field notes/substrate level

Agent Patterns

Understand the concepts and patterns of AI engineering at the substrate level: LLMs, loops, tools, and agents as they actually work. For technical readers who want to reason clearly, not just use abstractions.

single callloop

same weights, more passes

Index

the substratea stateless model. no memory between calls.

The Loop→

PATTERN 01/CONTROL FLOW

A stateless call invents an API that does not exist. The same model, in a loop, ships a verified fix.

Tools→

PATTERN 02/ACTION SURFACE

One tool that can do anything leaks a live secret. Four tools scoped to the task cannot.

Context→

PATTERN 03/WORKING SET

Uncompacted, a transcript re-proposes the fix it already ruled out three tool calls ago. Curated, it remembers.

Verification→

PATTERN 04/GROUND TRUTH

Unverified, the loop ends the moment the model says all tests pass. Verified, it doesn’t stop until they actually do.

Memory→

PATTERN 05/PERSISTENT STATE

Stateless between sessions, the agent asks the same clarifying question every morning. Given memory it can write to, it asks once.

Planning→

PATTERN 06/GOAL DECOMPOSITION

Improvised step by step, the agent calls a ten-file refactor done after step three. Planned up front, every step stays accountable to the goal.

DelegationComing soon

PATTERN 07/TASK DELEGATION

OversightComing soon

PATTERN 08/HUMAN OVERSIGHT

your agent