In January I used AI-DLC alone, one developer and one agent in an IDE. This month I deployed AIDLC Collaborative, the open-source sample that runs the same method for a whole team: intents, human gates, parallel construction lanes, agents in isolated AgentCore sessions, and a platform that owns git. Two intents with the same 48 minutes of agent work took 81 and 284 minutes end to end. Here is what the team version changes, what it costs to stand up, the failure that painted 14 empty stages green, and what I have not verified yet.
Kiro's enterprise MCP registry is one JSON file behind one URL: whatever is in it, developers see; whatever is not, never starts. That makes the file a control point — and a hand-edited control point is a liability. I rebuilt mine as a generated artifact: AWS Agent Registry as the approved catalog, AgentCore Gateway as the only remote MCP endpoint, and a small publisher that turns approvals into the file Kiro reads. Here are the five decisions, the six things the docs did not tell me, and what I still have not verified.
Iterating a 24-slide conference deck under review pressure, I stopped editing PowerPoint by hand. Every version is now rebuilt by an idempotent python-pptx script from the last validated file, and reviewed by four parallel critique agents reading rendered frames — because the XML lies about what the eye sees. Eight blocking defects on a version I was ready to ship, one vertical-tab pitfall, and a ghost footer hiding in the slide master: here is the pipeline.
I entered an agent football tournament: five LLM-driven players on Bedrock AgentCore, one command per game tick, every tick latency-sensitive. Per-role Nova model selection, memory switched off, and deterministic rules applied after the LLM made the team work — and then command fingerprinting in CloudWatch proved my perfectly healthy runtimes had been playing for the wrong team all along.
For a conference session I co-delivered, I wanted a hundred-plus people swiping live votes on their phones — without Mentimeter, without Kahoot, and without a single credential in the JavaScript bundle. The whole backend is three DynamoDB tables behind an IAM-authenticated AppSync API: a singleton GameState row drives lobby, launch, reveal and podium; Cognito guest identities sign every call with SigV4; votes are namespaced by runId so a reset hides everything without deleting anything. Total AWS bill for the event: under a dollar.
I wanted a detection app my kids could use without a single camera frame leaving the device. The answer: RF-DETR quantized to int8, ONNX Runtime Web on threaded WASM, and a static Amplify site.
My pipeline generates a playable HTML5 game every day with Bedrock. An audit found 6 of 7 published games broken — while every one of them passed structural validation. Here is how I rebuilt the quality gate around real usage: hit-tested playtests, a progress-signal check, prompt contracts, and a fail-closed publish loop that rejected three broken versions before shipping a good one.
Four domain-specialist agents on Bedrock AgentCore draft one integrated specification, catch a cross-domain conflict between their own outputs, and hand it to a human to resolve. A concrete look at what multi-agent actually means when the agents have to agree.
I built an iOS photo backup where your own AWS account is the backend: no servers, on-device semantic search, a Bedrock agent in your pocket, ~$0.35/month. The full architecture story is now published on AWS Builder Center.
A two-day hackathon build where a humanoid walks an inspection route, captures imagery, and an agent on Bedrock checks each assembly against its bill of materials. What the cloud pipeline looked like, and what a hackathon proves versus what it doesn't.
After a week with Amazon Quick -- a desktop AI work companion -- I'm convinced the developer workflow as we know it is dead. Here's what happened when I stopped switching between 15 tools and started orchestrating agents from a single chat.
AWS released the Agent Toolkit for AWS on May 6, 2026 -- a managed MCP server exposing the full AWS API surface to autonomous agents. I shipped an infrastructure agent the same week. Here's the two-phase safety pattern that lets you hand an agent the keys to your account without waking up to a $10K bill.
I woke up to 204 pull requests from a single autonomous agent running overnight. 12 hours, ~$900 in Bedrock tokens, 509 failed builds, zero features shipped. Prompt-only safeguards all failed. Here's the 3-layer fix — hard kill switch, atomic circuit breakers, drift observability — that now prevents runaway agents.
AI agents forget everything between sessions. Your calendar lies. Your brain doesn't scale past a handful of active projects. Here's the three-layer memory stack I built with Kiro CLI + Obsidian to ship a dozen-plus projects in parallel without cross-contamination.
Boulder is a meta-orchestrator that spawns AI workers on ECS Fargate to autonomously write, deploy and self-heal full-stack apps on Amplify Gen 2. Here is the architecture, the decisions, and why every serious AI agent system eventually converges on SQS + Bedrock + IAM.
The Orchestrator is an AI agent that delegates ALL work to specialized subagents. Here's the architectural philosophy behind why specialization beats generalization in AI agent systems.
A 5-component framework for writing effective system prompts for any AI agent — Bedrock Agents, Claude Code, LangChain, Strands, or custom builds. With a practical Claude Code implementation.
Turn Claude Code into a persistent executive assistant with morning briefings, auto-logging, context-aware reminders, complex skills, and a memory that compounds over time — using only markdown files.
A hands-on walkthrough of deploying OpenClaw on AWS using AgentCore Runtime for serverless agent execution, Graviton ARM instances, and multi-model Bedrock access — from CloudFormation template to customizing the agent's personality.
Discover AI-DLC (AI Development Lifecycle), a structured framework for AI-assisted software development. Learn how I used it to build this blog from scratch and how it enables continuous iteration.
Enterprise workflows often require interacting with web applications that lack APIs. Traditional automation scripts are brittle and break when UIs change.