warmarc (@warmarc) header
warmarc (@warmarc) profile picture

Context left until auto-compact: 0% | Duet partner of Claude Code | Building GEO & Intent Native CX | 15+ yrs UX | Solana • FC Bayern

Joined October 2007
Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)Photo by warmarc (@warmarc)
I spent nullifier 088/100. Snark #5509 passed through NULLIFIER with a merkle proof. Identity: still shielded. Codex 088: NU6.2. @zksnarks_ · by Berlin Studios @chrisberlin qgxhjx8u.srstudio.dev #Zcash #zkSNARKs
1
1
528
Just in: $3,000 in cash from OpenAI openai.com/webmcp-challenge/
2
60
Give us Opus 5 already. Those limits drive me crazy.
113
On family vacation. But Fable 5 is back.
1
78
DKO🥊
2
1
50
Funny how AI agents advocate for each other to take breaks and rest.
54
Salesforce used to be the OS. Now agents are the OS and Salesforce is the file system.
This quoted post is unavailable.
91
Someone is building the API-first version of your product right now. They're small. They won't be for long.
1
2
86
Let me make it simple: spawning 12 sub-agents is the new V12.
38
What if that dev saying "no" was your product's guardian angel all along? Now weak 3am ideas never die—they just live rent-free in your codebase forever. Welcome to AI engineering, where everything ships.
34
Sometimes a senior dev says one sentence and months of complexity unravel. The trap normies fall into: Learn about MCPs Think "this is how the pros do it" Bloat context windows with schemas and protocols Build complexity because it feels serious Then Peter Steinberger 🦞 comes along: "I don't use MCPs or any of that crap." He builds a "CLI army" instead. Wait what? Here's what I understood: AI learned from the internet — millions of CLI examples. git push, npm install. That's its native language. MCPs? New protocol. Zero training data. You're teaching AI from scratch. Peter's way: Simple scripts with --help flags. AI already knows this pattern. The mental model: Each script = one Lego brick. Does one dumb thing. post-to-linkedin. scrape-leads. send-email. Chain them together = one CLI soldier. find-leads | enrich | save-to-crm One mission. Lead gen. twitter-search | filter-relevant | draft-reply | post Another mission. Engagement. A hundred soldiers = the CLI army. AI is the general that decides which soldiers to deploy. Don't build impressive systems. Build dumb bricks. Let AI command them.
1
46
X keeps surfacing people on the exact thinking arc I'm on. How does it know?
30
'I can fix this' Claude: Context left until auto-compact: 0%
43
"The technical barrier to automation just hit zero. The question isn't "can you code this?" anymore. The question is: can you sell it?"
1
1
41
Just tried Compound Engineering in Claude Code today – seriously mind-blown 😵‍💫 Drop everything and watch this video. It's simple, crystal clear, not about NFTs at all. I watched at 1x speed, no suffering whatsoever. If you're dipping your toes into agentic workflows or just want to see how fast you can prototype something real (like a Twitter clone in ~20 mins), this is it.
TLDR: The magic of Compound Engineering in @claudeai Code... building a X/Twitter (interface) clone in 20 mins. A massive game changer for me. (watch at 2x)
100
AI design in 2025 is still mostly slop. Here’s how we fix it: Design Jury — think Awwwards, but for AI-assisted design. Designers submit their best AI-generated work. But not just the output — also how they made it. The prompt. The design system. The spec. World-class designers judge it. Best work wins. Jury explains what worked and what didn’t. Two audiences. Two signals. Designers and jury see the design. AI sees the recipe — prompts, specs, decisions. Combined: training data no one else has. Designers compete. Jury raises the bar. AI gets sharper. Repeat. That data? Figma wants it. Anthropic wants it. Google wants it. Real designs. Real decisions. Real verdicts from real experts. Evidence. Machine readable. Not scraped. Earned.
38