Agent patterns
Agent pattern decks
These are short visual guides to designing AI agents. Each deck takes one problem, such as handing a task to an agent, previewing the agent鈥檚 plan, or keeping a register of agents, and works through it in ten to twelve slides that take a couple of minutes to read.
The decks draw on the agent patterns library, a set of forty-one interaction patterns for agent products. Each deck is a free PDF download, and you do not have to sign up for anything to get it.
- Download
Eight ways agent products lose trust
The deck covers eight failures that recur across released agent products and in the 2026 research. Its main recommendation is to have the agent hand back receipts, such as the plan it followed, screenshots, the sources it opened, and a note on what it could not verify.
- Download
The six stages of agent handover
The deck walks through the six stages of handing a task to an agent: brief, plan, run, check, verify and learn. It explains what each stage is for and names brief and verify as the two stages people notice are missing.
- Download
Five minutes before the agent starts
The deck covers five things to agree with an agent before it acts: the plan preview, the editable plan, the clarifying question, the agreed limits and the autonomy setting. It recommends starting with the plan preview, where the agent states in plain words what it will do.
- Download
Reading a crowd of agents
The deck covers what a board of parallel agent runs has to show. It recommends listing any run that is waiting on a person at the top.
- Download
Your newest teammate has a spend cap
The deck covers the agent register, which records each agent鈥檚 owner, who may give it tasks, what it is allowed to write, and its spend cap. It recommends starting by putting an owner on every agent.
- Download
Can an agent use your product?
The deck shows what an agent reads on a checkout page. It recommends publishing an llms.txt file as a low-cost way to find out whether agents already reach your product.