Open navigation
WonderWhy.ai
Fausto Albers
I build agent systems for messy real work.
My work sits between operations, cognitive science, and production AI: agents that use tools, keep state, face evidence, and survive contact with industrial constraints.
What I am working on now
The common thread: take agents out of the demo room and put them where mistakes carry a cost.
Steel-industry R&D
EU-funded work on agents that read fabrication drawings and produce auditable machine programs.
Industrial Digital Twins
GenAI R&D with the Industrial Digital Twins research group at Amsterdam University of Applied Sciences.
AI Builders Club
A practical builder community for people shipping agents, tools, memory systems, and production AI.
What I am learning
Most agent failures start in memory, evidence, authority, and feedback.
- 01
Agents get reliable when something outside the model can check their work.
- 02
Memory earns its keep once it changes what the agent does next time.
- 03
Improvement is a durable change to a tool, skill, or policy that holds up outside the episode that produced it.
- 04
A model that scores its own work learns to satisfy the judge, so judgment lives in a part it cannot rewrite.
- 05
The failures that hurt are the ones where an answer compiles, renders, or sounds right while staying wrong.
Where this shows up
A few public traces. The private work is more detailed, but the pattern is visible enough.
Thinking in public
Notes on agent systems, coding agents, memory, evaluation, and the operating layer around AI.
Work with me
I am useful when the work involves agents, memory, evaluation, technical documents, or teams making AI part of real operations.
Email Fausto