Service line 06 — Agents
Agents with a job description.
Production agent systems with tool access, measured quality, and a person at every gate that matters.
Automations or Agents
Automations run a process the same way every time. Agents make decisions inside the process, and are shipped with an evaluation set and a human gate. If the work needs a model to decide something, it's an Agents package.
What's included
SCOPED BRIEF
One job per agent, defined narrowly enough to evaluate. Agents that do everything do nothing well.
TOOL ACCESS
Read and write access to the systems the job requires, with permissions and rate limits enforced.
EVALUATIONS
A scored test set built from your real cases, so quality is measured rather than asserted.
HUMAN GATES
Consequential actions wait for a person. Agents produce; they don't decide.
OBSERVABILITY
Traces of every run — prompt, tool calls, cost, latency — so behaviour is auditable after the fact.
ORCHESTRATION
On fleets: routing, shared memory, and retries across agents, with one control surface over the pipeline.
Packages
Agent Pilot
$4,500One scoped agent with real tool access, an evaluation set, and a human gate.
- — One scoped agent
- — Tool and data access
- — Evaluation set and scoring
- — Human approval gate
- — Deployment and monitoring
- — 2 revision rounds
14 DAYS · HUMAN REVIEW BEFORE DISPATCH
Agent Fleet
$12,000A multi-agent pipeline with orchestration, observability, and human checkpoints.
- — Multi-agent pipeline
- — Orchestration and routing
- — Shared memory and tools
- — Observability and traces
- — Human checkpoints
- — Runbook and handover
- — 3 revision rounds
45 DAYS · HUMAN REVIEW BEFORE DISPATCH