Lilly Tech Systems
AI Agents
Master the craft of building AI agents end-to-end. Frameworks, architectures, capabilities, domain use cases, and production operations — everything you need to build agents that ship.
AI Agents is our deepest single track, which reflects our conviction that agentic systems are where the most interesting failure modes and the most valuable wins live right now. An agent is what you get when you combine a language model with tools, memory, and autonomy over multiple steps, and the interesting part is that each of those three ingredients adds an entire new class of risk, latency, and debugging challenge. Most agent projects fail not on the model choice but on how the tools are scoped, how memory is truncated, and how failures are recovered.
We deliberately cover the building blocks and the orchestration together because, in practice, you cannot reason about one without the other. A well-designed tool with a weak planner wastes the tool; a strong planner with a leaky sandbox is a liability. The lessons include patterns from shipped systems, common anti-patterns we have seen cost real money, and practical guidance on when an agent is the right answer versus when a deterministic pipeline would have been cheaper, safer, and easier to debug.
All Agent Topics
50 topics organized into 5 categories spanning the full agent stack — from frameworks to domain agents to production operations.
Agent Frameworks
LangGraph Agents
Build stateful, graph-based agent workflows with LangGraph. Master nodes, edges, conditional routing, persistence, and the patterns that power production agents at scale.
6 LessonsCrewAI Multi-Agent
Build collaborative multi-agent crews with CrewAI. Master roles, tasks, processes (sequential, hierarchical), and the patterns for teams of specialized agents.
6 LessonsMicrosoft AutoGen
Build conversational multi-agent systems with Microsoft AutoGen. Master GroupChat, AssistantAgent, UserProxyAgent, and the patterns for agent-to-agent dialogue.
6 LessonsLlamaIndex Agents
Build RAG-native agents with LlamaIndex. Master ReActAgent, FunctionCallingAgent, query engines as tools, and agent workflows over your indexed data.
6 LessonsPydantic AI Agents
Build type-safe, production-grade agents with Pydantic AI. Master typed dependencies, structured outputs, validation, and the FastAPI-style developer experience for agents.
6 LessonsHuggingFace Smolagents
Build code-writing agents with HuggingFace Smolagents. Master CodeAgent vs ToolCallingAgent, the code-as-action paradigm, and integration with HF models and tools.
6 LessonsOpenAI Assistants API
Build managed assistants with the OpenAI Assistants API. Master threads, runs, file search, code interpreter, and the patterns for production OpenAI agents.
6 LessonsClaude Agent SDK
Build agents on the Claude Agent SDK. Master tool definitions, tool result patterns, computer use, and the Anthropic-recommended best practices for production agents.
6 LessonsGoogle Agent Development Kit (ADK)
Build production agents with Google's Agent Development Kit. Master Gemini integration, tool use, evaluation, and deployment to Vertex AI Agent Engine.
6 LessonsOpenAI Swarm
Build lightweight multi-agent systems with OpenAI Swarm. Master agent handoffs, context variables, and the minimal-orchestration pattern for routing between specialists.
6 LessonsAgent Architectures
Agentic RAG
Add an agent layer on top of RAG to handle multi-step retrieval, query routing, and self-correction. Master corrective RAG (CRAG), self-RAG, and adaptive RAG patterns.
6 LessonsAgentic Workflows
Design deterministic workflows with agent steps. Learn when to choose a workflow over a fully autonomous agent and how to combine the two for reliability and flexibility.
6 LessonsStateful Agents
Build agents that remember across runs. Master state stores (Redis, Postgres), checkpointing, conversation state, and recovery patterns for resumable agent execution.
6 LessonsEvent-Driven Agents
Build agents that react to events from queues, webhooks, and message buses. Master Kafka/RabbitMQ integration, async patterns, and high-throughput agent pipelines.
6 LessonsHierarchical Agent Systems
Build manager-worker agent hierarchies. Master delegation patterns, task decomposition across layers, and the supervisor-subordinate models used in CrewAI and LangGraph.
6 LessonsSwarm and Mesh Agent Networks
Build peer-to-peer agent networks with no central orchestrator. Master agent discovery, gossip protocols, consensus, and emergent multi-agent behavior.
6 LessonsAgent Pipelines
Compose agents into linear pipelines. Master pipe-and-filter patterns, intermediate state contracts, and the testing strategies that keep pipeline agents reliable.
6 LessonsReflection and Self-Critique Agents
Build agents that critique and improve their own outputs. Master Reflexion, Self-Refine, and the meta-cognitive loops that boost agent quality 20-40% on hard tasks.
6 LessonsAgent Capabilities
Agent Code Execution (Sandboxes)
Give agents the ability to run code safely. Master Docker sandboxes, E2B, Modal, RestrictedPython, and the security patterns for executing agent-written code.
6 LessonsAgent Web Browsing
Build agents that browse and interact with the web. Master Playwright, Puppeteer, browser-use, and the patterns for reliable web automation by LLMs.
6 LessonsAgent Computer Use
Build agents that control a full desktop. Master Claude Computer Use, OpenAI Operator, screen capture, mouse/keyboard control, and the security model for OS-level access.
6 LessonsAgent File System Operations
Give agents safe file system access. Master read/write/edit tools, sandbox boundaries, glob patterns, and the patterns used by Cursor, Claude Code, and Aider.
6 LessonsAgent SQL and Database Access
Let agents query and update databases safely. Master text-to-SQL agents, schema introspection, query validation, and the read-only/write-mode boundaries for production safety.
6 LessonsAgent API Integration
Connect agents to external APIs. Master OpenAPI tool generation, MCP servers, authentication patterns, and rate limit handling for API-heavy agents.
6 LessonsVoice Agents
Build voice-enabled agents with real-time speech. Master OpenAI Realtime API, Deepgram, ElevenLabs, telephony integration, and low-latency conversational patterns.
6 LessonsVision Agents
Give agents eyes. Master GPT-4V, Claude vision, Gemini vision, document understanding, and the patterns for image-grounded multi-step reasoning.
6 LessonsLong-Term Agent Memory
Build agents that remember across days and months. Master Mem0, Zep, episodic vs semantic memory, memory consolidation, and the recall patterns that beat raw vector search.
6 LessonsAgent Knowledge Graphs
Give agents structured world knowledge with knowledge graphs. Master Neo4j, GraphRAG, entity extraction, and the patterns for graph-augmented agent reasoning.
6 LessonsDomain Agents
Customer Support Agents
Build production customer support agents. Master ticket routing, knowledge base RAG, sentiment-aware responses, escalation, and integration with Zendesk and Intercom.
6 LessonsCoding Agents
Build coding agents like Claude Code, Cursor, and Aider. Master file editing, codebase navigation, test execution, and the SWE-bench evaluation methodology.
6 LessonsResearch Agents
Build deep research agents that produce report-quality output. Master multi-step research, source verification, synthesis, and the GPT Deep Research-style architectures.
6 LessonsSales SDR Agents
Build AI SDR agents that prospect, qualify, and book meetings. Master lead enrichment, personalized outreach, multi-channel sequencing, and CRM integration.
6 LessonsMarketing Content Agents
Build agents that generate, schedule, and analyze marketing content. Master brand voice, multi-channel publishing, A/B testing, and analytics integration.
6 LessonsData Analyst Agents
Build agents that explore datasets, generate insights, and produce dashboards. Master pandas, SQL, chart generation, and the patterns for autonomous data analysis.
6 LessonsDevOps and SRE Agents
Build agents for incident response, deployment, and infrastructure ops. Master log analysis, runbook execution, Kubernetes operations, and integration with PagerDuty.
6 LessonsFinancial Analysis Agents
Build agents for financial research, modeling, and reporting. Master 10-K analysis, financial data APIs, model building, and compliance considerations for fintech agents.
6 LessonsLegal Research Agents
Build agents for legal research, contract review, and case law analysis. Master citation grounding, redlining, jurisdiction handling, and the patterns to avoid hallucinated case law.
6 LessonsHealthcare Triage Agents
Build healthcare triage and clinical-support agents. Master HIPAA-compliant pipelines, symptom assessment, EHR integration, and the safety patterns for clinical AI.
6 LessonsRecruiting and HR Agents
Build agents for resume screening, candidate sourcing, and HR workflows. Master JD-resume matching, interview scheduling, bias-aware screening, and ATS integration.
6 LessonsPersonal Productivity Agents
Build personal AI assistants for email, calendar, tasks, and notes. Master Gmail/Calendar integration, task management, daily briefings, and proactive agent patterns.
6 LessonsAgent Operations
Agent Deployment Patterns
Deploy agents to production. Master serverless agents, container deployment, Vertex Agent Engine, AWS Bedrock Agents, and the cost-vs-control deployment tradeoffs.
6 LessonsAgent Cost Optimization
Cut agent bills 50-90%. Master model routing per task, prompt caching for agent loops, iteration caps, and the cost-per-task metrics that drive optimization.
6 LessonsAgent Testing Strategies
Test agents the way you test code. Master unit tests for tools, integration tests for loops, snapshot tests for trajectories, and CI patterns that catch regressions.
6 LessonsAgent Observability
See what your agents do. Master traces, spans, metrics, and the observability stack (LangSmith, Arize Phoenix, Helicone) for production agent debugging.
6 LessonsAgent Tracing Deep Dive
Master agent tracing across frameworks. OpenTelemetry, OpenInference, custom spans, and the patterns to debug 100-step agent runs without losing your mind.
6 LessonsAgent Security Hardening
Harden agents against prompt injection, tool abuse, and credential theft. Master least-privilege tools, input sanitization, output filtering, and the agent attack surface.
6 LessonsAgent Rate Limiting and Quotas
Stop runaway agents from blowing your bill. Master token bucket rate limiting, per-tenant quotas, circuit breakers, and the patterns to fail fast under abuse.
6 LessonsAgent Versioning and Rollback
Version, ship, and roll back agents safely. Master semantic versioning for prompts, model + prompt + tool versioning, canary releases, and instant rollback patterns.
6 LessonsAgent A/B Testing
Run rigorous A/B tests on agent changes. Master traffic splitting, statistical significance for agent metrics, multi-arm bandits, and online evaluation patterns.
6 LessonsAgent Failure Recovery
Build agents that gracefully recover from failures. Master timeout handling, partial-result patterns, dead-letter queues, and the patterns that turn flaky agents into reliable ones.
6 LessonsWhy an Agents Track?
Agents are the fastest-moving frontier in AI engineering. This track focuses on the patterns that survive contact with production.
Framework-Agnostic
Learn LangGraph, CrewAI, AutoGen, LlamaIndex, Pydantic AI, Claude SDK, and more — pick the right one for each problem.
Production-First
Every topic emphasizes deployment, observability, security, and cost — not just the demo loop.
Domain-Specific
Concrete patterns for support, coding, sales, analyst, devops, legal, healthcare, HR, and personal-productivity agents.
Capability Coverage
Code execution, web browsing, computer use, voice, vision, long-term memory, knowledge graphs — the full capability surface.