Raduan Al-Shedivat

path to one person one billion dollar company 🙂 x.com/0xRaduan/st...

doesn't get better than this:

coding agents showed that everyone should use a monorepo.

ASCII is all you need:

CC shocked that it was able to ship 3 weeks worth of work in 1 day:

Interesting thing from "inference" newsletter: They write post methodology on how they wrote the post with LLM tools.

multi clauding

Everything is just a system prompt + llm + tool calls + while loop.

cooking something great obviously with claude code:

Build Calm Confidence Software:

claude code as a way to learn new english words:

Talking to Claude about how premium software should feel:

it's happening: I am migrating to uv. cc. @charliermarsh @astral_sh

Loading screen. Onboarding almost ready.

Windsurf team keeps shipping: If they continue at this pace, this will be a super interesting space to watch (and compete in). x.com/windsurf/st...

farza on high quality launch videos / cluely:

with every slop fake headline that perplexity pushes in their app, the less I will use it

Using GPT-5 in @windsurf after wave 12, and it's noticeably better agent than what they had previously. Using high reasoning, it's slow, but instruction following is next level. Integration with DeepWiki also feels like an unfair advantage at this point. Very well done.

ffmpeg is all you need:

If only people would think in decades: Then they would care more about proof of work... Than any cluely-level short term marketing stunt to get views on some post. Stop posting rage baits on this timeline and start building something great.

We need grouping of tool calls back in Cursor. This is getting out of control quickly...

From @zeddotdev blogpost by @conradirwin. My .02 cents: This will always be like that - humans in the drivers seat, but LLMs remove a ton of inefficency and help you translate inputs into outputs. Which means you will have flatter org structure and just faster pace of work.

Gergely's keeps bashing Cognition, though their product keeps improving week over week, and many competitors copy what they've built into @DevinAI. Might be saying more about the audience of Gergely (devs from big tech, who haven't used Devin), than about the Devin itself.

I want to use gpt5. But it's super disappointing around 50% of the time, and wastes my time. How do people use it successfully here?

asking chatgpt to filter out SEO driven content

Just can't imagine it getting better than this.

Write stuff down: via @simonw

Just realized there is a way to make Cursor Agent a main chat window. This should be the default view in a post IDE world. cc. @ryolu_

everything is good about posthog, but it's super slow

The more agency I give to an agent the more surprised I am with results.

chatgpt reading vs code docs: "the page is long"

Running these in parallel: - claude code - cursor - cursor-agent - windsurf - codex Any other to try? cc @devinai - I need a devin CLI.

Built a message queue for ChatGPT like in @cursor_ai: x.com/0xRaduan/st...

Which blogs I should read to learn more how people are using gpt 5?

Early testing of gpt5: - amazing at following instructions - follows everything I told it in rules, similar to gpt4.1. - I let it debug issues on it's own - I still can't see using it instead of sonnet/opus for long-running high level use cases. - It's amazing in Cursor

agentic coding pro tip: make all your devloop tools output token efficient

all your sources, one home:

not gonna lie, every time I see em-dash now questioning whether it's ai written or not not that it's a bad thing - I will still read the whole thing, if it's from good source. but would double-check it in case it's some hallucinated text. so not a bad side effect.

Talk to your users:

x.com/julianweiss...

Sidebar News. Inspired by @dubdotco / @steventey; built with the help of @21st_dev / @serafimcloud

Anthropic: - releasing 2% increase - realizes everyone is releasing something bigger - we are planning to release substantially larger improvements to our models in the coming weeks agents are accelerating x.com/AnthropicAI...

figuring out how to make more delightful pictures for changelog:

Great time to be building x.com/OpenRouterA...

what if you had a separate inbox, but just for newsletters? > filter out emails by sender > read them / preview links that they sent > get AI summaries / key takeaways for each email > see stats and block useless senders in one click shipped and available at summate[.]io

ask ai to write design docs, once it understands your code

Very interesting: since chatgpt launch, data centers spending only increasing and office construction is only decreasing ignore this data at your own risk.

Good take from @balajis. Doubling down on your prompt and building better systems for verification is what you need to be spending your time on. Improving inputs. Verifying outputs.