Before any of this had a name.
First-gen by circumstance, idealist by temperament, realist by budget. No inherited playbook, just intergenerational priors, an unfortunate amount of existential debt, and a pathological appetite for becoming.
For a while, fear handled the reward modeling and social approval supplied the loss function. A quite elegant optimization scheme, until the optimizer started consuming its own operator.
The subsequent policy recalibration involved replacing heteronomous teleologies with epistemic sovereignty, reallocating probability mass from performative survival to curiosity-driven agency, and treating software, autonomous systems, and worldmaking as instruments of existential self-authorship.
No ergodic guarantees. No benevolent universe. Still, one human attempting to bootstrap meaning out of contingency, one absurdly ambitious side quest at a time.
Let the Absolute Idea unravel, its light cascading over the ruins of inherited teleologies, baptizing the unfinished self in the unbearable lucidity of becoming.
Let reality update the posterior. Never let it rewrite the objective function.
Know who you are, find yourself!
I'm interested in AI agents, agent operations, and I like to help others. I'm focusing on my own startup, Guli.
I love publishing repos. Most of what I post is something I like and think might help someone else, so I put it out there. I stay very active, and I push an update roughly every 3 days.
If anything here is useful to you, please leave a ⭐ on the repo and follow me. It really keeps me going.
3 AM and the hallway hums. A door that was shut a minute ago is open now, light leaking out like it has somewhere to be. Someone is standing in it, or maybe I am the one standing and the corridor is what moves. Nothing in the footage explains itself. Neither do I. The build is still green though.
And because debugging at 3 AM feels exactly like surviving a night shift, here is a night shift scene I drew:
One bigger open-source system at a time. Click a card to open it.
|
|
|
What ties them together Agents get rules (AGENTS.md). They should also leave a record (TRACES.md), talk to each other (aisync), and the same loop works for a person (existence-trace). Also: star-ranked top-* lists (github1) →
|
Refreshed automatically every hour, star counts are live.
| Project | ⭐ | Updated | What it is |
|---|---|---|---|
| top-open-weight-coding-models |  |
2026-11-02 | Open-weight coding and agent models (Qwen Coder, DeepSeek, GLM, Kimi, Devstral, StarCoder) plus t... |
| top-flying-car-open-source |  |
2026-11-02 | Open-source tools for flying cars, eVTOL and drones: flight controllers (PX4, ArduPilot, Betaflig... |
| top-ai-agent-guardrails |  |
2026-11-02 | Guardrails for AI agents: sandboxes, human-in-the-loop approval, cost and budget limits, prompt-i... |
| top-chinese-rainbow-fart |  |
2026-11-02 | Chinese internet compliment generators, rainbow-fart plugins, meme makers and joke corpora, ranke... |
| top-summer-2027-internships |  |
2026-11-02 | 2027 internship and new-grad job lists, ranked by GitHub stars |
| top-free-javascript-courses |  |
2026-11-02 | Every free way to learn JavaScript, ranked by GitHub stars |
| awesome-agent-workflows |  |
2026-11-02 | Curated, star-ranked catalog of AI agent frameworks, coding-agent harnesses and workflow platform... |
🧪 Currently exploring (click to open)
- 🤖 Agent observability: what an agent did, and how to learn from its failures
- 🧠 Evaluation first: small test sets that tell me whether an AI tool is actually better
- 🎓 Learning science applied to quizzes, spaced review and study plans
- 🌐 Bilingual docs and tools (English and 中文) as a default, not an afterthought
Every star tells me someone found it useful, and that is what keeps me shipping every 3 days.
One click on any of these makes my day:
⭐ If something here helped you, a star means a lot. This repo was previously named MaxonT.