Wayland Zhang
AI researcher and serial founder. Building agent runtimes and quantitative trading systems — and writing about the engineering discipline that makes them work.
Previously founded a social advertising platform acquired by AdChina (Alibaba Group). Studied Computer Science at the University of Toronto ('03-'07), where I attended Professor Geoffrey Hinton's class.
I started sharing what I was learning about AI in 2024. The community grew to over 100,000 practitioners across Bilibili and TikTok. That momentum pushed me to write three technical books and build two open-source platforms.
waylandzhang[]gmail.com
Current Work
Agent runtimes
Making local and cloud agents reliable enough for delegated work: memory, permissions, tool use, replay, and observable workflows.
Systems writing
Turning production lessons into books and essays that connect model internals with the engineering discipline around them.
Reasoning research
Exploring where fixed Transformer loops strain: dynamic computation, recurrent structure, manifolds, and personal context.
Projects
Production-grade multi-agent platform with deterministic replay, budget enforcement, and enterprise observability. Built with Rust, Go, and Python.
AI agent runtime powered by Shannon — Mac file ops, shell, GUI automation — with complex task delegation via Shannon Cloud workflows.
AI-Driven Quantitative Hedge Fund
Featured Essays
All essaysThe Four Realms of Neural Networks
A cultivation-novel reading of deep learning: PDE solvers, manifold geometry, gauge fields, quantum attention, and personal AGI.
Chern's Question, Answered by Two Worlds at Once
How a question from manifold topology reappears in rock mechanics and neural networks, pointing toward the same deeper geometry.
The AI Agent Harness
What Kocoro, Claude Code, and Codex-style systems converge on when the harness becomes the product.
How Torn Is a Trained Mixture-of-Experts?
A measurement-driven look at whether MoE routing creates real geometric discontinuities inside released models.
Information Theory Is All You Need
The clean bridge from Shannon entropy to cross-entropy loss, compression, and the limits of model intelligence.
Research Themes
Production agents
Tool execution, memory, permissions, observability, and workflows that survive real users.
Model internals
Attention, tokenization, MoE geometry, compression, and why architectures behave the way they do.
Dynamic reasoning
Alternatives to fixed token loops: adaptive computation, recurrent structure, and personal context.
Memory Dynamics
The Tensor Logic memory model, recurrent state, and how durable personal context changes what agents can become.