Agentic Systems
Multi-agent orchestration, tool use, and evaluation harnesses. How far can autonomy go before reliability breaks, and how do we push that line?
AI & Product Research
The Lab
Applied research with a shipping deadline. The Lab exists to find what is newly possible and drag it into production before the rest of the market notices.
Operating principle
Experiments,
not opinions.
Every week the Lab runs structured experiments against frontier models and new product surfaces. Each one ends in an artifact: a working prototype, a benchmark, or a written post-mortem. Nothing stays a hunch. The results feed the Agency's client work, the Foundry's own products, and the Signal's publishing.
Research streams
Product Interfaces for AI
Beyond the chat box. Canvases, ambient agents, generative UI, and the interaction patterns that make model capability legible to humans.
Applied Retrieval & Memory
Long-horizon context, knowledge systems, and memory architectures that let products actually learn their users.
Model Economics
Cost curves, routing, distillation, and the unit economics of AI products. What becomes viable with every price drop, and who gets there first.
Geospatial
Location intelligence, spatial data pipelines, and models that reason about places and maps. What changes when geography becomes a first-class input to every decision?
Knowledge Graphs
Graph databases, graph-based agentic memory, and knowledge graphs as the substrate for a durable company brain. How do organizations build memory that compounds instead of evaporates?
Read the findings.
Lab output ships publicly through the Signal: essays, teardowns, and benchmarks.
Go to the Signal →