Before your next website build, give the planning as much attention as the tools. In this video, I walk through BUILD, the process I use to take a website from the first business conversation through to launch and handover. B: Business discovery Understand the business, its customers and the requirements. What does the website need to help people do? U: User experience and technical planning Map those tasks for customers and staff, then plan the pages, admin features and services needed to support them. I: Implementation Build and connect the features you've agreed, checking them as you go. L: Launch preparation Test complete tasks and their failure cases. A working checkout also needs the right order record, stock update and customer email. D: Delivery and ongoing ownership Launch, check the live site and hand over access. Make sure the owner can operate it and knows who is responsible for ongoing upkeep. I demonstrate the process with a real website in the full walkthrough. Watch the full video: https://www.youtube.com/watch?v=1m0mvtrBVXo Resources below the video 1. BUILD-guide.pdf — keep this beside you during the client call. It includes the discovery questions and the checkpoints for each BUILD stage. 2. BUILD-AI.md — give this file to your AI assistant with your call transcript or notes. It helps you turn the conversation into a draft brief and work through the next stages. Start with the PDF for your conversation. Use the AI file afterwards, then check the brief with your client.

One instruction file for Claude Code and Codex. AGENTS.md holds your project’s rules—things like writing style and when to ask for your approval. Claude Code now supports it too (v2.1.277+). By default, it falls back to AGENTS.md when no applicable CLAUDE.md is present. Already have both? Compare them, resolve disagreements and keep the useful rules. Check your setup before deleting anything. Official details https://code.claude.com/docs/en/memory#agents-md

My new video is live on YouTube! I ran OpenAI Astra for a week. It is my favourite model so far. I get a more finished first draft than I did with Claude, with fewer hallucinations. It also loves burning tokens. I was hammering it for most of that week on extra-high reasoning or Ultra, with a lot of windows open. I maxed out my credits on the $200 plan, used OpenAI's resets three times, and still blew through the lot. Ultra is the same model at the same reasoning level, and it burns far more tokens. So I would start on Low with a task I can judge, check the result, and step up only when the work needs it. On the Artificial Analysis chart, Astra Low scores 46 for intelligence at $0.82 per task and GPT-5.6 Sol Max scores 47 at $1.99. Those are benchmark task costs, not anyone's bill. Which model is doing most of your work right now, and what made you pick it? https://youtu.be/T_CdGSK0AXk