Operate and improve
Once your agents and pipelines are built, we run and maintain them for a fixed monthly fee. When your agents' logbooks show a job repeating the same way, we automate it and monitor it, and you check the results. More of the work runs on its own over time, and what we built stays current without a new project.
This is for you if
- Your agents and pipelines are running, and you want them kept current as your systems and rules change.
- Your team does the same job the same way every week, and you want it to run on its own.
- You would rather not start a new project each time a source system or a rule changes.
When something changes, it comes back to you
A job that runs on its own does not stay that way by default. If something changes, it comes back to you for approval: the agent prepares and you approve, until your team decides the job can run on its own again.
| What happens | Who approves | |
|---|---|---|
| A source system changes | The pipeline reports the change instead of guessing, and the agent that maintains it prepares a fix | You, before the fix goes live |
| A rule changes | The rule is rewritten, and the jobs that use it go back to "the agent prepares, you approve" | You, until the job runs cleanly again |
| A new market, supplier or system | The new input does not repeat yet, so the agent prepares and a person approves | You, until it repeats and is written down |
| Money leaves, something goes outside, a commitment, an unclear rule, a step the law gives to a person | Always goes to a person, however long the job has run without a mistake | Always a person |
What you get
- The result
- More of the work runs on its own. Work that repeats and is written down gets automated and monitored, and you check the results.
- Kept current
- The agent that builds and maintains your pipelines and agents keeps them up to date. It shows a plan and changes nothing until a person approves it.
- Automation is earned
- A job is automated when your agents' logbooks show it repeating the same way, its exceptions are known and one person owns the result. Your team decides, job by job.
- Watched, not forgotten
- Every automated run is monitored and written to the logbook, and exceptions go to a person.
What we do, what you do
| We | You |
|---|---|
| Run and maintain your pipelines and agents, and keep them current. | Approve each change before it goes live. |
| Read your agents' logbooks and propose the jobs that are ready to be automated. | Decide, job by job, what runs on its own, and name who owns the result. |
| Automate those jobs, test them with evals and monitor every run. | Check the results, and handle the exceptions that come to you. |
| Turn your corrections into rules, so the same mistake is not proposed again. | Tell us when something in your business changes. |
Worth knowing
We prefer this model, because the learning builds up over time. Each month adds entries to the logbook, corrections to the rules and words to your company dictionary, and that is what lets the next job be automated. The agent that builds and maintains your pipelines and agents works in your repository, so all of it stays with you. At DuffBeauty, one of Gathena's agents built the pipelines, the warehouse, the accounting agent and the way to ask questions in Claude Team, and maintains them in Claude Code; the nightly pipelines run on their own and are monitored.
Your questions
- How is it priced?
- A fixed monthly fee, agreed before we start. You pay for the result, never for the hours our agents work.
- What does "automation is earned" mean?
- A job runs on its own only once your agents' logbooks show it repeating the same way, it is written down, its exceptions are known and one person owns the result. There is no fixed number of clean runs: your team decides.
- Can we stop?
- Yes. You own the repository, your agents and the agent that builds and maintains them, and you can maintain all of it without us.
- Where does our data live?
- Your data, pipelines, warehouse and the agents' tools stay in your own cloud, and you control access. The model runs at the provider you chose.