Everything Agents

Tutorials for building reliable agents.

Real code, real bugs, real fixes. An agent is a loop with tools and an LLM that picks its own next step; these tutorials are about making that pick predictable. Starts with Agentforce, Salesforce's agent platform, and grows from there. Free to read.

A frame for what's here

Most agent systems sit somewhere between a chatbot that decides everything and a scripted flow that decides nothing. The tutorials here live in that middle zone. The shorthand for it is guided determinism: let the model choose, but constrain what it can choose from. Reliable without going rigid.

The spectrum of agent controlAgents fall on a horizontal axis from free-roam, where the language model picks every action, to fully scripted, where every transition is a deterministic gate. The refund agent in this tutorial sits in the middle: guided determinism.FREE-ROAMLLM picks every actionFULLY SCRIPTEDevery transition is a gatechatbotRalph Wiggum loopguided determinismreliable agentIVR / form flowno LLM

Tutorials

  • Guided determinism for agents using AgentScript14 lessons · 73 min

    Combining LLM flexibility with deterministic gates in Agentforce. Built around one real refund agent.

  • Demystifying Agent Harnesses14 lessons · 74 min

    What's actually inside an agent harness, how the planner-tools-executor loop runs, and where each design knob lives. Coding agents are the running example; the patterns generalize.

In the queue

  • Vibe coding agentsCoding fluently with an agent in the loop while keeping the system reliable. The instincts and the guardrails.
  • Testing agentsWriting test specs, asserting on tool calls, and reading run results. What stays deterministic and what doesn't.
  • Observing agentsPulling session traces, parsing them, and turning behavior signals into actionable diagnostics.
  • Autotuning agentsClosing the loop: using observations to autotune prompts, gates, and tool choice without sacrificing correctness.

Sign up in the header to get a note when these go live.