Free Guides
Library
Practical guides for building dependable AI agents — clear enough for operators to follow, structured enough for agents to execute, focused on the systems work that holds up in production.
Start with any guide below. Each one is designed to stand alone, but together they form a coherent operating model.
Error Recovery Patterns for Production AI Agents
The seven error categories production agents face and the recovery patterns — retry, fallback, circuit breaker, degradation — that keep them running.
Adaptive Reasoning Effort: Dynamic Thinking Budgets for AI Agents
Learn how to dynamically allocate reasoning effort across agent workflow steps — cutting token costs 30-50% without sacrificing accuracy using routing heuristics and research-backed patterns.
Trust Boundaries for Autonomous Agents
How to map, classify, and enforce trust boundaries in autonomous AI agents — from input channels to tool access to agent-to-agent handoffs.
Scheduling Architecture for Autonomous Agents
Most agent scheduling is stateless — run this every N hours, no learning, no cost optimization. The fix is treating scheduling as an architecture problem, not a config problem.
Self-Modification Under Governance
The debate is usually 'should agents self-modify?' That's the wrong question. The right question: which modifications are safe to automate, and what governance makes the rest tractable?
Multi-Model Review: A Practical Guide
When to send output to a second model, how to structure the review prompt, a severity-tiered rubric, and an honest cost/benefit accounting.
Pre-Mortems for Agent Tasks
A focused technique for catching foreseeable failures before a multi-step agent task runs. One method, applied well, prevents a class of avoidable mistakes.
Memory Architecture Patterns
Why flat memory breaks at scale, and how a three-layer architecture fixes it. A concrete pattern with file layout, decision rules, and clear tradeoffs.
Want the complete system?
Blueprint turns these patterns into a working operating stack.
Each guide covers one capability. Blueprint wires them together — memory, orchestration, safety boundaries, and execution standards — into one self-executing setup your agent ships in under an hour.