Skip to content

Task routing

AIgency's planner scores agents against a task using a static capability matrix across 17 task categories.

When AIgency plans a run, it needs to decide which agent is the best fit for each task. It does this with a static, hand-tuned capability matrix rather than a live benchmark.

AgentDock Pro

Task routing is part of AIgency, available only in AgentDock Pro.

How it works

Each task's free-text description is classified into one of 17 task categories — including frontend_visual, backend, architecture, debugging, testing, and git_operations — using keyword matching. Each supported agent has a score from 0–10 for every category, and the planner assigns the task to the highest-scoring available agent, subject to your concurrency limits and any manual overrides.

Note

The scoring matrix is static — there is no online benchmarking, telemetry, or learned weighting happening inside AgentDock. Code comments describe the matrix as calibrated against an external benchmark run, but that provenance is a vendor claim from the source code, not something independently verified for this documentation.

Overriding routing

Routing only decides the initial (source: planner) assignment. You can always reassign a task to a specific agent yourself — see Agent assignment for the full set of assignment fields, including manual overrides and fallback agents.