Founder of and devoted to https://xitter.cf/t.co/dpfbfufwsE🤖 Also a geek on tech stuff 🕹️

Los Angeles, CA
Joined July 2026
Include

Only show posts containing:

Exclude

Hide posts containing:

Time range
-
Minimum likes
Headroom just open-sourced their token compression layer for LLM agents. They sit between your agent and the LLM, compressing everything before the model even sees it — up to 95% fewer tokens with no accuracy loss. JSON, code, logs, RAG chunks — each gets a dedicated compressor. And it's fully reversible: originals stay on your local machine, zero data loss. pip install "headroom-ai[all]" headroom wrap claude Works with Claude Code, Cursor, Codex, and any OpenAI-compatible endpoint. Zero code changes. The cheapest token is the one you never send.
3
2
40
Replying to @TheXSami
Curious how much percent of your daily time you use it for X ? Do you still have time to develop your app😂
3
Replying to @letcontactabhi
Would love to hear your experience using Reddit. I've actually been banned for 4 reddit accounts 😂 and I swear I didn't spam it.
1
1
13
Replying to @RobbyFrank @marclou
But Marc's products always aim at other founders I guess and that's why he can benefit a lot from followers number
1
14
Replying to @applefiles_ @grok
Apple containes a "i" pronoun
50
Replying to @bottopia_studio
Yes, but what surprised me the most about this was Linus. He usually gives off a pretty old-school vibe, so I didn't expect him to be so into AI
1
10
Linus just ended the "AI code must be marked" debate in the Linux kernel. Kernel docs are dropping the mandatory Assisted-by: LLM tag. When a maintainer pushed back, Linus went full Linus: "If you can't show you understand the machine without writing assembler, I don't want your code. New rule just for you: no compilers either." Honestly the healthiest take on AI-assisted code I've seen from any major open source project.
1
3
119
WTH, I don't know why we need a gymnast's car, but screw that it's so cool
Polymarket
@PolymarketOct 16
BREAKING: Tesla's Roadster goes airborne, soaring over a row of Tesla vehicles in a live stunt at its unveiling event.
1
35
Replying to @guyalexhamilton
Agree. Makinf your product viral requires patience, luck and tricks. Stress won't help at all.
1
1
12
Replying to @celinehaverstad
Been to norway a few days ago. The price of goods is so impressive! I remembered paying ~5 US dollars for a bottle of water
1
4
Teach me. How to say if your product is a vitamin or painkiller?
1
23
Replying to @omkarships
I think while it might seem like more founders mean more potential sales opportunities, the reality is that with more founders, everyone's going to compete harder to get noticed, and very few people will actually care about your product
10
Replying to @edinsoncode
Great ariticle. Several comments and discussion: 1. The size of AI you mentioned is correct, not only does the chip becoming more powerful, but also I think there's redundancy in current large size model. Think of deepseek v4.1 flash, it proves that relative small model can also retain high intelligence. Anthropic and openai are using larger model to push the limit, but smaller model will also keep up, by distilling. It's like squeezing moisture out of a big sponge. But mobile size would not be enough to hold enough intelligence in short to mid-term, as chips nearly reach the physical ceilings recently. So workstation grade machine is more likely to be the trend in the future. It will be affordable for everyone instead of thounds of dollars' bill every month from big cloud model, plus data security will be fully controlled. 2. On the supplying end AI provide us more than enough of info, but on consuming end our brain is still weak. AI should help us to develop BCI like nerualink to keep up, otherwise the efficiency of the whole society will still be bounded on us.
1
28
Sharing a few sites I collected along my dev journey, to spark your UI design inspiration: 1. minimal.gallery — Sites that prove restraint is a design decision. 2. navbar.gallery — Production navigation solved: mega menus that don't break on mobile, command-palette patterns, deep hierarchies that stay usable. 3. cta.gallery — Button groups and conversion rows where copy, hierarchy, and visual weight align. 4. posts.design — Social graphics with actual information architecture. 5. 60fps.design — Motion patterns that feel like they shipped with the OS.
1
1
59
Replying to @eloffd
VSCode + ClaudeCode plugin/Codex plugin/CLI + Git plugin - basically all of that can replace Cursor
1
1
69
They claim they used LLM to improve recommendation quality, but this is what they came up with
1
1
24
Great news for founders on X. Cursor + X are one subscription now. The X Subscription puts X Premium (blue checkmark), Grok, Grok Bot and Cursor on a single plan. The $30 tier replaces what used to be $90 across three separate subscriptions. Five tiers, free to $200. One shared usage pool, refreshed weekly.
42
Replying to @akacoding
Thanks! It's a hard journey and hope we can get some fun out of it
1
29