liveBook · Manning

Pattern Selection Card

As explained in Designing AI Agents

A book-specific aid for choosing among recurring agent execution topologies.

The Pattern Selection Card turns a business problem into a first-pass agent architecture worth testing. It maps cognitive-function demand and execution topology to candidate patterns; it does not automatically select or certify a design. section 2.3.5 section 10.6

Read the card from left to right:

  • ASSESS: Rate each cognitive function’s demand as None, Light, or Heavy. Record the task completion contract, smallest baseline, and first deviation. section 2.3.5 section 10.6
  • ROUTE: Use the binding deficit, dependency shape, collaboration demand, and time budget to choose processes and a single-agent or multi-agent topology. section 2.3.5 section 10.6
  • SELECT: Use the function-topology coordinates to choose candidate patterns. Record consequential seam contracts, budgets, and a removal condition. section 2.3.5 section 10.6
Table 10.6 Pattern selection card
Panel
Decision content
Required evidence
ASSESS
Task Completion Contract; seven-function demand rating; smallest baseline and first deviation
Bound workload, baseline run, external acceptance gate
ROUTE
Binding deficit; dependency shape; collaboration demand and time budget; process and topology choice
Failure trace, topology preconditions, rejected alternative
SELECT
Candidate coordinates and patterns; consequential seam contracts; budgets and removal condition
Pattern preconditions, evaluation plan, disconfirming signal, rollback or recovery rule

The card pairs each panel’s decisions with the evidence needed to support them. section 10.6

Argus code review

For a read-only code review agent, ASSESS rates Perception, Reasoning, and Reflection Heavy; Governance is Light, and Memory, Action, and Collaboration are Light or None. ROUTE favors a single agent, with Route or Orchestrate as candidates. SELECT proposes Context Triage, Complexity-Based Routing, and Generator-Critic as a testable starting architecture. section 2.3.5

Treat the card as an architecture hypothesis, not an acceptance record. A separate evaluation receipt records whether the proposal passed its experiment. Test the composition against the baseline, including its seams, and remove a mechanism if the same workload passes without it. section 10.6

Compare token costs only among candidates meeting the same required outcome, authority boundary, and acceptance evidence. When those conditions are equivalent, prefer the design that achieves the required quality with fewer tokens. section 2.3.5

How Designing AI Agents covers it

How deep it goes
Specifies three decision panels and their evidence requirements, while leaving candidate generation open rather than providing an automatic selector.
Examples
Uses a read-only code review agent to demonstrate demand ratings, topology choice, and candidate patterns.
What you can do after
Practice/perform a first-pass architecture selection and evaluate the proposal against a baseline and acceptance gates.
How it connects
Connects cognitive functions and execution topology to pattern selection, evaluation, and token cost.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Designing AI Agents. The section links point to the parts of the book it draws on.