after evaluating Sublime Text for 15 years, I have finally purchased a license.

Microsoft open-sourcing Copilot for VSCode feels like a concession to Cursor. Cursor’s so far ahead, MS would rather open it up and hope startups build competing tools on top than risk losing the entire VSCode ecosystem to forks.

I used to pick the ‘MAX’ models in Cursor Agent thinking they’d be better, but it was so slow I’d just write the code myself. TIL: leaving it on ‘Auto’ makes it way faster and still smart. That’s when the magic happens and I am now vibe coding complex rust database code!

using Opus 4 puts Claude Code heads above every other code agent out there. The only way to sustainably use Opus 4 is via Claude Max plan. I really only use Cursor for the auto completion now.

biggest hidden news today is the Apple Containerization Framework which allows you to run docker images natively on mac without needing to download Docker #WWDC25

failling out of love with Cursor Tab. It doesnt seem to know the fields of the structs. So the generated code almost always refers to non-existent fields. Most of the time I have to delete it. Feels especially jarring after using Claude Code which has almost perfect accuracy.

Linear + Notion looks cool until you have tried pasting a JQL query in Confluence and getting a dynamically updating table of Jira issues!

seems they took the pre-trained Grok 3 model and scaled RL to the max. its why it was originally gonna be grok 3.5. as expected, pre-training is fully saturated at this point. #Grok4

no coding model 🫤 thats all i was looking foward to. Claude Code remains king for now #Grok4

new #Superman movie is basically a SNL parody skit with a vfx budget. they even have the guy from SNL in the movie...

using Claude Opus with the new 'ultrathink' mode in Claude Code. I am now burning $5 per message lol! wonder how long Anthropic can sustain this.

The only way Cursor competes with Claude Code at $20/mo: Kimi-K2’s low cost inference. Sonnet 4-level quality + enough requests to make it feel unlimited.

in your Claude Code prompt, append 'ultrathink with multiple agents' and watch it burn an entire sun's worth of energy for your query.

this. html is dated. instead of evolving the rube goldberg of hacks on top, fresh start using canvas is the path i am for.

interesting that its Opus only for now. no Sonnet. the only economical way to use Opus is via Claude Code. using via Cursor/Copilot through API will result in insane bill

gpt 5 router trying to guess how much thinking budget to assign each message == scrum folks trying to put story points on engineering tasks

Pitted Codex CLI (GPT-5) against Claude Code (Opus 4.1) on a nasty debugging task: compare two branches, trace the root cause. No contest. Claude Code runs laps around GPT-5.

after a long session with Claude Code when you think it really did root cause a complex bug for you, and then you see this .... 🤥🤦‍♂️

Bitbucket is Atlassian's hedge against Github Issues. The day Github starts investing into Issues seriously enough to pull users from Jira, is the day Atlassian starts making serious investment into Bitbucket. Until then, both sit in maintenance mode.