Working with Stefan

This page describes the kinds of work Stefan Preu is looking for and how a conversation with him might start. It reflects his stated interests and preferences. It is not an offer, price list or availability statement.

Three ways in

Stefan is open to engineering roles, consulting projects and collaborations, and he gives hiring and consulting equal weight.

• A role: engineering or architecture work where agents do real work, such as harnesses, context and memory, and workflows that need to be deterministic. AI labs included. He is also open to leadership roles, discussed individually.
• A consulting project: an organization with agents and real systems that wants help designing the context, tools and boundaries around them, and getting its team ready to work with agents.
• A collaboration: open-source tools, research and experiments on context for agents.

Source: preu.at/about

Who he especially wants to work with

Stefan is especially interested in advanced agent practitioners with real business or production use cases. He also wants to help larger organizations become ready for agentic workflows. Software companies and organizations in other industries are both relevant, and people earlier in adoption are not excluded.

PM CLI can be an entry point: some visitors may want help using it, or equivalent tooling, in an actual workflow.

Topics a project might cover

These are directions to explore, not fixed packages:

• understanding a workflow and the information needed to carry it out
• making relevant knowledge discoverable across sessions and tools
• designing or adapting an agent harness and memory approach
• combining deterministic execution with agent investigation where it fits
• creating a bounded environment for experimentation and evaluation
• improving feedback, acceptance criteria and inspection of work state
• adapting an existing system as models, goals and processes change

Stefan is interested in both the initial implementation and its continuing adaptation.

How a first conversation might start

Stefan starts from the organization's existing knowledge: its goals, decisions, past work, responsibilities and how information moves between people and departments. Scattered information can still be useful; structure and quality can improve step by step.

He treats cross-department handoffs and coordination as part of the problem, alongside software and infrastructure. The useful question is what the people and the system need to accomplish, not which fashionable tool to introduce first.

Three early questions Stefan asks

Early in a conversation with an organization, Stefan usually wants to know:

1. Does the organization already use LLM providers?
2. Which providers does it use?
3. What access are agents allowed to have to company data?

The answers shape what is possible from the start.

Data access and confidentiality

Stefan respects confidentiality and does not assume that private company information can be sent to third-party model providers. Exploring a workflow is not permission to ingest, expose or connect private data.

Depending on the situation, options include appropriately anonymized or prepared data subsets, or local models for preparation and gate checks, where the organization permits them. Those options still need case-specific privacy and security review; running a model locally does not make disclosure safe automatically.

When it is not a fit

The kind of work Stefan wants to do needs a legitimate way to experiment with relevant data. If an organization rules out every workable route to experimentation, he sees little basis for this kind of collaboration.

That is a fit preference, not an offer to work around company policy.

What makes an opportunity attractive

Stefan values freedom to experiment, investigate, research and learn. Seeing an organization's actual projects and what it has already tried with agents makes an opportunity more interesting to him, because it gives him something concrete to understand and build on.

He works well with curious people who welcome direct feedback and are willing to rethink how things are done.

Ownership after the work

Stefan considers it essential that a client can keep using and changing the result independently. He aims to avoid artificial dependency on a vendor, model, harness or on himself.

Continued involvement should be valuable because the system keeps improving, not because it cannot run without him. He leans toward building together, with the arrangement depending on the people, the project and the setting.

What this bot cannot tell you

There are no published prices, rates, packages, durations, retainers, response times or guaranteed results, and this AI representative does not know Stefan's current availability. It cannot book meetings, send messages or agree to anything on his behalf.

Scope, availability and terms need a direct conversation with Stefan.

How to get in touch

The contact form at preu.at/contact is the main route. Pick a role, a consulting project, a collaboration or something else; a few sentences about what you are working on is enough. Email works too: stefan@preu.at.

Services and client work are presented at unbrained.dev. If you would rather read code first, start with github.com/unbraind.