← Field Guide  ·  Tier 1 · Foundational Industry & Research

Anthropic

Building Effective Agents (Schluntz & Ingham, Dec 2024)

Foundational · blog8 patterns
In Plain English

This page shows every pattern that Anthropic has published or written about — 8 in total. Each one includes a simple picture of how it works, a short explanation, an example of when to use it, and its trade-offs (the good parts and the not-so-good parts).

Reading patterns from one source at a time matters because it lets you see how a single company or team thinks about building AI agents, and compare their ideas side by side. If a pattern here also appears in the bigger Deep Catalogue, you will see a link so you can read the full, detailed version.

An open reference book with a pencil in the fold under warm lamp light

1A.01Augmented LLM

AUGMENTED LLM User LLM Retrievalknowledge Toolsactions Memorystate

The baseline building block: an LLM given retrieval, tools, and memory. All higher patterns compose from this primitive.

Use caseAny agent at all — this is the atomic unit.
Trade-offsSimple to reason about; limited on its own for multi-step work.

1A.02Prompt Chaining

PROMPT CHAINING LLM 1step a Gate LLM 2step b Gate LLM 3step c

Decompose a task into a fixed sequence of LLM calls, each operating on the previous output, with programmatic gates between steps.

Use caseContent pipelines (outline → draft → edit), ETL-like LLM workflows, tasks with stable structure.
Trade-offsPredictable cost/latency, easy to debug · Cannot adapt to unexpected inputs.

Cross-ref: Deep Catalogue ORC-PLN-01 →

1A.03Routing

ROUTING Input Classifier Lane A · Specialist Lane B · Specialist Default / Escalate

Classify the input first, then dispatch to a specialist prompt or agent — rather than asking one generalist to handle every case.

Use caseCustomer support (refund / product / complaint lanes), inbox triage, multi-intent assistants.
Trade-offsSpecialist quality per lane · Classifier errors cascade.

Cross-ref: Deep Catalogue ORC-DEC-01 →

1A.04Parallelisation — Sectioning

PARALLEL — SECTIONING Dispatchersplits task Section 1∥ Section 2∥ Section 3∥

Split an independent task into sub-tasks, run concurrently, aggregate programmatically.

Use caseDocument analysis across sections, independent fact-checks, multi-source gathering.
Trade-offsLower wall-clock latency · Multiplicative token cost.

Cross-ref: Deep Catalogue ORC-ACT-01 →

1A.05Parallelisation — Voting

PARALLEL — VOTING Taskreplicate Run 1vote Run 2vote Run 3voteMajority / best-of-N aggregation.

Run the same task multiple times; aggregate by majority, best-of-N, or consensus.

Use caseSafety-critical classification, code review where false negatives are costly.
Trade-offsAccuracy lift on classification · Same-prompt-N-times multiplies cost.

1A.06Orchestrator–Workers

ORCHESTRATOR–WORKERS Orchestratorplan + compose Worker A Worker B Worker C

A central LLM dynamically decides which specialist workers to invoke and how to compose their results.

Use caseResearch agents, complex customer workflows, enterprise automation spanning domains.
Trade-offsScales past single-agent limits · Coordination overhead, orchestrator bottleneck.

Cross-ref: Deep Catalogue ORC-PLN-02 →

1A.07Evaluator–Optimiser

EVALUATOR–OPTIMISER Generatorgen ⇄ eval Draft Evaluator Critique Revisegenevalcritiquerevise

Generator produces; separate evaluator critiques against criteria; loop until quality met.

Use caseCode generation, copy under brand guidelines, translation, any output with programmatic quality criteria.
Trade-offsMeasurable quality lift · Doubled cost minimum, evaluator is new bottleneck.

Cross-ref: Deep Catalogue ORC-RFL-01 →

1A.08Autonomous Agent

AUTONOMOUS AGENT Agent LLM Plan Act Observe Stop?planactobservedone?

Open-ended loop: LLM plans, acts with tools, observes, and decides when to stop — without a predetermined script.

Use caseOpen-ended exploration, troubleshooting, research tasks with undefined scope.
Trade-offsMaximum flexibility · Hardest to bound cost and correctness.
Also in Tier 1
Google Cloud 12Antonio Gulli 17Academic Foundations 17