Prompt Systems
Move beyond isolated prompts into reusable techniques, composed workflows and working sessions.
Explore solution →Books, tools, experiments and working solutions for agents, context, memory, language, retrieval and AI applications.
The books work through difficult subjects from first principles. The goal is not to stop at explanation: each path should lead toward something that can be built, tested or used.
Build AI memory from storage and retrieval through provenance, temporal state, unfinished work, context selection, behavioural evaluation, trust, and the boundary with learning — adding only the mechanisms that measurement earns.
Read 19 chapters → 36A first-principles investigation of how AI can transform language from the primary carrier of communication into one representation within a larger system of semantic mediation, attention, personalisation and delegated action.
Read 36 chapters → 27Context as a compiled, observable artefact: see what a model actually receives, learn what capacity, content, representation, time, authority and scope each demand, then build a deterministic Context Compiler, deliver its output to a runtime, observe that it arrived, and test whether it helped.
Read 27 chapters → 30Build from a callable model to explicit context, durable state, controlled actions, independent verification, and a deterministic runtime that decides what happens next.
Read 30 chapters → 26Explore what it means for information to become geometry, when that geometry can be trusted, and how representation, similarity, retrieval, calibration, cross-space alignment, and compression combine into an embedding runtime that knows its own limits.
Read 26 chapters → 25Build browser-native AI from the capability boundary upward: local models, observability, WebMCP tools, agent security, and a user-owned browser policy engine.
Read 25 chapters → 21Turn language-model behavior into an experimental variable, then learn when optimization is measurable, when improvement is trustworthy, and what DSPy contributes to the process.
Read 21 chapters → 19 Artificial life, complex systems and computational emergenceAn experimental search for life-like organization in computational systems — through emergence, causality, persistence, material turnover, hidden state and finite computation.
Read 19 chapters → 60Progress from deterministic debugging to diagnosing models, evidence, trajectories, and AI-generated work.
Read 60 chapters → 16Explore why language models hallucinate, how hallucination can be measured and evaluated, where individual detection methods fail, and how evidence, verification, policy, abstention, and memory gates can be combined to build reliable systems around stochastic models.
Read 16 chapters → 16Understand PyTorch from first principles — from tensor geometry and autograd to model structure, transforms, attention, debugging, performance, compilation, reproducibility, and a GPT-style model built from scratch.
Read 16 chapters → 13Learn to read modern AI models by decomposing them into smaller mechanisms, then rebuilding quality scorers, value and policy heads, recurrent reasoning systems, optimizers, and preference rankers in PyTorch.
Read 13 chapters → 11Build reliable AI agents by understanding the mechanisms underneath them: control loops, actions, validation, planning, state, tools, memory, search, and external verification.
Read 11 chapters → 47Go beyond basic agent loops into orchestration, evaluation, reliability, multi-step reasoning, memory, and production-grade agent systems.
Read 47 chapters → 32Think, create, research, write, and build with AI through open conversation, clear intent, structured reflection, iterative collaboration, and amplified creation.
Read 32 chapters → 59Build cellular automata from the smallest local rules, then follow them through measurement, computation, continuous artificial life, learned dynamics and reproducible engineering. A first-principles investigation of what simple local systems can do — and how to tell what the evidence actually supports.
Read 59 chapters → 12Explore how LLM agents grow from simple conversations into reflective, tool-using, memory-enabled and multi-agent systems—and how those systems can become collaborators, companions, interfaces, and extensions of human agency.
Read 12 chapters →The practical destination of the work: projects, diagnostics and tools that turn the underlying ideas into something usable.
Move beyond isolated prompts into reusable techniques, composed workflows and working sessions.
Explore solution →Small, focused applications that wrap useful AI workflows in an accessible interface.
Explore solution →Experiments and practical patterns for local models, browser-native tools and user-controlled AI.
Explore solution →An AI agent says: Done. The task is complete. That sentence is almost worthless. The agent may have: edited the wrong file, changed the right file incorrectly, skipped part of the …
Read →An AI agent often fails for a surprisingly ordinary reason: it commits too early. It finds one plausible next action, follows it, and then spends the rest of the run trying to make …
Read →AI Agent Forgets Previous Work? Add Working, Semantic and Episodic Memory An agent can use the right model, call the right tools, execute the right plan, and still behave as if …
Read →An agent can have a perfectly capable model and still behave badly because its tools are badly designed. This is one of the most common agent failures in production: user goal ↓ …
Read →An AI agent that keeps calling the same tool, revisiting the same page, rewriting the same file, or repeatedly saying “I’ll try again” is not displaying persistence. It is …
Read →A surprising number of agent failures are not really model failures. The model may be perfectly capable of writing each individual step. The failure happens because the system …
Read →