Hi everyone and welcome to this week’s edition of AI-Native Builder Weekly.
My Skills for Planning AI Coding Projects
Here are the Skills and templates I use in my own AI coding projects with Claude and Codex. Copy the install prompt and ask your agent to install them for use in your next project:
- Planning process overview
- /brainstorm skill (and brainstorm.md template)
- /create-spec skill (and project_spec.md and feature_spec.md templates)
An ask: I plan to share more about my current process for building with coding agents in the coming weeks. Let me know which aspects you’re most interested in or any problems you’re facing that you’d like help with (just reply).
Useful AI tools and resources
- FAQ on AI Evals by Hamel Hussain
- GPT-6 Astra for motion design animations
- The Morning Newspaper (plus GrokBot Template)
- Product Management in the age of AI by Josh Elman at a16z
- Inspo MCP: Design inspiration for your agents from real websites
What I’m reading, watching, and listening to
How Databricks rolled out GPT-6 Astra to 3500 engineers: Patrick Wendell (co-founder and CTO of Databricks) posted an insightful note about how they tested and rolled out GPT-6 Astra to their team. I found a lot of good ideas about how to pilot new model adoption in large engineering teams and how to encourage teams to explore the jagged frontier while keeping costs in mind.
Ben Thompson on Business Models in the AI Era: I’m a long time fan of Ben Thompson (and loyal Stratechery subscriber) so I’m definitely biased, but this was an insightful, wide-ranging conversation on the business models underpinning major AI players (OpenAI, Anthropic, NVIDIA, Google and more).
Notable launches this week (9/20)
- Projects in Claude Code: A project is one conversation: Claude splits the work into threads, runs them as parallel cloud sessions, passes context between them, keeps shared memory, and keeps going when you leave. I’m particularly excited about the new Claude Projects, as I shared on Twitter:
This is big.
Anthropic basically built in "The Orchestrator Rule" to the Claude product, where you just tell Claude all the thing you want, and Claude will spin up threads to get it done (plus run cloud agents to work vs locally on your machine).
Very excited to try this out. https://twitter.com/claudeai/status/2100632677904744716
Projects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop.
In beta today for select Pro and Max users in cloud sessions; coming to all Claude users soon.
- Claude Docs, Slides and Claude Design are also just part of Claude now.
- Claude Cowork and chat are now One Claude: You can fire off a quick question and do knowledge work in a single place Claude keeps working even after you close your laptop, using local files and apps when your computer is open and its own cloud computer when it isn't.
- Claude Code (finally) adds support for AGENTS.md
- Codex Multi-account plugin support: Support for multiple accounts across most plugins, covering personal, work, and side-project contexts.
- OpenAI plan to launch something big next week (according to Sam Altman)
That’s all for this week! Reply to this email to let me know what you found most useful.
Happy building,
Avthar
You're receiving this because you subscribed via Maven, my paid courses, or signed up directly. This is a weekly email to help you build, lead, and thrive with AI. Feedback welcome!