Lots of folks feeling FOMO about MCP. Here's the problem it solves:

Who has built some sick Cursor agent custom modes? @kregenrek echo.hive Ian Nuttall Ray Fernando Mckay Wrigley Also Pontus Abrahamsson — oss/acc we need a new section for cursor.directory for these (especially if they embrace a json config)

In Cursor having a custom agent mode is nice just for the fact that at any conversation turn, you can encourage more specific tool use. Trying to integrate something tricky? Tell it to be liberal in using its “Search Files” and “Read Files” tools.

Respect to Ryo Lu and team for the automatic, relevant icons in custom modes. A really nice touch.

o3 deep research with Github integration is insane

I think Claude makes the most sense, brand and model behavior wise to power Apple devices. It’s clearly the high-eq model without a distribution home. Deep focus on safety and security and from my experience its users seem to skew higher income.

I just pulled the trigger last night on Claude Max, then the next day o3-Pro drops!?

It’s time for the major AI labs to embrace a react-flow based agent builder GUI. Google DeepMind OpenAI Developers Anthropic It’s the unlock: coding/tech barrier is too high for 70% of companies. Microsoft Research is paving way with their power Platform and bending Power Automate.

I’ve made more progress in the last week with Claude Code than I have in a month with Cursor & Windsurf. The context volume clearly matters. I’m betting it’s my hesitancy to use the Cursor Max Mode vs the flat fee for Claude Max. Cursor should have usage based Max Plan.

I like Claude code because it works better. I don’t romanticize the command line: I would prefer if it was an IDE

If you want to become AI native, you have to be willing to try, tinker, and fail over and over. The people who get stuck are those who are waiting for someone to teach it to them.

I love Cursor, I like that I have the option to run background agents on my phone. But, I’m struggling to see, given Cursors required use of variable cost max models, what it does that OpenAI Codex (fixed rate) or Jules (free) doesn’t do?

I think the windsurf and Claude code PM deals over the last two weeks have taught us the value of the mid-level devs over the department and company leads in this AI era

ChatGPT agent should have a special routine for registering the warranties on users consumer products, with memory to log what’s been warrantied. Would help Pro users pay for itself Kevin Weil 🇺🇸

Cursor should just sell to Anthropic. At those API prices, they absolutely need each other.

The AI bubble take is such a boomer one, over indexed on the latest OpenAI model. Google is making meaningful progress, we’re weeks away from bigger leaps on Opus coding.

"AI will widen the gap between superstars and everyone else," per WSJ