Liam
3.1K posts



@skeptrune Wouldn’t you say it depends on the modality? Like Claude code for X sounds like a developer tool while cursor for x seems more open
English

@ThePrimeagen Thank you for your efforts in almost every technological change for giving it a chance and showing how to learn/master something new!
Not a huge fan of vibe coding myself, but it’s cool to see this.
English

🚨🚨 I GET TAUGHT HOW TO VIBE CODE🚨🚨
🚨🚨 I GET TAUGHT HOW TO VIBE CODE🚨🚨
🚨🚨 I GET TAUGHT HOW TO VIBE CODE🚨🚨
x.com/i/broadcasts/1…

English
Liam retweetledi

We're adding @opencode as a new default to Omarchy 3.3. It's going to have live themeing, just like neovim, ghostty, obsidian, and all the other key tools. github.com/basecamp/omarc…

English

@theo @staysaasy Both have shipped good software that is profitable with multitudes of users. That’s notable given >95% never reach that.
I think you of all people would agree that this is a challenge very hard to achieve.
English

@staysaasy He has not shipped anything notable, much less usable, since selling Minecraft over 10 years ago.
I would have said similar for DHH if he hadn’t done the deep omarchy dive recently + his maintenance of Rails
English

@NathanFlurry @neondatabase @PlanetScale @supabase @CockroachDB You’ll run into latency issues with HTTP over TCP
English

Nice (HTTP support):
- @neondatabase serverless driver (supported by @PlanetScale too)
- PostgREST (supported by @supabase)
- ClickHouse HTTP driver
- DynamoDB already supports it
Naugy (TCP required):
- @CockroachDB
- Cassandra
English

@crazibeat1 @adnansahinovich @expo Most companies style from scratch since they know their users what something customer and unique
English

@adnansahinovich @expo Same reason I’m using flutter, I prefer declarative ui styling, I don’t want to use stylesheet.
And who wants to style all components from scratch?
I rather use heroui native or something else and just customize it minimal to my taste
English

@akramcodez More like the LSP ecosystem is unreal. That’s the true gem that makes it all work
English

@XenZee2025 But due to this problem it is a problem an MCP server needs to concern itself with. It should be as lean and unique as possible.
English

With the traditional approach, the output of each tool call must feed into the LLM's neural network just to be copied over to the inputs of the next call, wasting time, energy, and tokens. So technically, it is a problem of the LLM, not the MCP server. You can wrap the MCP tool in TypeScript.
English

MCP is quite cool but can easily bloat your tools.
With every new tool, hallucinations and token usage seems to increase 🫠
Csaba Kissi@csaba_kissi
MCP is the API for connecting to AI. Utilize it.
English





















