Roadmap
Every vertical Agentsia Labs publishes is a long-term commitment: quarterly refreshes per release, automatic retests when frontier models ship. Naming a quarter publicly is a forcing function, so we only date what we are confident we can defend. Three verticals are in design for 2027 with no dated commitment yet.
Adtech is the operator-of-origin vertical for Agentsia. RTB has a hard end-to-end latency budget (100 ms in the OpenRTB spec; 50 ms for the DSP bid decision inside it), and frontier APIs do not credibly meet that budget. Decisioning is high-volume, narrow, and scorable against ground truth on most axes. A clean first vertical.
Fintech decisioning rhymes with adtech: narrow, high-volume, regulated, latency-sensitive. Rubric work transfers. Credit decisioning, fraud typology, and transaction-pattern reasoning expose a different set of capability axes.
Statutory interpretation, contract analysis, regulatory reasoning. Rubric design is harder because ground truth is partially subjective; expert panel structure is being designed first.
Clinical triage, decision support, documentation quality under HIPAA-aligned constraints. Scoring rubric requires clinical reviewer availability; panel recruitment is in progress.
ADAS perception edge cases, in-vehicle agent reasoning, driving-telemetry interpretation. Multimodal elements make the evaluator architecture more demanding. Scheduled after the three text-first verticals ship.
Have a commercial vertical that deserves an Assay and is missing from the list? Write to ammar@agentsia.uk with a short brief.