Do we need an agent, a simple automation or a better workflow?
We answer that by looking at the result people will actually use. A fixed process may only need a script. An agent may be appropriate when varying inputs, multiple tools and decisions interact. The task and its constraints guide the choice.
Typical starting situations
- Documents are regularly compared or summarised.
- An AI prototype looks convincing, but its quality has not yet been evaluated transparently.
- A team wants to understand access, approvals and correction effort before an integration.
What we clarify together
These elements describe a possible scope. We agree which outcomes your project needs.
- Workflow description
- Inputs, desired output, people involved and exceptions. This reveals where support is needed.
- Comparison of possible approaches
- An existing tool, fixed automation or an AI system: assumptions and open questions are considered side by side.
- Pilot framework
- Suitable sample data, evaluation criteria, permitted actions and a decision point for continuing.
- UnderstandTask & Limits
- DesignApproach & Experiment
- EvaluateResult & Failure cases
- Hand overResponsibility & Operations
A concrete starting point
A team wants to compare several proposals. Before choosing a model, it clarifies which details belong in the comparison, how missing prices are handled and who reviews the result. The first experiment can be limited to synthetic documents. This is a possible approach, not a report of a client engagement.
Further reading
Evaluating AI agents
A transparent evaluation framework for results, failure cases and permissions.
Alpine-X
My own project approach to AI workspaces. The project page describes its development status.
Preparation & scope
For the first conversation, a description of the current workflow and an anonymised example of the desired result are enough. Confidential documents or access are only needed after agreeing a suitable transfer method.
Technical guidance does not promise a particular automation rate. Data protection, suitability and required approvals must be clarified before use. A pilot does not yet take unsupervised responsibility in live operations.
Frequently asked questions
What does an analysis deliver?
Depending on the agreed scope, this might include a workflow description, a comparison of approaches and a proposal for a limited experiment. The proposal specifies the actual deliverables before work begins.
Does an AI solution need to be selected already?
No. A recurring task can be the starting point. Existing systems and simpler approaches belong in the comparison.
Can this lead to implementation?
Yes. After clarification, potential implementation is discussed separately with Stoicera. Analysis, development and ongoing operations each have their own scope.
Describe your project
One concrete question is enough to start. The project brief helps organise your goal, existing systems and open questions. Save it locally as text and send it yourself afterwards.