Reserved Asset 🔒
https://t.me/addyosmani
Not affiliated with the brand ❌
https://addyosmani.com
Available for legal sale ✅
https://fragment.com/username/addyosmani
https://t.me/addyosmani
Not affiliated with the brand ❌
https://addyosmani.com
Available for legal sale ✅
https://fragment.com/username/addyosmani
Addyosmani
Cognitive Surrender
Cognitive offloading is delegating to the AI and still owning the answer. Cognitive surrender is when the AI's output quietly becomes your output and there i...
https://addyosmani.com/blog/cognitive-surrender/
Available for legal sale ✅
https://fragment.com/username/addyosmani
Available for legal sale ✅
https://fragment.com/username/addyosmani
👍1🤔1
O’Reilly Online Learning
Web Performance Engineering in the Age of AI
On today's web, performance isn't just a nice-to-have—it's essential. A slow or unstable experience drives users away, while a fast, reliable one builds trust and keeps them... - Selection from Web Performance Engineering in the Age of AI [Book]
https://www.oreilly.com/library/view/web-performance-engineering/9798341660182/
Available for legal sale ✅
https://fragment.com/username/addyosmani
Available for legal sale ✅
https://fragment.com/username/addyosmani
👍1🤔1
🤖 The Evolution of Development: Claude Code Launches AI "Agent Teams"
Remember when we were amazed by simple code autocomplete? We are rapidly moving past the era of single AI assistants into the world of fully autonomous AI Engineering Teams.
In his latest article, Addy Osmani dives deep into an experimental feature in Claude Code — Agent Teams (The Code Agent Orchestra). Instead of working with a single virtual developer, you can now deploy a fully coordinated engineering squad right inside your terminal.
🧠 How it works:
Instead of a single model tackling a massive task sequentially (and drowning in context), a parent Orchestrator (Lead) breaks down the goal into structured sub-tasks. It then spawns independent Claude instances — like a Backend Developer, a Frontend Developer, and a QA Engineer — each running asynchronously in its own tmux pane with an isolated context window.
⚡️ The Key Difference: Sub-Agents vs. Agent Teams
* Old Way (Sub-Agents): Hierarchical and isolated. The main agent gives a task to a sub-agent, which returns a result. The sub-agents don’t know each other exist.
* New Way (Agent Teams): True peer-to-peer collaboration. Teammates communicate directly, share API contracts, sync via a shared task list (tasks.json), and use file-locking to prevent code conflicts without constantly bottlenecking the Lead.
🔬 Powerful Use Cases:
1. Adversarial Debugging: Got a complex, silent bug? Deploy a small team of agents, give each a different hypothesis, and let them debate. They challenge each other's assumptions to find the root cause much faster.
2. Parallel Architecture Reviews: Assign one agent to look for security holes, another for performance bottlenecks, and a third for test coverage simultaneously.
🛠 How to try it:
It's currently an experimental feature toggled via your Claude Code settings.json. *Pro tip from Addy:* The sweet spot is 3–5 agents. Any more, and the coordination overhead and token costs can outpace the parallel execution gains.
The Big Picture: We are moving toward a world where developers act less like manual coders and more like engineering managers or orchestra conductors — defining specs, monitoring the dashboard, and approving PRs from an autonomous AI workforce.
🔗 Source: https://addyosmani.com/blog/claude-code-agent-teams/
#AI #ClaudeCode #SoftwareEngineering #TechTrends #Coding
Available for legal sale ✅
https://fragment.com/username/addyosmani
Remember when we were amazed by simple code autocomplete? We are rapidly moving past the era of single AI assistants into the world of fully autonomous AI Engineering Teams.
In his latest article, Addy Osmani dives deep into an experimental feature in Claude Code — Agent Teams (The Code Agent Orchestra). Instead of working with a single virtual developer, you can now deploy a fully coordinated engineering squad right inside your terminal.
🧠 How it works:
Instead of a single model tackling a massive task sequentially (and drowning in context), a parent Orchestrator (Lead) breaks down the goal into structured sub-tasks. It then spawns independent Claude instances — like a Backend Developer, a Frontend Developer, and a QA Engineer — each running asynchronously in its own tmux pane with an isolated context window.
⚡️ The Key Difference: Sub-Agents vs. Agent Teams
* Old Way (Sub-Agents): Hierarchical and isolated. The main agent gives a task to a sub-agent, which returns a result. The sub-agents don’t know each other exist.
* New Way (Agent Teams): True peer-to-peer collaboration. Teammates communicate directly, share API contracts, sync via a shared task list (tasks.json), and use file-locking to prevent code conflicts without constantly bottlenecking the Lead.
🔬 Powerful Use Cases:
1. Adversarial Debugging: Got a complex, silent bug? Deploy a small team of agents, give each a different hypothesis, and let them debate. They challenge each other's assumptions to find the root cause much faster.
2. Parallel Architecture Reviews: Assign one agent to look for security holes, another for performance bottlenecks, and a third for test coverage simultaneously.
🛠 How to try it:
It's currently an experimental feature toggled via your Claude Code settings.json. *Pro tip from Addy:* The sweet spot is 3–5 agents. Any more, and the coordination overhead and token costs can outpace the parallel execution gains.
The Big Picture: We are moving toward a world where developers act less like manual coders and more like engineering managers or orchestra conductors — defining specs, monitoring the dashboard, and approving PRs from an autonomous AI workforce.
🔗 Source: https://addyosmani.com/blog/claude-code-agent-teams/
#AI #ClaudeCode #SoftwareEngineering #TechTrends #Coding
Available for legal sale ✅
https://fragment.com/username/addyosmani
Addyosmani
Claude Code Swarms
Claude Code now supports agent teams - coordinated swarms of AI agents that research, debug, and build in parallel. What was feature-flagged is now real. H...
YouTube
Introducing Gemini 3 Flash
For too long, developers have faced a difficult trade-off: choose a model that is intelligent but slow and expensive, or one that is fast and cheap but lacks deep reasoning capabilities.
Today, we are eliminating that compromise with the release of Gemini…
Today, we are eliminating that compromise with the release of Gemini…
https://youtu.be/gi3cr9ucJx8?is=cwKg2eWFukmUbepQ
Available for legal sale ✅
https://fragment.com/username/addyosmani
Available for legal sale ✅
https://fragment.com/username/addyosmani
👍2
🤖 Your AI Coding Agents Need a Manager: How the Developer's Role is Evolving
In the near future, the highest-leverage developers will look like async-first managers running a small fleet of parallel AI coding agents. The skills that make someone a strong tech lead or engineering manager—clarity, delegation, verification loops, and async communication—are suddenly translating directly into being good at AI coding.
At scale, AI coding stops being a prompting problem and becomes a management problem. The workflow naturally splits into two side-by-side modes:
1. Local, high-touch sessions: Where you stay human-in-the-loop. These are for architecture decisions, tricky refactors, and product nuance where taste and judgment dominate.
2. Cloud or background sessions: Running asynchronously for focused, bounded tasks (straightforward features, migrations, test generation, doc updates). You fire them off, context-switch to something else, and come back to review the pull request.
4 Manager Skills That Transfer Directly to AI Coding:
1️⃣ Clear Task Scoping (Write a brief, not a vibe)
Ambiguity kills agent effectiveness. A practical agent brief should cover the exact outcome, context, constraints, non-goals, and acceptance criteria. A great tactic is putting durable rules into an AGENTS.md file in your repo—giving agents the map, conventions, and definition of done before they start writing, much like onboarding a new human developer.
2️⃣ Smart Delegation
A surprising trap is over-delegating parts that are actually human work: product intent, API design tradeoffs, and architectural boundaries. Some mechanical implementation or boilerplate can be fully handed off, while work touching shared interfaces should be delegated only with clear checkpoint reviews.
3️⃣ Verification Loops
Agents can generate low-quality work at incredibly high speed. Set up early quality gates: require the agent to run test suites, lint, and typecheck before proposing a PR. You can even use a two-agent pattern: Agent A implements, Agent B reviews for correctness, and Agent A applies the feedback.
4️⃣ Async Check-ins
If you are running 5 to 10 parallel sessions, you cannot afford to constantly scroll through logs. Treat agents like remote reports: set a check-in cadence (*"If you haven't made progress in 15 minutes, stop and report blockers"*) and ask for status updates in a predictable format (What changed? What's next? What are the risks?).
The Real Challenges of the Agent Lifestyle:
* Merge conflicts multiply fast: Parallel agents touching adjacent code creates chaos. Solve this by isolating tasks using git worktrees so instances run in separate directories and don't interfere with each other.
* "Should we build this?" becomes the bottleneck: When building code gets cheap, you start building everything. AI doesn't remove the need for judgment; it raises the value of it. Prioritization and saying "no" matter more than ever.
A Simple Operating System for AI Orchestration:
Plan like a manager ➡️ Spawn like an orchestrator ➡️ Monitor async ➡️ Verify aggressively ➡️ Integrate carefully ➡️ Retro (update your AGENTS.md so the next run starts smarter).
🔗 Source: https://addyosmani.com/blog/coding-agents-manager/
Available for legal sale ✅
https://fragment.com/username/addyosmani
In the near future, the highest-leverage developers will look like async-first managers running a small fleet of parallel AI coding agents. The skills that make someone a strong tech lead or engineering manager—clarity, delegation, verification loops, and async communication—are suddenly translating directly into being good at AI coding.
At scale, AI coding stops being a prompting problem and becomes a management problem. The workflow naturally splits into two side-by-side modes:
1. Local, high-touch sessions: Where you stay human-in-the-loop. These are for architecture decisions, tricky refactors, and product nuance where taste and judgment dominate.
2. Cloud or background sessions: Running asynchronously for focused, bounded tasks (straightforward features, migrations, test generation, doc updates). You fire them off, context-switch to something else, and come back to review the pull request.
4 Manager Skills That Transfer Directly to AI Coding:
1️⃣ Clear Task Scoping (Write a brief, not a vibe)
Ambiguity kills agent effectiveness. A practical agent brief should cover the exact outcome, context, constraints, non-goals, and acceptance criteria. A great tactic is putting durable rules into an AGENTS.md file in your repo—giving agents the map, conventions, and definition of done before they start writing, much like onboarding a new human developer.
2️⃣ Smart Delegation
A surprising trap is over-delegating parts that are actually human work: product intent, API design tradeoffs, and architectural boundaries. Some mechanical implementation or boilerplate can be fully handed off, while work touching shared interfaces should be delegated only with clear checkpoint reviews.
3️⃣ Verification Loops
Agents can generate low-quality work at incredibly high speed. Set up early quality gates: require the agent to run test suites, lint, and typecheck before proposing a PR. You can even use a two-agent pattern: Agent A implements, Agent B reviews for correctness, and Agent A applies the feedback.
4️⃣ Async Check-ins
If you are running 5 to 10 parallel sessions, you cannot afford to constantly scroll through logs. Treat agents like remote reports: set a check-in cadence (*"If you haven't made progress in 15 minutes, stop and report blockers"*) and ask for status updates in a predictable format (What changed? What's next? What are the risks?).
The Real Challenges of the Agent Lifestyle:
* Merge conflicts multiply fast: Parallel agents touching adjacent code creates chaos. Solve this by isolating tasks using git worktrees so instances run in separate directories and don't interfere with each other.
* "Should we build this?" becomes the bottleneck: When building code gets cheap, you start building everything. AI doesn't remove the need for judgment; it raises the value of it. Prioritization and saying "no" matter more than ever.
A Simple Operating System for AI Orchestration:
Plan like a manager ➡️ Spawn like an orchestrator ➡️ Monitor async ➡️ Verify aggressively ➡️ Integrate carefully ➡️ Retro (update your AGENTS.md so the next run starts smarter).
🔗 Source: https://addyosmani.com/blog/coding-agents-manager/
Available for legal sale ✅
https://fragment.com/username/addyosmani
Addyosmani
Your AI coding agents need a manager
In the near future, high-leverage developers look like async-first managers running parallel AI coding agents. The skills that make someone a strong tech lea...
👍1👌1
X (formerly Twitter)
Addy Osmani (@addyosmani) on X
Meet Gemini 3.5 Flash! ⚡️ It's fast, great for building rich UIs + agents and is stronger from coding to long-horizon tasks and multi-step workflows.
https://x.com/i/status/2056788908021788811
Available for legal sale ✅
https://fragment.com/username/addyosmani
Available for legal sale ✅
https://fragment.com/username/addyosmani
X (formerly Twitter)
Google (@Google) on X
Introducing Gemini Spark ✨
It’s your 24/7 personal AI agent that helps you navigate your digital life, taking action on your behalf, and under your direction.
🧠 It runs on Gemini 3.5 and is built on @Antigravity, so it can perform long-running tasks easily…
It’s your 24/7 personal AI agent that helps you navigate your digital life, taking action on your behalf, and under your direction.
🧠 It runs on Gemini 3.5 and is built on @Antigravity, so it can perform long-running tasks easily…
https://x.com/i/status/2056791134295273554
Available for legal sale ✅
https://fragment.com/username/addyosmani
Available for legal sale ✅
https://fragment.com/username/addyosmani
👍3