Build Retrieval That Can Explain Its Sources
Design retrieval around authorized evidence, stable document revisions, useful context, and claim-level provenance instead of a larger context window.
Read the guideThe engineering field guide
Build agents.
Engineer reliable
behavior.
Practical patterns for predictable behavior in systems that reason.
Explore the field guide
- 01Observe
Capture context.
Define the input. - 02Propose
Generate a plan.
Make constraints explicit. - 03Validate
Check the evidence.
Review the action. - 04Commit
Execute with durability
and traceability.
Systems
Patterns
Tools
Better outcomesDiscipline turns
capability into value.
Patterns
Tools
Better outcomesDiscipline turns
capability into value.
Featured guide
Deterministic Patterns for Agentic Applications: A Practical Guide
Build reliable agent systems with validated proposals, pure state transitions, durable observations, scoped approvals, and transactional execution intents.
Agent architecture16 min readLatest articles
Choose a Workflow Before You Choose an Agent
Match autonomy to the actual uncertainty in a task, with a practical decision process for workflows, bounded agents, and human escalation.
Read the guideDesign Tool Contracts an Agent Can Use Correctly
Make tools discoverable, narrow, and recoverable with explicit schemas, authorization boundaries, useful results, and honest failure semantics.
Read the guideMake Retries Safe Before You Automate More
Design operation identities, retry budgets, and reconciliation paths that prevent an interrupted request from becoming duplicate work.
Read the guideObserve Agent Runs and Enforce Useful Budgets
Trace the complete task, separate waiting from work, and enforce cost and latency limits without losing track of uncertain side effects.
Read the guidePut Approval at the Point of Commitment
Bind human approval to a concrete action, authenticated identity, and resource version so execution still matches what was reviewed.
Read the guide