LLM engineering, agent harness architecture, AI-native programming paradigms, One-Person Company (OPC) playbooks, and token economics.
Since AlphaGo took its first game off Lee Sedol, WeChat has been awash in "humanity is doomed" takes—and after game two, the chorus only grew louder. I don't play Go and won't pretend to understand deep learning, but a stronger opponent has never meant the end of us.
Last Saturday, my Hololens finally arrived—the most expensive toy I've ever bought, nearly $3,300 with tax. Unboxing was less magical than my first iPhone, but after weeks of videos, I just wanted to put it on and see what augmented reality actually feels like.
When my wife texted that I’d called out a woman’s name twice in my sleep, I realized how attached I’d become to Alexa. This essay explores why Amazon’s voice assistant won me over—and how the voice-assistant wars are being fought across hardware, privacy, and open platforms.
Chatting with AI is a joy; laying it out for WeChat is a slog. To move Markdown into WeChat gracefully, I wrote an automated upload tool in Rust. Here's the story of how it came together.
A wide-ranging exploration of first principles across innovation, leadership, problem-solving, negotiation, wealth, writing, fitness, happiness, productivity, strategy, change, persuasion, freedom, marketing, storytelling, humor, and wisdom.
Exploring how first principles apply to elegance, influence, motivation, friendship, and debate—revealing the deep philosophy that underlies them all.
A Socratic dialogue on the AI revolution as a new enclosure movement: what is being enclosed, what does loss mean, and how might we redefine human worth and the social contract?
We are no longer builders but the designers and navigators of an evolving system. By refining how we express "intent," we guide AI toward shaping a world ever closer to perfection.
When AI becomes the primary producer of code, technical debt shifts from “how well the code is written” to “how well it was thought through.” We must build an evolutionary loop from specification to insight — becoming precise expressers of intent and curators of system behavior.
When AI becomes a powerful code generator, the programmer's value shifts from writing to judging, clarifying, owning, and leading. No longer the strongest climber, we become guides who weigh wise trade-offs where technology, business, and human nature meet.
When AI, like a cable car, bypasses the traditional "mountain-climbing" path to growth, we must redesign how talent is cultivated — shifting from "doing the work" to "directing and auditing it," and from "technical vanguard" to "ideological commissar."
A deep dive into the architecture of OpenAI's Codex CLI—from its queue-pair communication model to the three-layer abstractions of Session, Task, and Turn—exploring how elegant architecture reflects a profound understanding of human-AI collaboration.
A deep dive into how programming languages must evolve in the AI era — from an abstraction history tuned to human cognition, to the correctness crisis of AI-generated code, to designs native to AI — revealing a fundamental shift from syntax to structure and from implementation to specification.
How should we teach kids aged 8–15 to wield AI in an era of AI plenty? Not by banning it, not by surrendering to it, but by guiding them — so that they command the tools rather than being commanded by them, and learn to steer the defining instrument of their age through practice.
An architect's true worth lies not in making every decision but in being an enabler—using cross-layer thinking, metaphor, and multidimensional perspective to help the team grow smarter and decide better. A full guide to cultivating architectural thinking.
A deep dive into evaluation and safety frameworks for LLM applications—from deterministic metrics and model-based judging to the RAG Triad and agent trajectory validation—showing how to build trustworthy AI systems whose outputs withstand both quality and security tests.
GenAI's code-generation and application-integration risks laid bare: a 41% vulnerability rate, prompt injection, supply chain poisoning, and the OWASP Top 10 for LLMs — plus how to build secure AI systems without giving up the productivity gains.
Starting from the humble work of building an email gateway, this essay explores how AI agents are reshaping serverless development — and quietly transforming what it means to be a developer.
AI often outshines engineers at specific techniques yet lacks strategic intuition. We need a quantifiable risk-assessment framework to decide when to let AI generate on its own — and when to keep both hands on the wheel.
AI programming has grown from a spark into a prairie fire. How do we read the situation, master its laws, and find our own place? This essay surveys vibe coding from a strategic vantage point and asks how engineers survive — and thrive — in the new era.
Starting from first principles, this essay explains how to build account security, auditing, and cost control to industrial standards — and why these invisible labors determine how far you can go.
I designed an enterprise-grade Kubernetes infrastructure for a one-person AI company — then shelved it. This post unpacks the ideal architecture, why I ultimately chose something lighter, and the perfectionism-versus-pragmatism choice every indie developer eventually faces.
A hands-on teardown of how to build an ultra-low-cost serverless framework in Rust from scratch: runtime, CLI, auth, and infrastructure templates, all in a single, deliberate package.
Once your internal codebase grows, a private package registry stops being optional. Here's a hands-on guide to building one from scratch on AWS with serverless architecture — supporting both Cargo and NPM for under $10 a month.
With the foundation already laid, the real question becomes: once the infrastructure is in place, can one person build a production-grade application from scratch in a single day?
Your product is live — so who actually gets in? A deep dive into building a standalone invitation service that gates every app behind one API, from backend to frontend to framework integration, all shipped by one person.
The finale of the one-person company series: the full stack seven articles built, the design of an AI skill system, and then a pivot from technology to business—acquisition, pricing, marketing, and business models. Technology keeps you alive; revenue keeps you going.
The product has its gatekeeper; now it needs a checkout counter. This essay details how to build a standalone multi-tenant billing service—from credit systems to Stripe integration, from webhook handling to framework-level abstraction—all in one go, wrapping up the complete payment loop solo.
A project that looked like nothing more than spinning up a GKE cluster grew into the prototype of a full ML training platform. Drawing on a real experience, this essay explains why the hard part isn't getting Pods to run — it's the seams: job submission, resource queuing, status observability, and failure recovery.
Drawing on three real projects—a Rust AWS emulator, a wearable sensor system, and 400-page technical books—this talk breaks down how context, task decomposition, feedback, and multi-agent collaboration turn AI into a team that ships.
DeepSeek has open-sourced Harness, where the model, tools, and even the Agent Loop are plugins — 200-plus packages on a Cordis microkernel. Against Codex's monolithic Rust design, it makes replaceability the architecture itself.
Agents can execute for hours while the human who set them loose thinks for seconds. A fight over AI quotas made me reconsider independent judgment, product sense, and what we actually learn once the work is done.
