We find the work worth handing to agents, then measure whether it pays.
Four steps, each a fixed fee agreed before the work starts, and each ending in something you can measure. The assessment is the usual first step; a team that has already defined its process can start at the pilot.
The method
Every engagement runs the same loop, on one process or on many. It is how we decide what to build, and how you decide whether it was worth it.
Find the work worth automating
We sit with the people doing the work, list where the hours go, and rank each process by volume, by what a mistake costs, and by how much of it an agent could do from records you already keep.
Define it
We write down what the agent does, what it reads, what it may change, where it must ask a person first, and what good output looks like. That definition becomes the test the agent has to pass.
Measure the agent-to-human ratio and the quality
How much of the work the agent did, how much a person still did, and how often the result was right the first time. If the numbers do not hold up, the process does not ship.
The engagements
Each step builds on the one before it, and each stands on its own, so you can stop after any of them with something that works.
Assessment
Teams that know agents should be doing more of their work, but not yet which work.
Find and define. Interviews with the people doing the work, a walk through the systems they use, and a ranked list of the processes worth automating, with the data each one needs.
- 2 to 3 weeks
- Fixed fee, quoted per engagement
Deliverables
- A ranked shortlist of processes, each with its expected agent-to-human ratio and its risks
- A written definition of the top candidate, ready to build
- The measures a pilot will be judged on, agreed before it starts
Production pilot
Teams with one process defined, by us or by them, that want it running on real data.
One agent for one process, built on Cortex and connected to your systems with your team, on the models you choose. It asks before it changes anything, and it works alongside the people doing the job until the numbers say it can carry its share.
- 6 to 8 weeks
- Fixed fee, quoted per engagement
Deliverables
- One agent in production on your systems and your models
- The integrations it needs, with access scoped to what the process requires
- A measurement report: agent-to-human ratio and quality against the targets set at the start
- A runbook your team owns
Enablement
Teams that want to build and run the next agents themselves.
Hands-on work with your engineers and process owners on the whole loop: finding and defining a process, building and testing an agent, and measuring it. We build the next agent together, with your team at the keyboard.
- 4 to 6 weeks
- Fixed fee, quoted per engagement
Deliverables
- A second agent, built by your team with ours alongside
- Your own playbook for finding, defining and measuring the next process
- An evaluation set and a review routine your team runs without us
Managed agent ops retainer
Teams with agents in production that want someone watching them.
We run what is live: watch quality and the agent-to-human ratio, review every failure, and update agents as your processes and models change. New processes go through the same find, define and measure loop.
- Monthly, ongoing
- Fixed fee, quoted per engagement
Deliverables
- A monthly report on volume, agent-to-human ratio, quality and model spend
- Fixes and updates for the agents in production
- A standing shortlist of the next processes worth automating
What the agents run on
The agents we build run on Cortex, our agent platform, on the models you choose, including models on your own hardware. You can switch models later without rebuilding an agent. The recorded demos show it working.