Mat-GPT

App Awards 2024 Best New Products: - Clay - Perplexity - Polymarket Time Saver of the Year - Hey (email) Time Waster of the Year - Venmo (search) Biggest Losers - Titan (reservations, rlly?) - Superhuman - Apple iOS Best hardware - Oura - Framework Most awaited - Pocket AI

@GergelyOrosz Just put a 1-liner in the system prompt to bias towards web searching Helps a ton

Claude just keeps regressing for me, day after day. I swear that until a few days ago, when Claude did not know something, it kicked off a web search, figured out, and answered. Now it just refuses to do the work that I pay for. It's like showing you the middle finger. Really?

@kentcdodds @claudeai @Cloudflare Makes perfect sense

@matiaphone @claudeai @Cloudflare It has no inference. It's just well organized primitives with codemode

I had 🐨 Kody build this for my 2.5 year old son …pe-color-match.kentcdodds.workers.dev

@kentcdodds @claudeai @Cloudflare @kentcdodds Is “Kody” actually an LLM? Or just a codemode mcp? Like does the client submit a prompt or call the “Kody” tools via codemode?

Another cool thing that I could do is, if I really wanted to, I could tell Kody to create a GitHub repo for this and turn it into a real software project with CI and tests and all that. And Kody could just do that. Then I could hand things off to Cursor Cloud Workers from Kody to make adjustments.

@RhysSullivan I run a consulting firm for PE firms and am trying to set up executor for my customers. They use Cowork which only supports OAuth-enabled remote MCPs. Goal is to deploy a single executor per customer that they can connect to in Cowork. Whats the easiest way to do this? Or do you think it'd be easier to create a custom worker / codemode MCP for each customer instead?

@kentcdodds @kentcdodds how do you handle auth with the cloudflare entry point AND all the tools that get used in codemode?

I think I'm liking the direction of a personal AI assistant built using MCP and Cloudflare Dynamic Worker Loaders (code mode). I love that I can talk to it through ChatGPT, Claude, or Cursor (existing subscriptions). I just need to build out a good memory system (biggest risk).

@kentcdodds So very similar to executor by @RhysSullivan

@matiaphone Yep

🔥 Oh here we go! I've built a personal assistant agent that works in ChatGPT, Claude, Cursor, anywhere that supports MCP. I just asked claude (in the browser, not claude code) to create a github issue on my project and then fire off a cursor cloud agent to solve it.

@RhysSullivan @GilFeig Arbitrary code execution is obviously the future Just need a simple standardized way to expose an API to the execution layer Which brings us back to….. MCP

@GilFeig >Good MCP has 2 tools. Search tools, and run tool. This doesn't feel right - long term this is going to introduce a problem where if you have 20 MCP servers added each with their own search and run it's not composable You can write MCPs like normal, just need an execution layer

If you hate MCP, you don't understand MCP. You have a bad setup/harness and blame it on MCP, which is nothing more than an extremely thin wrapper. The argument that it's slow is rooted in poor implementations that load up thousands of tools and eat your entire context window. But that's not necessary and it's not how the best AI companies are using MCP in practice. MCP is great because it's a lightweight standardization similar to REST and API protocols. It gives the world the ability to provide a minimal, single interface into their products. It then plugs nicely into all LLMs. MCP or not, we need that. Good MCP has 2 tools. Search tools, and run tool. You don't need anything more than that. You expose millions of tools to your agent, and they run quickly and eat close to zero context. CLI tools are great too. But they're heavier to produce, heavier to set up, require running agents to have CLI access, and give up a ton of security guarantees. MCP was invented when CLIs already existed for a reason. Many reasons. We're not suddenly enlightened by CLIs. Nothing has changed.

@OMGitsRomi @convex yep

@convex down …

@me_jd_solanki @mattpocockuk @cole_medin Use Pencil to spin up quick wire frames. Include them in the PRD as guidance

When doing AFK/Ralph how do you handle guided UI generation? For example, How to provide wireframe or design when it builds new page or feature cc: @mattpocockuk @cole_medin

@mattpocockuk random typescript question for you Is there ANY way to build a discriminated union derived from an object? I have a set of const objects that I want to dynamically build validators for These config objects have a set of "type - id" pairs that I want to discriminate on But object.values collapses the type and the resulting validator ends up as a union of all possible types / ids Is there a way to maintain the discriminated union dynamically? Or just typescript limitation?

@aidenybai Lumen

how do you deslop / review things nowadays? looking at you Claude Code users

@zeeg @vercel @jarredsumner

So @vercel you gonna make a full package manager for skills or does someone else gotta? This thing is a mess

@tomkrcha 2 quick things. 1. The agent inside pencil doesn't have permission to access any of the code (even though the .pen file is in my repo) and there is no way to give it permission. I'm trying to have it create a style guide based on my project. 2. How can I access the official design kits / style guides?

Excited to launch Pencil INFINITE DESIGN CANVAS for Claude Code > Superfast WebGL canvas, fully editable, running parallel design agents > Runs locally with Claude Code → turn designs into code > Design files live in your git repo → Open json-based .pen format

We used to use editors with code review features Soon we'll be using code reviewers with editor features

@tomkrcha One flag — the demo file says “Explore Design Kits and Style Guides at the top — but I can’t find them anywhere!

@dillon_mulroy Can I beta test? Dying for this

making the most performant code review tool this side of the mississppi

@mattpocockuk What's the case for storing instructions in docs/ vs. skills?

Bad AGENTS​.md files can make your coding agent worse and cost you tokens. Here's a prompt you can use to clean them up - PLUS a full guide for folks wanting to learn more. Enjoy: aihero.dev/s/KQLczQ

All the vibe coders gave up right when it became possible

Buy snap