Gunnar Grosch
Principal Developer Advocate AWS | Engaged to the wonderful @arcticf0xx.com | Opinions are my own | gunnargrosch.com
- Happy Friday, everyone! We're celebrating with proper Italian kebab pizza and proper Italian pineapple pizza! 🍕
- Reposted by Gunnar GroschExcellent insights from Ben Kehoe (@ben.kehoe.io) via Gunnar's (@gunnargrosch.com ) summary with the full talk video in embedded. I would encourage you to have a read and watch the talk. builder.aws.com/content/37A0...
- 🎯 DEV Track Spotlight: Breakout Sessions & Code Talks from AWS re:Invent 2025 I've covered 27 incredible sessions featuring AWS Heroes and speakers from AWS. From multi-agent AI systems to serverless, Rust performance to legacy modernization. #AWSreInvent #DEVTrack #AWSCommunity #AWSHeroes
- 60 unique sessions, 93 speakers covering: GenAI & Agentic AI (Strands Agents SDK, Amazon Bedrock), Developer Tools (Kiro, Q Developer), Security, Infrastructure (serverless, edge), Modernization, and Data (DSQL). Real implementations, production patterns, lessons learned from practitioners.
- These aren't vendor pitches - they're practitioners sharing real code, real demos, and real learnings from building production systems. AWS Heroes solving actual problems with hard-won experience. 📝 Full summary with all 27 Breakout Sessions & Code Talks: builder.aws.com/content/37r7...
- Part 27 of DEV Track Spotlight 🤖 Together with @readysetcloud.io, I demonstrated how to build production-ready multi-agent AI systems at AWS re:Invent (DEV415). The key? Treat agents like distributed systems with A2A protocol, MCP for typed interfaces, and RISEN prompts for deterministic behavior.
- Best demo moment: Allen's "destroyed package" scenario where contents caught fire, got run over by multiple cars, then the ashes caught fire again. 🔥🚗 The agents handled it autonomously - no predefined rules needed. That's the power of multi-agent choreography.
- Full code samples, RISEN prompts, and SAM template available on GitHub. 📖 Deep-dive: builder.aws.com/content/37lC... 🎥 Session: www.youtube.com/watch?v=9O9z...
- Part 26 of DEV Track Spotlight 🚀 Shridhar Pandey (AWS) and I built a complete serverless image generation app - backend, frontend, observability - using Kiro CLI. Session: Serverless Full-Stack in Action: AI-Driven Developer Experience #AWSreInvent #DEVTrack #AWSCommunity
- We used Kiro CLI with AWS Serverless MCP server. The MCP server transformed Kiro from just suggesting CLI commands to actually executing them with tools like sam_init, sam_build, sam_deploy. Built: API Gateway + Lambda + Bedrock + S3 + React frontend + CloudWatch observability
- Key takeaways: Break architecture into steps, use steering files for control, security by default, observability is integral, seamless console-to-IDE transitions. Shridhar's advice: "Don't be shy with your prompts. The more specific, the better!"
- 📝 Full deep dive: builder.aws.com/content/37l3... 📺 Watch: www.youtube.com/watch?v=168m...
- Part 25 of DEV Track Spotlight 🤖 DEV332 with Joy Chakraborty (AWS) & Elizabeth Fuentes Leone (AWS) "Simple things should be simple to build and deploy" - building multimodal AI agents with Strands Agents SDK & Amazon S3 Vectors #AWSreInvent #DEVTrack #AWSCommunity
- Strands Agents SDK: 3M+ downloads, minimal code for production agents. Demo showed analyzing architecture diagrams, docs & 6-min glacier video with scene-by-scene summaries. MCP integration for remote tools, conversation management with sliding window/summarization strategies.
- Amazon S3 Vectors: First cloud object storage with native vector support, billion-vector scale, subsecond latency. Travel assistant demo remembered user preferences across sessions using metadata filtering. Elizabeth's tip: "Describe tool capabilities, not service names!"
- Part 24 of DEV Track Spotlight 🦀 DEV405 with @darko.rup12.net (AWS Principal Developer Advocate) "With the world of LLMs, with coding assistants, with everything at your fingertips, you don't need to write Python anymore. You can write Rust." #AWSreInvent #DEVTrack #Rust
- Live demo highlights: • To-Do MCP Server with type-safe parameters • Thermal receipt printer integration using recibo crate • Hardware control - LLM analyzes code, creates tasks, physically prints them on receipts! Rust MCP servers: <1 second startup vs 4+ seconds for TypeScript
- Key tech: rmcp crate with #[tool_router], #[tool_handler], #[tool] macros. Standalone binaries, zero dependencies, cross-platform. Production examples: IAM Policy Autopilot (AWS open source), Grimoire MCP (code pattern storage)
- 📝 Full deep dive: builder.aws.com/content/37fJ... 📺 Watch: www.youtube.com/watch?v=mr7Z...
- Part 23 of DEV Track Spotlight 🤖 DEV331 with Erik Hanchett & Du'An Lightfoot (AWS Developer Advocates) "Agents still need compute, storage, databases. All the knowledge you have is still valuable up until this point and beyond." #AWSreInvent #DEVTrack #AWSCommunity
- Multi-agent food tracker demo: user-facing agent, orchestrator, 2 MCP servers (recipe + nutrition via Gateway). Built with FastMCP in Python. Converted existing Lambda functions to MCP servers without code changes. Memory demo showed user preferences persisting across sessions!
- Key topics: Amazon Bedrock AgentCore (serverless runtime, 8-hour workloads), Memory (semantic search, user preferences), Gateway (semantic tool search saves tokens/costs), Kiro (spec mode + vibe mode), full CloudWatch observability Du'An: "Start with observability from the beginning!"
- 📝 Full deep dive: builder.aws.com/content/37c1... 📺 Watch: www.youtube.com/watch?v=oZWj...
- Part 22 of DEV Track Spotlight 🤖 DEV403 with @jameshood.dev (AWS) & @maishsk.com (AWS) "I am also a former AI skeptic. My previous experience with AI was trying flashy demos on real world code and watching it crash and burn." Then he discovered Agent SOPs. #AWSreInvent #DEVTrack #AWSCommunity
- Agent SOPs = structured markdown docs that guide AI through complex tasks with RFC 2119 constraints (MUST, SHOULD, MAY) Amazon has thousands in production. The brilliant part? Use AI to write them. Real workflow: PDD → Research → Design → Code Task Generator → Code Assist (TDD with 97% coverage)
- Key distinction: "Vibe coding is amazing for prototypes, not very good for production code." Active participation required - read research, review designs, understand implementations. AI accelerates, doesn't replace engineering judgment.
- 📝 Full deep dive: builder.aws.com/content/37Zh... 📺 Watch: www.youtube.com/watch?v=cf-W...
- Part 21 of DEV Track Spotlight 🚀 DEV304 with Elizabeth Fuentes Leone (AWS) & Rossana Suarez (AWS Container Hero, Naranjax) Werner Vogels' most famous quote: "Everything fails all the time." Do you find out from dashboards or angry customers? #AWSreInvent #DEVTrack #AWSCommunity #AWSHeroes
- AI-powered observability in CI/CD with 3 intervention points: • Pull Request - risk assessment • Pre-Deployment - AI blocks unsafe deploys (critical safety gate!) • Post-Deployment - validates success Results: 200 alerts → 5, MTTR 2hrs → 15min (8x faster) 📉⚡
- Live demos showed: 🤖 Amazon Bedrock with multiple models 🔧 Strands Agents SDK ☸️ Amazon EKS + Prometheus/Grafana 📱 Telegram notifications 🚀 GitHub Actions integration AI blocked bad deployments automatically with detailed explanations!
- Rossana: "AI is a tool. It makes you stronger, faster, better. Don't be afraid of AI. Use it and you will be successful." Everything is open source and available today! 📝 Read the full deep dive: builder.aws.com/content/37Wv... 📺 Watch: www.youtube.com/watch?v=TsKK...
- Part 20 of DEV Track Spotlight 🗄️ DEV315 with Namrata H Shah (AWS Community Hero) & Rob Koch (AWS Data Hero) "When databases can discover, recommend, and optimize on their own, they stop being reactive systems and become intelligent partners." #AWSreInvent #DEVTrack #AWSCommunity #AWSHeroes
- Covered 4 AWS database services: • RDS - Query optimization, instance rightsizing, storage optimization • Redshift - Distribution keys, skewed queries, predictive scaling • Aurora - Replication lag, connection management • DynamoDB - Hot partitions, throttling prevention, cost optimization
- Demo with Kiro CLI & MCP servers showed agents analyzing week of data, recommending T3 medium → T3 small downsize, forecasting traffic spikes before they happen. Start small, measure results, then scale!
- 📝 Full deep dive: builder.aws.com/content/37Ty... 📺 Watch: www.youtube.com/watch?v=3MA5...
- Part 19 of DEV Track Spotlight 💰 DEV318 with Steph Gooch (AWS) & Kenneth Attard (AWS Community Hero, Betsson Group) "It's Friday morning, you've had an easy week... And then you get this email, your AWS budget has gone off" #AWSreInvent #DEVTrack #AWSCommunity #AWSHeroes
- Real optimizations from Betsson targeting $100K savings: • Graviton migration (EC2/RDS/Lambda) • S3 gateway endpoints (free!) • Centralized NAT gateways (breakeven at 10 VPCs) • Regional optimization (N. Virginia saved $2,438/month vs São Paulo) • WAF Challenge vs CAPTCHA ($0.40 vs $4 per 10K!)
- AI-powered tools: AWS Cost Optimization Hub, Amazon Q Developer optimization feature, Kiro CLI with MCPs for automated implementation Prevention: Service Control Policies + Context files = optimize once, apply forever
- 📝 Full deep dive: builder.aws.com/content/37RE... 📺 Watch: www.youtube.com/watch?v=vvdj...
- Part 18 of DEV Track Spotlight 🤖 DEV313 with Haowen Huang (AWS) & Cyrus Wong (AWS AI Hero) "By the 2030s, we will see an agent economy operating like a global neural network of interconnected agents." Voice-controlled humanoid robots, robot dogs & drones! #AWSreInvent #DEVTrack #AWSCommunity
- Real production system from Hong Kong: • Amazon Nova Sonic - Stream-in/stream-out for natural, interruptible robot conversations • MCP servers in Lambda - Translate voice commands to robot actions • Strands Agents SDK - Handles complex multi-step reasoning automatically
- Key topics: Serverless architecture (IoT Core/Lambda/Bedrock), prompt engineering with Enum types, parallel execution for multiple robots, Kiro CLI for code generation Cyrus: "Use frameworks. Don't be lazy. Simple framework will already include solutions!"
- 📝 Full deep dive: builder.aws.com/content/37OF... 📺 Watch: www.youtube.com/watch?v=ZKqV...
- Part 17 of DEV Track Spotlight 🎲 DEV330 with Olivier Leplus & Tiffany Souterre (AWS) Building a multi-agent AI Game Master with Strands Agents SDK - live coding that made complex agent orchestration look simple! #AWSreInvent #DEVTrack #AWSCommunity
- The demo featured 3 specialized agents: • Character Manager - Creates/manages characters with DB access • Rules Agent - RAG system with D&D rulebook • MCP Dice Server - Handles dice rolls Key protocols: MCP (Model Context Protocol) & A2A (Agent-to-Agent) for tool discovery & agent communication
- Live coding highlights: Built agents in just a few lines of code, demonstrated model flexibility (Amazon Bedrock, Nova Premier, Anthropic, Mistral, Ollama), created MCP servers, implemented A2A communication Patterns apply beyond gaming - customer support, troubleshooting, research!
- 📝 Full deep dive: builder.aws.com/content/37LE... 📺 Watch: www.youtube.com/watch?v=UxjP...
