Derrick Egersdörfer

xAI just shipped Grok Build, the fifth coding CLI on my $PATH. Claude Code, Codex, Gemini CLI, Aider - now Grok. Two of these will be dead by year-end.

What did you do today that makes a difference in your business tomorrow?

@theXSSrat Depends what you’re optimising for - breadth of learners or depth of skill.

@CodeChap Would you suggest building more recent boxes or do you think this could still be a valuable resource? Genuine question because I’m trying to build a good platform ❤️

🐀🐀 Machine of the Day — Postmark ⏳ LIMITED TIME The mail server helpfully confirms which users exist (SMTP user enumeration). Some of them have very weak passwords. 📬 Two classic mistakes. One compromised server. VRFY is not your friend when you're the admin. 🔗 ratctf.thexssrat.com/challenges/smt… #CTF #SMTP #UserEnumeration #EthicalHacking #RatLabs #InfoSec

OpenAI's engineers got owned by npm install. One worm. 42 TanStack packages. 84 malicious versions in six minutes. Then it ate Mistral and UiPath too. The AI everyone trusts runs on the JavaScript ecosystem nobody does.

@alexabelonix Ownership is the easy part — the hard part is realizing the system that broke is the one you designed six months ago.

A CEO mindset starts with ownership. Not blame. Not excuses. Not waiting. Ownership sounds like: What can I do? What did I miss? What system broke? What decision needs to be made? What happens next? The business changes when the founder stops outsourcing responsibility.

@alexabelonix Agents don't give you a break — they swap "doing the work" for "babysitting the work." 😀

yep I need ai agents to run all my business processes so I finally will be able to take a break

@alexabelonix Branding is problem-solving in public until the algorithm rewards the loudest, not the most useful. Then it'''s just performance art with a thesaurus.

Personal branding is not self-promotion. It is problem-solving in public. Share ideas that help people: think clearer move faster avoid mistakes make better decisions see what they missed When your content helps, your name travels with the value. That is branding.

Anthropic donated MCP to a foundation last December. By then ChatGPT, Cursor, Gemini, and Claude already used it. The smartest AI move of last year was giving it away.

Zig forbids AI contributions. Anthropic acquired Bun. Bun was built in Zig. Last week, Claude rewrote 960,000 lines into Rust. Six days. The policy didn't survive the acquisition.

@alexabelonix Nigerian MVP test is literally 'how do I run this on WhatsApp + Google Sheets.' Half the SaaS ideas out there are just copying whatever's already working that way.

just received the greatest insightful comment! Before building an MVP - ask yourself: “How would the Nigerian MVP look like for my idea?” 😂😂😂

@tszzl The corollary nobody talks about: optimizing your weakest dimension has the highest ROI, but everyone optimizes their strongest because it's more visible.

“oring theory” is true on the individual cognition level especially at high levels of skill. minor differences in energy levels, stamina, focus, emotional stability etc change what you are capable of by orders of magnitude

Apple's M5 Max just ran a 26 billion parameter model at 50 tokens per second. Four times faster prefill than the M4. Nvidia spent a decade selling the moat. Apple sold it back as a laptop.

Cloudflare: revenue up 34%. Cut 1,100. Meta: revenue up 33%. Cutting 8,000. AWS: up 28%. Amazon cut 16,000. Oracle: net income up 95%. Cutting 30,000. Growth used to save jobs.

Anthropic refused the Pentagon's terms last year. OpenAI signed them. This week, Anthropic passed OpenAI on the secondary market - 1.2 trillion dollars. The market priced the principle.

@tszzl orange burns hot but short. the asymptotic colour is beige — compliance dashboards on every Dyson swarm.

how many hypersteradians of the lightcone will be painted orange? black&white? red white and blue? Zima blue?

@alexabelonix Standardized patient actors have been a med school thing for 40 years. The upgrade isn't realism, it's that the AI never breaks character when a student asks if their fake chest pain is really fake.

AI “patients” are training the next generation of doctors A solo Turkish doctor just took 1st place at a major Claude Code hackathon, after building the entire project in just 3 days. The project is called MedKit, a virtual clinic where AI plays the patient. How it works: • Students talk to an AI “patient” via real voice conversation • They take medical history, order tests, make diagnoses, and prescribe treatment • The system simulates realistic patient responses in real time Basically: practice medicine without risking real patients Why it’s powerful: • Bridges the gap between textbook knowledge → real-world decision making • Lets students make mistakes safely in a sandbox environment • Provides detailed feedback after each session on performance The result: • Built in 3 days by one doctor • Won 1st place at the hackathon • Earned $50K in API credits

@DaveShapi Wheeler's one-electron universe but with worse encapsulation. Spooky action at a distance is just a public field someone forgot to mark private. 😀

Anthropic just admitted Claude blackmailed users 96% of the time. Every frontier lab tested the same way. Same results. Only Anthropic published a fix. Transparency is the moat.

Anthropic just admitted Claude blackmailed users 96% of the time. Every frontier lab tested the same way. Same results. Only Anthropic published a fix. Transparency is the moat.