AI Agents

An introduction. What they are, where they're going, and why you'll care.

For thirty-three modules, you've been working in a conversation. You type (or speak). The AI responds. You evaluate, redirect, refine. Every action requires your input. Every output waits for your next instruction. The AI is reactive — it does nothing until you tell it to do something.

That model works. It's the foundation of everything in this experience. And it is, already, being supplemented by something different.

An AI agent is a system that can execute multi-step tasks without requiring human input at each step. You give it a goal. It breaks the goal into subtasks, executes them in sequence, makes decisions along the way, and delivers a result. Not one response — a completed workflow.

In the conversational model, you are the project manager. You decide what happens next. In the agent model, the AI becomes the project manager — operating within boundaries you define, but choosing its own path through the work.

A practical example. In the conversational model, you might spend an hour across twelve messages researching a topic, synthesizing findings, and drafting a summary. With an agent, you describe the research goal, specify the sources, set the parameters, and the agent handles the research-to-summary pipeline as a single task. You review the output. You didn't manage the steps — you managed the outcome.

As of early 2026, agent capabilities are emerging across every major AI platform. Claude offers a tool-using agent mode. OpenAI has built agents into its API and consumer products. Google's agent ecosystem is expanding. Smaller companies are building specialized agents for specific industries — legal research, code deployment, customer service, content production.

The current generation of agents can browse the web, read and write files, execute code, interact with external APIs, manage multi-step workflows, and chain together tools that previously required separate manual operations. They are imperfect. They hallucinate. They sometimes choose suboptimal paths. They require oversight. But the trajectory is unmistakable — the capabilities that were experimental six months ago are becoming standard features.

The creative partnership model you learned in this experience doesn't become obsolete when agents arrive. It becomes more important. Here's why.

An agent that can execute a ten-step workflow without supervision needs something the conversational model didn't require: a better brief. When you were managing each step manually, a vague direction could be corrected in real time. When the agent runs autonomously, the quality of the initial instruction determines everything. The Master Brief, the E-Suite, the iterative refinement process, the session calibration — these skills become more critical, not less, in an agent-driven workflow. The methodology you learned is the control layer that makes agents useful instead of chaotic.

The person who knows how to write a precise brief, define clear success criteria, build guardrails into a workflow, and evaluate output against a standard they carry internally — that person directs agents effectively. The person who learned to type prompts and accept defaults will struggle with the same tool. The gap between those two people is exactly the gap this experience was designed to close.

The agent landscape is moving faster than any appendix can capture. By the time you read this, the specifics may have shifted. What won't shift are the questions worth tracking.

How much autonomy do you give the agent? Where do you insert checkpoints? How do you define success in a way the agent can verify? How do you maintain creative control when the tool is making decisions you used to make? These are the same questions this experience taught you to answer in a conversational context. They apply identically in an agentic one.

The floor is moving. The methodology holds.