Blog Archive
Token Budgets per Agent Run: Allocating, Tracking, and Enforcing
Manage token budgets across agent runs. Allocate, track, and enforce token limits to control costs and prevent runaway agent loops.
Designing Tool Schemas That Models Won't Misuse
Design tool schemas that prevent LLM misuse. Use clear descriptions, strict type validation, and defensive patterns to keep agents on track.
Prompt Injection Defense for Tools: Context Separation and Tainting
Defend against prompt injection in agent tools using context separation and data tainting. Prevent tool outputs from hijacking agent reasoning.
Queue-Driven Agents: Feeding Workloads from Kafka into a LangGraph Fleet
Build queue-driven agent systems with Kafka and LangGraph. Process agent workloads at scale with guaranteed delivery, ordering, and fault tolerance.
LangGraph for Production: State Machines, Checkpointing, and Subgraphs
Take LangGraph to production with state machines, checkpointing, and subgraphs. Build reliable, observable, and scalable agent workflows.
Clinical Documentation Assistants: Building a HIPAA-Aware Agent Pipeline
Build HIPAA-compliant AI agents for clinical documentation. Handle PHI securely, maintain audit trails, and automate medical note generation with guardrails.
A Guide to Building a Checkout Agent on Stripe Payment Links
Build an AI-powered checkout agent using Stripe Payment Links. Automate product recommendations, handle cart management, and guide users through payment.
Agent Change Management: Peer Review, Staging, and Rollback for Prompt Updates
Implement change management for AI agents. Review prompt updates in staging, test against evals, and rollback safely when production behavior degrades.
Distributed Tracing for Multi-Agent Systems
Implement distributed tracing for multi-agent systems using OpenTelemetry. Track tool calls, reasoning chains, and inter-agent communication across your fleet.
Approval Gates: Pausing Agent Workflows for Human Sign-Off
Implement approval gates in agent workflows. Pause execution before critical actions, route to human reviewers, and resume with approval or rejection.
Agentic Retrieval: Letting the Agent Decide What to Fetch
Move beyond static RAG with agentic retrieval. Let agents decide what to search, when to stop, and how to combine information from multiple sources.
Golden Trajectories: Curating Human-Approved Agent Runs
Build golden trajectory datasets by curating human-approved agent runs. Use these validated paths to fine-tune agents, create evals, and improve reliability.