Blog
Notes from the workbench
Engineering deep-dives, product thinking and field notes on building serious AI systems.
Engineering
How AST-based pruning cuts 93% of your context tokens
A walk through the parsing pipeline that turns a 1.2M-token monorepo into a focused 86K-token skeleton — without losing the signal.
May 19, 2026·8 min read
ProductDesigning the Labyrinth Workflow Composer
Why we built a visual node canvas for multi-agent orchestration, and the routing primitives behind it.
Apr 27, 2026·6 min read
LLMOpsThe metrics that actually matter for production agents
Latency, throughput, spend and generation quality — how to instrument them and what to alert on.
Apr 2, 2026·10 min read
EngineeringMulti-model routing: picking the right LLM per request
A practical framework for routing between frontier and small models to balance cost and quality.
Mar 15, 2026·7 min read