AI helps developers learn more and code more productively, but it writes the what but loses the why. This makes maintenance and collaboration difficult. @awakenjake and @gregce10 with @specstoryai shared a their workflow to keep the "why" and stay in control (including OSS templates and resources to help you get started). The brief overview is: Plan: Use Markdown files (spec md, tasks md) to set clear and detailed guardrails and define the work before you code. Implement: Guide the AI as a pair programmer. Don't just accept; intervene and make ad-hoc corrections as it works. Reflect: After the session, sync the entire chat history. Use the AI to analyze that log and extract a clean decisions md file. This decisions md file becomes the summary for your pull request. It shifts the code review from focusing only on what changed to including why it changed, giving you a durable record of intent. Learn more 👇 elite-ai-assisted-coding.dev…

there are only two ways to get any actual speedup from ai coding agents: 1) work alone, 2) work as one (trunk-based) everything else just pushes the bottleneck to team communication, which in many cases was already the actual bottleneck

Got the chance to sit down and chat with @mitchellh about AI disclosure in open source contributions, thoughtful use of AI coding assistants, and the future of version control. About half of contributions coming in to Ghostty are AI generated/assisted. Maintainers and contributors can take steps to help each other out when it comes to these contributions. newsletter.specstory.com/p/m…

digitalocean.com/community/t… Get good at open source contributions using AI coding agents. "AI is not replacing open source contributors. It is giving them new ways to participate. Sustainable collaboration depends on transparency, respect, and careful review. Tools like SpecStory preserve not only code, but also the reasoning behind it, which makes reviews faster and more instructive for everyone."

With GitHub’s coding agent, a new pattern has emerged: group PRs. One person kicks it off, but multiple teammates jump in, shaping design together in real time. Reviews are turning into jams, less critique, more creation. It’s not a code review anymore. It’s a code jam, the developer version of FigJam. #github #codingagent #ai

We’re all figuring out the right tool for the job, which means being able to bounce between agents seamlessly. SpecStory now supports Codex CLI so you can bring all your AI coding sessions from the past to Codex.

Agentic chat histories are rich enough to recreate a codebase if it was AI-native. open.substack.com/pub/specst…

Companies that don’t hire younger SWEs will lose out in the next five years. At Shopify, we have a leaderboard for AI usage, and the Top 100 (out of 8,000 employees) is packed with the small cohort of New Grads who joined with me last year and some cracked Staff Engineers.

“AI suggestions are opportunities to learn”. 💯 Just wrote about this in the context of open source contributions using AI newsletter.specstory.com/p/h…

Never lose your AI coding chats💥 SpecStory automatically captures every AI-assisted chat and prompt as local Markdown - instantly searchable, version-controlled, and shareable with your team. - Revisit decisions weeks later and understand the reasoning behind a code/feature change - Debug faster by tracing how a solution evolved across prompts - Review PRs with context and see the exact conversation that led to the code. 🎥 Here’s a quick demo: see how SpecStory captures your Cursor sessions right into .specstory/history/ so the reasoning lives alongside your commits. 👉 Works today in VS Code with Copilot, Cursor, and Claude Code.

Here's the whole PR:

Software development is undergoing a meaningful shift. Traditionally, the act of building software was often judged by its output: PR commits, features shipped, tickets closed, etc. But anyone who’s built systems knows the real work has always been reasoning through tradeoffs, constraints, and design choices. That picture is now changing. AI coding tools are making it easier than ever to generate code. The real challenge isn’t code generation. It’s clarity: deciding what to build, why it matters, and how it should behave. In other words, defining the requirements rather than writing the syntax. And that shift changes how we think about building software. And that’s not just a product manager’s job anymore. Every role now influences the product in meaningful ways. Engineers make product decisions when they choose performance over simplicity. Designers shape the product when they adjust flows that reduce friction. SREs and infra teams define the product when they set standards for reliability and scalability. Even support and customer success teams contribute when they bring forward constraints from real users. The future of software development will not be defined by code alone. It will be defined by the conversations, the decisions, and how clearly we capture and communicate the requirements.

And of course, use specstory.com to save all your Claude Code histories as markdown

So happy to see that others agree with us that a clean, beautiful place to create agent-ready specs can make requirements engineering more zen 🪷 Wish we could send everybody a Mac if they don't already have one. (But don't worry, we have love for Windows/Linux fans too.. building our Cursor/VSC extension pushed us deep into WSL land).

Sundays I often caught between “I should relax” and “I should get a jump on the week”. It’s especially acute today since tomorrow is a US holiday. The thing I’m most interested in is this word “should”. I think it’s the root of this feeling and the struggle of productivity.

If you work with lots of different projects in Cursor/VSC and the list in File->Open Recents isn't enough for you, try Ctrl (not Cmd)+R. It gives you a long list of recent projects with type-to-filter.

Simple to include in your README markdown: