, INP, CLS)
✅ WCAG 2.1 accessibility standards
✅ Stack-agnostic (Works with React, Next.js, Vue, Astro, or plain HTML)

Instead of fixing performance after you deploy, you can now tell your agent:

"Audit this page for web quality issues" or "Optimize performance and fix Core Web Vitals"

It's like pair-programming with a performance expert. This is an unofficial project and something I personally use on my apps, but wanted to share it with y'all in case you find it useful. You can also pair this with the Chrome DevTools MCP so let your AI agent "see" what you do in the browser.

Get started with this one-liner:

npx skills add addyosmani/web-quality-skills

I hope you find it helpful! 

Demo in the video.

#ai #programming #softwareengineering

"Can you read 900 words per minute?"

RSVP is a way of displaying information, like words, one at a time, very quickly, in the exact same spot on a screen.

Instead of your eyes moving across a page to read a paragraph, your eyes stay fixed in one place while the content "flashes" by. It is essentially speed reading, where the content moves so your gaze doesn't have to.

#programming #softwareengineering

"The next two years of software engineering" five questions to ask yourself and your team with recommendations to consider

My latest free write-up: https://lnkd.in/g4EU-f5Z ✍ 

Software engineering is at a strange inflection point.

AI coding has evolved from autocomplete to autonomous agents. The hiring boom gave way to efficiency mandates. And a new generation of developers is entering the workforce having used AI assistance from day one.

I wrote about five questions that may shape our industry through 2026:

1. The junior developer question: Will entry-level hiring collapse as AI automates junior tasks or rebound as software spreads into every industry?

2. The skills question: Will core programming skills atrophy as AI writes most code or become more critical as humans focus on oversight and the hardest 20%?

3. The role question: Will developers shrink into code auditors or expand into orchestrators designing AI-driven systems?

4. The specialist vs. generalist question: Will narrow specialists find their niche automated or will T-shaped engineers who blend depth with breadth thrive?

5. The education question: Will CS degrees remain the gold standard or will faster paths (bootcamps, portfolios, employer training) overtake them?

For each question, I explore two contrasting scenarios and offer concrete recommendations for both junior and senior developers.

The consistent thread: change is the only constant. I hope some of the ideas here help you stay in the loop regardless of which future arrives.

#ai #programming #softwareengineering

Vibe Kanban: orchestrate multiple AI coding agents in parallel. Free and 100% open-source!

Switch between Claude Code, Codex Gemini CLI, and track task status from a single dashboard. Repo: Repo: https://lnkd.in/gA_UPD8a

Vibe Kanban is essentially a project management layer for your coding agents. Instead of just staring at a terminal while Claude Code or Gemini CLI does its thing, this gives you a UI to:

1. Orchestrate multiple agents: You can run them in parallel or sequence.
Visualise the work: It puts tasks on a board so you can actually track status (hence the Kanban name).

2. Centralise config: Manage your MCP configurations in one spot.

3. Remote access: It handles SSH, so you can open projects on remote servers easily.

I've been enjoying trying out a few orchestration tools this last year, including Conductor, Jules, Claude Squad and of course Claude Code Web, GitHub Copilot Agent etc. While the tools you use might change, I think now is a great time to build your skills and comfort levels using parallel agents. 

#ai #programming #softwareengineering

Build a full-stack Deep Research Agent with React and Google ADK!

Research is one of the most popular use-cases folks have for agents. 

Today I'd like to show you our Deep Search Google ADK (Agent Development Kit) sample - a sophisticated, full-stack application that solves two major problems in AI development: lazy execution and misalignment.

It plans before it acts: instead of rushing to search, it pauses to strategize. it presents a research plan to the user first, allowing for a "human-in-the-loop" moment to refine the direction before resources are spent.

It self-corrects: it implements a "critic" loop. if the research findings are weak or gaps are found, the agent autonomously generates new questions and searches again until the quality bar is met.

This is also not just a notebook: this is a complete app including a React + Vite frontend, a Python + FastAPI backend, and deployment scripts for Google Cloud, making it a true template for production apps. Even if your Python is dusty, I promise you'll find it relatively straight-forward to navigate.

Check out the GitHub repo: https://lnkd.in/gbAe5Zck

...as well as the many, many agent samples we've got in the same place :)

#ai #programming #softwareengineering

"How to write a good spec for AI agents"

My latest free write-up: https://lnkd.in/g--r4kEu  ✍ 

Using an AI agent for coding or other use-cases? I keep hearing "just write a better spec" but the real problem isn't effort. You can write something that rivals an RFC and still watch the model fall apart once the context gets too large.

What's worked for me is aiming for specs that are clear, scoped, and designed for how agents actually process information, not how humans wish they would.

A few principles that consistently pay off:

Start with a high-level vision, then let the agent draft the details. Treat the first pass like a product brief. Review, correct, and refine before any code is written.

Structure the spec like a real PRD. Clear sections, explicit commands, testing expectations, style examples, and hard boundaries beat loose prose every time.

Break work into modular tasks instead of one giant prompt. Small, focused context wins over dumping everything into a single message.

Build in guardrails. Define "always do,” "ask first,” and "never do.” Encourage self-checks and verification against the spec before you review anything.

Plan first, execute second. Use read-only or plan modes where possible, then iterate. Specs work best as living documents that evolve with the project.

The takeaway: good agent specs aren't about writing more; they're about writing just enough structure and nuance to keep the model focused without overwhelming it.

If you've been frustrated by agents drifting, hallucinating, or missing requirements, the fix is usually upstream. The spec is the first thing you and the AI build together.

#ai #programming #softwareengineer

"My AI coding workflow in 2026"

Write-up: https://lnkd.in/gvcs_5sc ✍ 

I get asked about my workflow for coding with LLMs pretty often so I thought I would share some notes on what works well for me. 

Here is my 5-step workflow for this year:

1. Specs before code (The "15-Minute waterfall") 

Don't dive straight into generating code - plan. I start by brainstorming with the AI to create a comprehensive spec.md and a step-by-step project plan. We define requirements, UX, and edge cases before a single line of code is generated. Really improves quality. Cursor, Cline, Claude Code and many tools now encourage more planning before acting to support this workflow.

2. Context is king - without it, LLMs can make incorrect assumptions.

LLMs are only as good as the information you feed them. I use "context packing" - dumping relevant docs, existing patterns, and constraints into the session or project directory (using tools like gitingest or Context7). Don't make the AI guess or waste tokens trying to figure out what's in your head.

3. Manage task granularity - just like with traditional engineering, do work in well defined tasks or chunks, not all at once.

Break work into iterative chunks. "Implement one function/component" works infinitely better than "Build this app." I use a structured prompt plan, executing tasks one by one to keep the model on the rails. 

4. Review and test what's important - LLMs are often an "over-confident pair programmer" 

It's healthy to treat AI output like code that came from an over-confident pair developer. It might look right, but is it? Review and test everything important. I weave testing into the loop - generating tests alongside the code and refusing to merge until they pass. If testing feels too much for your project (maybe it's a personal project), at least review the diffs. 

5. Git is your save point - commit early and often. 

When an AI hallucinates or breaks the build (and it will), you need granular commit history to roll back to a safe state immediately. I've had so many experiences where I feel great 3 prompts in then the next 2 completely go off the rails. Tools increasingly support the idea of checkpoints and rollbacks, but for most workflows, just using git + committing can get you far.

In terms of models and tools: I have a core set of models (incl Gemini 3) and tools that I rely on pretty regularly, but I make time each week to try out what's new (patterns, open-source projects etc). 

So if the Ralph Wiggum loop, Gastown/Beads or other ideas trend they're on my radar and I can choose whether they make sense for my workflow.  I'm very big on orchestration right now and use CLIs for more complex local work. It's very much just continuous learning + trying things.

AI tools are incredible force multipliers, but they don't replace the need for engineering fundamentals. In fact, they reward them. The human engineer remains the director of the show.

#ai #programming #softwareengineering

Announcing "Product Engineering with AI" - a practical, 200-page book bundle on building AI-powered products from foundations → tools → real-world implementation!

Get it (and sample chapters) at: https://prodengineer.org  🚀 

If you're a software engineer or product manager trying to ship with agents, this is the playbook we wish we had. Shipping AI features today is less "write every line" and more "steer tools + editors + agents, then verify quality."

Hassan Djirdeh and I wrote Product Engineering with AI to make that shift concrete: AI dev platforms, AI code editors, and agentic workflows - plus prompt patterns, debugging, and UX practices that hold up beyond prototypes.

The bundle also includes bonuses:

• Product Engineering with Claude Code: hands-on PM workflows (parallel agents, commands, MCP integrations). Lots that also apply to Gemini CLI.
• AI in Practice: a 30-page case study collection of real production AI examples
• Product Engineering Insights in 2026: insights from experts across the field on how roles, tools, and workflows are changing
• Google Agent Development Kit (ADK) samples - 5 new agent code examples for product engineers including a PRD Studio, Customer & Meeting analysis and more.
• A 45-min workshop: how we built an AI-powered product from scratch
• A 30-min talk: designing human-centered AI experiences people trust

We hope you find it useful!

#ai #programming #softwareengineering

"The AI coding trends for 2026" (so far!)

My latest free write-up: https://lnkd.in/gjFFq_8N ✍ 

The AI coding landscape in 2026 has shifted fast. I just published a primer on the patterns and platforms defining how we build software this year. It's going to keep evolving and I plan to keep my write-up updated.

Here are the top trends shaping 2026:

🔄 The Ralph Wiggum pattern: We are moving from single-shot prompts to autonomous loops. Agents now run tests, hit errors, and fix their own code until the "completion tag" lights up. 

📦 Agent Skills: Expertise is becoming portable. Just like npm install, you can now install "skills" (like Vercel's performance rules or accessibility guidelines) directly into your agent's context.

🏗️ Orchestration > Assistance: Tools like Conductor, Vibe Kanban, and Gas Town are turning developers into fleet commanders. We aren't just pairing with one AI. We are managing parallel agents working in isolated git worktrees. I see this becoming a more default way we work.

🧩 Sub-Agents: Monolithic assistants are out. Specialized sub-agents (one for security, one for docs, one for tests) are in, reducing context pollution and hallucinations.

🦞 We are also seeing a massive rise in local-first tools like Clawdbot (with a lot of hype around it right now) and how we approach personal intelligence. 

Even at Google, we know that the industry feels like it's moving fast and I wanted to try making it just a little easier to stay updated as we all share and figure this stuff out :) Also included my fav. community posts as a bonus.

#ai #programming #softwareengineering #lifeatgoogle

Hi and welcome to my page!

I am a problem solver passionate about designing empathetic, scalable, and reliable software.

Currently focused on building world class experiential tech at Salt XC and exploring AI agents for new age workflows.

Previously spent the last 4 years building full-stack fintech experiences at companies from pre-series to super-unicorn scale. Focused mainly on scaling BE data pipelines and building elegant data heavy UIs.

You can find my resume and github linked below!

taplio

The Battle

Addy Osmani

Krishna Solanki

Followers

Posts

Engagements

Likes

Performance ScoreAddy Osmani's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Addy Osmani

Krishna Solanki

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Addy Osmani's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreAddy Osmani's LinkedIn Performance