VOL. 01AI · Memory
Memory architectures for AI agents.
Short, long, episodic, semantic — what each layer is actually for, and how the good agent systems wire them together.
Curtis
Blanchette,
ships systems.
I lead engineering at Metalab. Right now: AI-native client work, secure AI enablement, and the workflows that accelerate how engineers ship.
02Selected work
Eleven projects I'd be happy to be remembered for.
Each one shipped. Some are still in production; some were closed-source and are summarised. The full list is longer than is interesting to read.
“Curt is the best problem-solver I have ever worked with. We could get the outline of a business problem, sit on a meeting for two hours, and he'd have the technical requirements, scope, architecture diagrams, an ERD, and a low-fidelity design by the next day.”
03Learning Centre
Talking-head explainers from the workshop.
Short transmissions on the systems behind AI delivery — memory, retrieval, agent loops, the un-fun parts. Cross-posted from LinkedIn.
VOL. 02Coming soon
More transmissions inbound.
Talking-head explainers ship semi-frequently — typically the week after I post them to LinkedIn. Subscribe there to catch the next drop.
04Writing
Notes from the trenches.
Mostly on AI-native delivery, judgment as a priced layer, and what changes when execution gets cheap.
05Stack & tools
The tools I trust, more or less in order.
Software is a craft. These are mine. None of them are dogma.
languages I reach for
- TypeScript everywhere
- Python AI, data, control
- Rust latency-critical
- Java enterprise stints
- Bash too much, probably
- SQL fluent · Postgres
platforms & infra
- Postgres the one true DB
- Redis queues, caches
- AWS ECS · CDK · S3
- GCP Vertex RAG
- Terraform infra as code
- Vercel edge + RSC
AI / ML tooling
- MCP model context protocol
- Ollama local LLMs
- Anthropic Claude code + API
- Vertex RAG vector retrieval
- Pi.dev coding harness
- Shiki syntax highlighting
IoT & control
- ESP32 MicroPython
- MQTT telemetry spine
- OSQP convex QP solver
- EKF state estimation
- MPC model predictive control
- FastAPI control surface
frontend
- React 19 RSC + suspense
- Next.js 15 app router
- Tamagui RN + web
- Tailwind v4 CSS-first
- Angular legacy carried
- MDX long-form
how I work
- Trunk-based small, frequent
- Design docs before code
- RFCs for the big stuff
- Review SLA same day, always
- Senior-heavy from day one
- On-call fair rotations
06Now
What I'm working on this month.
A /now page in the Derek-Sivers tradition. Updated when it stops being true.
I lead engineering at Metalab — building AI-native client solutions, secure AI enablement for engineering teams, and the workflows that accelerate how engineers ship.
Last updated · 2026 · 04 · 12 — Open to staff & principal-level conversations
$cat ~/.plan
shipping ......... ai-native client work · metalab
enabling ......... secure AI for engineering teams
defining ......... workflows that accelerate engineers
writing .......... judgment as a priced layer
reading .......... "The Soul of a New Machine"
building ......... cortex MPC · mycelium IoT
avoiding ......... twitter, mostly
$uptime
on this earth — 13,400 days & counting
$_
Got something that shouldn't break? Let's talk.
Open to staff & principal-level conversations. I read every note; expect a reply within a day or two.