Advanced Autonomous AI Agents: Memory, Collaboration and Evolution in Action
summary
Builds .NET agents with Microsoft Agent Framework wired to models deployed in Azure Foundry, plus evals, tracing and an MCP-based tool server; shares a runnable repo with demos. Shows workflows (writer + critic loop), middleware hooks, scratchpad memory patterns and longterm memory tradeoffs, reports surprise failures like a learned system prompt being overwritten and context blowups up to 70k tokens. Compares models (Gemini 3, GPT‑5.1, Grock, FI series) and finds tool‑calling and memory management the real engineering pain.
AI-written summary · about
chapters (28)
- 0:00Welcome
- 1:13Autonomous AI agents in 2025
- 3:45Challenges in agentic systems
- 5:32Microsoft Azure AI Foundry
- 8:03Deploying models and agents
- 12:20Building a simple AI agent
- 14:06Agent memory with threads
- 15:31Adding tools to agents
- 17:44Integrating MCP tools
- 19:11Built-in tools in Azure AI Foundry
- 22:03Middleware for agents
- 24:13Metacognition in agents
- 27:26Evolving agents via prompts
- 29:33Agent workflows for tasks
- 34:54Latest AI models for agents
- 36:23Model selection for logic
- 41:50Cost and token efficiency
- 42:34Models in Azure AI Foundry
- 44:25Memory challenges in agents
- 46:34Context management issues
- 48:01Mixing models for performance
- 49:04Small agents vs large models
- 50:31Web browsing and engineering
- 51:34Economic data limits
- 52:37Math and memory in agents
- 54:04Summarization and temporal memory
- 55:11Prompt injection risks
- 56:15Demo and resources
related talks
39mEvolving Intelligence: Designing Autonomous AI Agents
26mAI powered automation & multi-agent orchestration
05h 17mBuilding AI Applications
36mBuilding AI Agents in .NET with Azure AI Foundry
25mBuilding AI Agents
46mAI Agents for the Rest of Us
29mPairing with the machine: How AI Agents are rewriting the rules of software development
56mBuilding AI Agents and Multi-Agent Systems with Azure AI