Here's how I do Real Engineering with Opus 5.5 (Workflow w/ Tools)

Star PandaCode on GitHub: https://github.com/aicodeking/pandacode Join this channel to get access to perks: https://www.youtube.com/channel/@aicodeking/join In this video, I'll show you how I engineer with Claude Opus 5.5 on long-running tasks, using real examples from PandaCode, the app I built with it. First, I'll explain why simple prompts work better than step-by-step workflows, and why you should ask your agent to write a benchmark before you ask it to make things faster. Then I'll talk about getting Opus to write thoughtful tests so that nothing reaches you before it passes them, and about giving it reference designs from sites like recent.design and Dribbble. Finally, I'll show you how I use orchestration with Agent Teams in PandaCode, with Opus as the architect, Sonnet as the workers and Codex as the reviewer, plus a few small tips for long tasks. -- Key Takeaways: 🧠 Keep your prompts simple: say what you want, how you'll know it's done and what the agent must not touch, then let Opus pick the path. 🔍 Opus 5.5 reads the codebase and finds a better order of work than a step-by-step workflow you wrote before you saw the problem. 📊 Ask your agent to write a benchmark first, so it has a number to improve instead of a vague "make it faster" goal. ⚡ With benchmarks, PandaCode's window opens in about a third of a second, and follow-ups went from about 8 seconds to about 4 seconds to the first word. ✅ Opus sometimes writes very few tests, so ask for a small set of thoughtful tests and make every change pass them before you look at it. 🎯 Design is subjective, so give the agent reference designs from recent.design or Dribbble instead of letting it guess. 💻 PandaCode has moodboards and a sketch tool, so your references stay with the project and every new task can use them. 👥 Always use orchestration: Agent Teams in PandaCode give you an architect that plans and reviews, and workers that run in parallel. 🔀 My usual team is Opus 5.5 as the architect, Sonnet 5.5 as the workers and Codex as the reviewer, and you can save named teams in settings. 👍 These habits don't make Opus perfect, but they let it work for a long time on its own and catch the basic problems before you see the result.