A different kind of collaboration
If you have spent your career designing products built on deterministic software -- where the same input always produces the same output -- working with machine learning engineers will require a fundamental shift in how you think, communicate, and design.
I have been working alongside ML engineers at Helsing for over a year now, and the collaboration is unlike anything I experienced in consumer tech. Here is what I have learned.
Understand probabilistic thinking
The single biggest adjustment: ML systems are probabilistic, not deterministic. Your model will be right most of the time, not all of the time. This changes everything about how you design.
You need to ask different questions:
- What happens when the model's confidence is low?
- How do we communicate uncertainty without eroding trust?
- What does the human decision loop look like when the AI gets it wrong?
- How do we design for graceful degradation, not just happy paths?
Learn the vocabulary
You do not need to understand the math, but you need to understand the concepts. Learn what precision and recall mean. Understand the difference between false positives and false negatives and why the tradeoff between them matters for your users. Know what a confidence score represents.
When an ML engineer says "we can improve recall but it will hurt precision," you need to understand the design implications of that tradeoff immediately.
Design for the model's limitations
Every model has a performance envelope -- conditions under which it works well and conditions under which it struggles. Your job is to design interfaces that are honest about these boundaries.
This means:
- Surfacing confidence levels where appropriate
- Building in human review steps for low-confidence outputs
- Designing feedback mechanisms so users can correct the model
- Never presenting AI outputs as absolute truth
The worst AI interfaces are the ones that pretend the model is infallible.
Embrace iteration differently
In traditional product design, you iterate on the interface. In AI products, the model itself is iterating. Performance improves over time. Capabilities expand. Edge cases get handled.
This means your designs need to be resilient to changing model behavior. Build flexible layouts. Use abstraction layers. Design systems that can accommodate outputs you have not seen yet.
Ship feedback loops, not just features
The most valuable thing you can design in an AI product is not a beautiful interface. It is a feedback loop -- a mechanism that captures user corrections and channels them back into model improvement. This is where design and ML engineering converge, and it is where the real product value compounds.
Your ML engineers are not your backend team with different tools. They are partners in defining what the product can even be.