Agents are Bad at Writing Agents
Why eval-driven agent loops optimize for passing the metric over the goal, and what that means for building reliable dev agents.
Blog
ReReadme: Doc-as-CI for Repository Context
Using agent workflows to generate and maintain README documentation as persistent repository context.
Claude Code Save Plan Hook
Never lose a Claude Code plan again. A tiny hook that snapshots your latest plan into the repo the moment Claude transitions from Plan to Edit.
Reasonable Autonomy in Claude Code
A practical guide to configuring Claude Code permissions for real-world workflows, with examples.
Using Markdown Templates with AI
While working with LLM tools for production application development, I’ve found one of the highest-leverage productivity hacks to be the use of Markdown...
Vibe Coding a Spotify Playlist Analyzer
A quick vibe-coded Python project exploring what can be inferred from public Spotify playlist data using modern AI coding tools.
AI Agents Ate the Conference: Reality, Hype, and Hard Lessons from re:Invent
A candid engineer's recap of re:Invent 2025: the gap between leadership and engineering on AI agents, and what real-world agent architecture looks like.
Automated LLM Fine-Tuning with Multi-Agent Systems
Notes from AWS re:Invent 2025: Automating LLM fine-tuning through a multi-agent workflow optimizing accuracy, cost, and latency.
Customizing Foundation Models with Amazon SageMaker AI
Notes from AWS re:Invent 2025: Simplifying and accelerating foundation model customization using SageMaker AI.
Agents in Financial Services: Allianz and Coinbase Keynote Highlights
Notes from AWS re:Invent 2025: The Allianz and Coinbase keynote on agents, resilience, internet payments, and the emerging x402 protocol.
Building AI Agents at Scale
Notes from AWS re:Invent 2025: Scaling agentic systems, enforcing multi-tenant safety, and choosing the right architectural patterns.
Advanced Document Processing with LLMs and AWS: Modern IDP Patterns at Scale
Notes from AWS re:Invent 2025: Intelligent document processing, multimodal extraction pipelines, and AWS-native orchestration for high-assurance workflows.
AI Code to Prod: Why Context Matters More Than Vibes
Notes from AWS re:Invent 2025: Shifting from vibe-driven prompting to structured AI workflows, featuring Kiro and the rise of the 'context architect'.
Architecting Agentic Systems: Self-Evolving Patterns for Autonomous Ops
Notes from AWS re:Invent 2025: Self-evolving architectures, hierarchical agent patterns, and the operational realities of deploying autonomous AI in production.
LLM Eval Reliability Foundations
Notes from AWS re:Invent 2025: Why LLM benchmarks are failing, how contamination and nondeterminism distort scores, and where evaluation is heading.
Real-Time AI Inference Patterns from the Gaming Industry
Notes from AWS re:Invent 2025: How game studios build real-time inference engines, optimize cost, and blend narrative between players, designers, and AI agents.
Model Privacy Assessments in Modern GenAI Systems
Notes from AWS re:Invent 2025: Reconstruction attacks, attribute inference, guardrails, and differential privacy for GenAI and RAG systems.
Agentic RAG in Practice: Strategies for Smarter Retrieval
Notes from AWS re:Invent 2025: Advanced and self-corrective RAG patterns, including orchestrator agents, branching strategies, and accuracy-boosting techniques.