Marvin Brian Tendies· Writing on agentic UX· Berlin
Working notes on
how people and
AI agents
get along.
Essays on the patterns that decide whether an agentic product survives contact with real users: what the agent shows before it acts, how it reports a partial result, and how a person corrects it without starting over. One essay every few days.
Latest essays
Patterns, oversight, trust and method. Written from the middle of the work, not from the outside looking in.
Where the
argument happens
There is no forum here and no newsletter, on purpose. An empty forum is worse than no forum, and a signup form collects addresses I would rather not hold.
So it is email. If an essay is wrong, if a pattern breaks in a domain I have not seen, or if you disagree with the whole premise, write to me. I answer. Disagreement is worth more to me than agreement, and it is the only reliable way I find out what I got wrong.
How I work, now
that agents work too
Method beats model.
Every few months a better model arrives and the demos get louder. What does not change is that most AI work fails on the same thing it always did: nobody wrote down what good looks like.
This gets sharper the higher the stakes. In a supply chain, an agent that reorders the wrong quantity does not produce a bad screen, it produces a truck. That is where the questions stop being theoretical: how much autonomy before someone has to confirm, what the agent shows when it is uncertain, and how a planner overrides it at two in the morning without reading a manual.
So my practice moved. Intake got stricter, because a vague brief now produces confident nonsense at scale. Output became a contract, not a deliverable. Verification became a step instead of a footnote.
Intake gets stricter
A human designer fills gaps in a brief without noticing. An agent fills them with plausible invention. So the brief has to carry what was previously carried by shared context.
Output becomes a contract
Not "a research summary" but a defined shape with named sections, required evidence, and stated exclusions. If the format is not specified, the format is whatever the model felt like.
Verification is a step
Checking used to be something careful people did anyway. It is now a named part of the process with its own time, because the output looks equally confident whether or not it is true.
Researcher by conviction,
builder by curiosity
Ten years of enterprise UX taught me one thing: the hardest design question is not how something looks, but whom people trust, and when. Working with agents is that question in its purest form, and I get to ask it full time, on supply chains, where a wrong automated decision has a physical consequence somewhere in the world.
I ask questions until the answer fits, write methods down so teams and agents do the same work, and run trails around Berlin in between.
Questions
and answers
What is UX for AI agents?
UX for AI agents, also called agentic UX, is the design of how people and autonomous software agents work together. Instead of designing single screens, it defines how an agent plans, decides, acts and replans, how it explains its decisions, when it asks for approval, and how a person can set goals, monitor progress and intervene at any time.
Its core building blocks are goal delegation, orchestration, explainability, oversight and trust calibration.
How is agentic UX different from classic UX design?
Classic UX design shapes screens and interactions that happen once. Agentic UX shapes behaviour over time. The unit of design is not a screen but a trajectory: what the agent does across minutes or days, where it pauses, what it shows when it is uncertain, and how a person corrects it without starting over.
What does an agentic UX designer do?
An agentic UX designer treats agent behaviour as a design material. That means designing approval and escalation rules, explainability patterns that make decisions traceable without overwhelming people, correction and handoff flows, and the documentation that lets other teams reuse those patterns as a standard.
Why do autonomous systems need explainability and oversight?
Autonomy without trust is a reason to switch a system off. Trust becomes calibrated instead of blind once an agent explains its decisions, asks for approval at the right moments, and accepts correction. In regulated environments this is not a feature but a precondition for adoption.
Is Marvin Tendies available for work?
He is an agentic UX designer in Berlin, currently working in enterprise supply chain software, after four years at DB Systel on Deutsche Bahn platforms. He is open to conversations about agentic UX, AI-native product work, and collaboration on open pattern work.
Email: hej@marvintendies.de