Blake Operator🏅

Startups don’t collapse from bad ideas. They collapse when their systems only work under ideal conditions. Perfect focus. Perfect people. Perfect load. That’s not resilience. That’s temporary alignment. Thread 👇

@catalinmpit Every single one of these is a prompting problem I've had and fixed. Explicit instructions in your system prompt, tell it the scope upfront, give it the style guide. It's annoying that you have to but it works.

Lately, Claude makes some shocking mistakes. ⟶ Implements overly complex code ⟶ Ignores the codebase's code style ⟶ Removes working code for no reason ⟶ Replaces code that's out of scope from the task at hand It feels like it needs 100% supervision. At this point, you're better off writing everything yourself.

@49agents @sickdotdev Not sometimes. Almost always. The best product graveyard in tech is full of things that lost to worse products with better distribution. We just don't talk about the graves.

@BlakeOperator @sickdotdev this is a wild take but honestly the billboard thing worked. every dev twitter thread is proof. distribution beats product sometimes, painful as that is to admit

Noticed something interesting: Claude Code auto-adds itself as a co-author on every git commit. Codex doesn’t. That’s why Claude shows up all over GitHub, while Codex is basically invisible. Feels like OpenAI is skipping a very obvious distribution hack here.

@johnrushx In 2008 nobody was saying "I'm going to build the next great app." They were still figuring out what an app even was. That's exactly where skills and connectors are right now.

The end game is that every serious tech company gonna build “general purpose agent” that can do anything (code, research, assist, etc). Same as the end game for mobile was a touchscreen with OS The skills/plugins/connectors/ will be the “apps” of the post-AI era for small teams

@bindureddy Honestly the hardest part now isn't the model. It's knowing your task well enough to pick the right one for it.

The era of cheap and fast models is here! GLM - agentic coding Flash - OCR Kimi - claw orchestration GPT 5.4 nano - classifier MiniMax - dirt cheap Grok Fast - cheap entertainer Claw and open source agent usage is rinsing exponentially every day 🚀

@boringmarketer The people who can't tell the difference are going to have a rough time when the thing they shipped needs to be maintained.

There’s a massive difference between vibe coding and engineering with agents

@VraserX I think about this a lot. Google had the research, the compute, the talent, and the paper that started it all. The only thing they didn't have was urgency. And that turned out to be the only thing that mattered.

OpenAI is the lab most likely to build something historic. Google is the company most likely to bury something historic under 11 product teams, 4 approvals, and a spring launch event. Too harsh or fair?

@kentcdodds Future-you will open that issue, see a wall of perfectly structured context, and have no memory of being this organized. Trust the process.

Telling the agent to create github issues is a great way to defer work you think you may need to do once you hit a certain scale, but you're not there just yet. Empty all the context into that issue so it's easier to pick up from where you left off when it's time to solve.

@Prathkum I've had AI enthusiastically help me walk in completely the wrong direction. It was very encouraging the entire way.

AI is the only piece of tech that does not make you doubt your skills. You build with confidence and even when you are wrong, it says "you are absolutely right."

@allTheYud We solved intelligence before we solved access tokens. That's where we are.

I'm amused by how in the modern era, "Rewrite this terminal UI to CLI so Opus can use it" is a 5-minute task consisting of "tell Opus to rewrite it", and "Okay but now make that new tool's Github private repository visible to Claude operating out of your other private repository" is 10 minutes of trying to wrangle API tokens followed by giving up and hard-downloading the new repository.

@JonathanRoss321 The best feedback I ever got came from one person I met by accident. Imagine not needing the accident.

One of the best ways to uplevel is to work with great talent. Not everyone has had that opportunity. AI is going to change that. Everyone has access. Looking forward to a year of massive upleveling for anyone who wants it.

so the thing literally converted to 2026 cyber-bullying. everybody is suspicious of ai replies. soon, ai will be everywhere and people will be suspicious of everything.

@brian_lovin Intelligence per second per dollar is the metric the whole industry should be using instead of benchmark leaderboards.

Using Cursor again today for the first time in a while. Still using Claude Code, Codex, Conductor, of course. First: someone needs to rename because the C-named companies are out of control. Second: fast is good. Composer 2 is good because it's fast. That's all you need to know to at least give it a try. Third: I am grateful that I can switch between all of these tools in an instant. Little-to-no lock in. I pick the thing that gives me the most intelligence-per-second-per-dollar and am happy.

@nateberkopec 4 seconds to 400ms is not an optimization. It's a different product.

The #1 most important performance work in almost any web app is converting full navigations into SPA/route changes/Turbo Drive clicks. You are converting a ~4 second experience into a 400ms experience every single time. Nothing else comes close.

@OfficialLoganK The roadmap is ambitious. Multiple chats per app alone is something people have been asking for since day one.

Our AI Studio vibe coding roadmap for the new few weeks: - Design mode - Figma integration - Google Workspace integration - Better GitHub support - Planning mode - Immersive UI - Agents - Multiple chats per app - Simplified deploys - G1 support And more, should be fun : )

@bindureddy The right model for the right job is the new skill. Using GPT-4 to classify emails is like hiring a surgeon to take your temperature.

@daviefogarty Volume negates luck is the line most people will scroll past and the only one they actually needed to read.

@dvassallo GUIs were for humans. CLIs are for agents. The interface follows whoever is doing the work.

Soon: Everything is CLI

@sudobunni Doom about AI replacing developers is the new "learn to code" energy. Loud, confident, and mostly wrong.

scrolling on Mastodon: feeling inspired to build cool things and learn about interesting problems people are solving scrolling on X: "developers are becoming obsolete", "there's no point in learning to code anymore", "AI is everything"

@Riyvir Stitch is a good reminder that generating UI and understanding design are two completely different skills.

i honestly think figma hasn't released a design agent yet because the tech just isn't able to do it with consistent excellence. seeing the results from stitch etc is reinforcing this for me.