Why this role exists
Many business owners know what needs doing but lack the time for the next step. Agents can prepare work: research, draft, create tasks. Because they access real data and take actions, the principle matters even more here: AI suggests, people decide – and every action must be traceable.
Your outcomes
- Agents with clearly limited permissions and defined approval points.
- Logged steps and sources that users can understand.
- Tested behaviour for errors, ambiguity and cancellation.
- Measurable quality per use case instead of impressive one-off demos.
Your day and scope
Breaking down workflows with Product and Customer Success, defining tools and permissions for agents, designing approval steps, building test scenarios, reviewing logs and fixing misbehaviour. Agents never act outside their approved permissions.
Boundaries: You own building, safeguarding and the quality of agents in Peak One. Single AI features without agent logic and the general data platform sit mainly with other roles.
Your first 90 days
First 30 days
Understand Peak One, the data model and possible agent use cases; name risks and approval points for each.
By day 60
Deliver one clearly scoped agent with logging, an approval step and test scenarios.
By day 90
Have evaluated the agent based on real usage and documented a reusable framework for further agents.
What you bring
- Hands-on experience with language models, tool calling or multi-step AI workflows.
- Solid software engineering fundamentals.
- Security awareness around permissions and data access.
- Careful testing, including unusual cases.
What you can learn with us
- Our specific tools and models.
- Our target customers’ workflows.
- Our internal permission model.
When this role is less of a fit
You want agents to act as freely as possible without designing approvals, logs and limits.
Compensation and conditions
We only commit to compensation, contract type, working hours and location once they are approved. We will tell you all of it concretely in the first call, before you invest time in further steps.
Your application process
Short application
Name, email and your answer to the role's entry question. CV optional.
First call
A personal conversation about the role, conditions and your questions. We cover compensation and location concretely here.
Work sample
An announced, time-boxed task in a fictional environment with synthetic data: design or build a simple agent, including permissions, an approval point and test cases. No real systems, no contact with third parties. People assess safety, traceability and explanation.
Decision
A person decides, never an AI. You always get a response.
Your entry question
„An agent should prepare follow-up tasks from open offers for a business owner. Where do you put human approval – and why exactly there?“
Apply now