Brainstorming with Claude Code for a longer time is underrated.
Some notes on building more confidently with Claude Code (CC) or other coding agents:
If you are building with coding agents like CC, you might be spending a lot of time on the initial prompt. But this is less iterative (not ideal for building) and leads to underspecified prompts that leave the agent guessing in most cases (what framework to use, version, project structure, etc.).
An alternative approach to this, especially if you are not confident in your prompt/context engineering skills, is to interact with your coding agent for multiple turns to help provide it with just enough details before it tries to build the feature/project. It's important not to overdo it. You will know when it's "just right" as you will see the coding agent eager to get started with implementing when it feels it has enough context (apparent from the more confident responses over time).
You want to be productive with the coding agent at the end of the day. Don't be strict with how much you spend on brainstorming, as it depends on the complexity of the feature/project v1. But this process can save you long hours/days of frustrating interactions with the coding agent in the long term. I know many of you are experiencing those today. This is one way you can avoid too many of those frustrating sessions.
I find that conversing with CC for longer helps to build context that CC can then use to build out features or entire projects (end-to-end) in some instances. It's no different from when you are interacting with teammates before deploying. This is particularly useful when you are getting started. What you notice quickly is that you become more aware of what CC lacks upfront, and these longer interactions become less frequent over time and as you work on other projects/features.
The reason I believe this works is that you are essentially narrowing the AI's scope of exploration. And you are helping to fill in project details that the coding agent would otherwise be guessing. Underespecified prompts are the worst, and this applies to coding and other areas.
When building with coding agents, don't underestimate how important your knowledge and skills are to improve the reliability of the agent. Human-AI-Brainstorming is one of the many underappreciated human-in-the-loop approaches that can give anyone (with prior programming knowledge) a significant edge in the age of more advanced coding agents.
Join the Google Chrome Built-in AI Challenge 2025 to create new web applications or Chrome Extensions using a suite of AI APIs and powerful built-in models, including Gemini Nano.
🦖 Submit by October 31 ↓
https://t.co/n3xQELsLCd
Spread the love for open source with #Hacktoberfest, a month-long celebration of open-source projects, their maintainers, and the entire community of contributors. https://t.co/hgQ0lVZWlq
made it to render irl and in the world of pokemon!
christina caught me coding this in between sessions. My friend graham said it’s just a matter of time until my posture is as bad as his lol. It comes with becoming a senior dev
I’m heading back to #RenderATL next week — this time as a Brand Ambassador for @ResilientCoders!
Last year was one of the most transformative experiences I’ve had in tech. Code, culture, and community took over ATL.
The Carnival of Code is almost here. See you there 🎪✨
Init swag just dropped 🔥
You can get your hands on these limited edition swags 👇
All you need to do is claim your Init ticket and share it on socials!
Claim your ticket and join the giveaway: https://t.co/E32xjovA1D