Patrick Erichsen

How are folks keeping prompts/rules in sync with their code? This feels like an obvious problem I'm not sure has been solved specifically for per-repo configs, e.g. local prompts/rules versioned alongside a codebase

⚡️Parallel tool calling landing soon in @continuedev most providers support this out of the box, should make planning mode (lot of read-only tool calls that can be run in parallel) noticeably faster!

This is already the case for popular CLIs like `gh`, even Claude Code docs recommend using the GitHub CLI over the MCP Uses a fraction of the tokens (no verbose tool defs), no risk of tool confusion, easier to understand

"schizophrenic mind-map of hyperdimensionally clustered snippets of code" is my new favorite way of describing codebase rag

💻 Agent mode just went local! Today, @MistralAI released Devstral, an Apache-2.0, 24B-parameter model trained for tool calling in real-world software development environments It reached #1 on SWE-bench for open-source models, but more importantly in our real-world testing it…

I'm excited to join @continuedev as CEO. After working with @tylerjdunn and @NateSesti first as a user, then advisor, I'm thrilled to have the opportunity to help build what I believe in: open developer tools that amplify developers' work without trying to replace them.

Watching how agents solve problems is such a great way to learn a new language Basically like pair programming with your language-savant co-worker that has infinite patience for your questions on why they are doing things the way they are

Nothing like morning coffee, browsing Discord feedback, and cranking out a few oneliner PRs ☕️

This is a game changer for monorepos!

Continue now supports glob patterns when determining which custom rules to apply 📏 Even better, Continue can generate your custom rules for you! Just ask something like "Create a rule to always use Tailwind when editing .tsx files" while in Agent mode

#4 feels like a missing piece of UX in all assistants today eg let me branch off the current context, ask a "quick question", and easily return to the convo

@Docker MCP Catalog 🤝 @continuedev Agent mode

The future of AI-native dev is local 🧠💻 @continuedev built an offline-first, TypeScript-native AI assistant, powered by @lancedb blazing-fast, privacy-respecting code search. No infra. No cloud. Just fast, smart dev workflows. Case study 👉 blog.lancedb.com/the-future-of-…

I'm going to call it right now. A lot of stuff is going to break on this mission. By design. As part of the plan. Don't get upset. I'm not saying SpaceX plans to fail. I'm pointing out that SpaceX has taken an ultraimportant principle from software engineering, and realized…

Writing docs pre-agents: manually include context, explain everything in detail, etc Post-agents: let the agent explore the feature itself, simple prompt to "now update the docs" at the end Working alongside agents to write docs is not only easier but quite fun

10M context length on Llama 4 Scout is wild - for the first time, I can now just shove the entire @continuedev repo into context using tools like @gitingest

🚀 First Open Source Contribution Merged! 🎉 Super excited to share that my first open-source contribution has been merged into @ContinueDev! 💻 Open-source collaboration is an incredible learning experience - more to come! 🚀 #OpenSource #AI #DevCommunity #JavaScrip

This is a great example of where specialized models shine @NateSesti and I spent a lot of time attempting to get a deterministic algo reliably working for Apply (and it's still useful for slower local models), but models trained on Apply are ultimately more reliable

Would not have guessed Russian is the 2nd most popular language on the internet

I've started to notice that quite often I'll reprompt LLMs for very trivial changes, eg I changed my mind on a variable name Imagine the human equivalent - forcing someone to start over from scratch, on the exact same task, because you want a different variable name