Designing for AI + Systems
Reducing complexity
Business Design Principles for Reducing Complexity in AI enabled systems
Aynne Valencia
Strategic thinking to move business from the generative to the agentic, assistive, anticipatory, and agentive.
Last update: Jan 24, 2026
Data is the Product
Your product is your content, and what you can do with it, with AI, that is mostly data
The adage: garbage in, garbage out, applies more than ever. Data works from input to get output. The more meaningful, factual data, the better the output.
Data works from input to processing, analysis, inference, and recommendation. This requires good. One must work with facts and context, and the job of the proxy is to make meaning.
Expecting to show meaningful results? Make sure it’s easy to ingest data that matters, that means if someone needs to upload data, map files, import tableau, etc. that yoru uploading and field mapping tools are optimzed to make that process smooth or better yet, use systems that parse the data for them, create a sample set to validate the data has been ported correctly (this also builds trust by showing what happens behind the scenes), allow users to make the necessary corrections.
Missing data? No problem. First, diagnose the problem:
Is this something users always seem to be missing? Then, make it easy by fetching the data for them, or give them a good experience with the data they do have.
Do your users know how the data should be formatted and what it is for?
Can you get the data from another input/source? Do you have a sensor that can be used? Is there a map or open-source data source you can grab to get the information without the user needing to intervene? Start there to make products that people will love.
Context Matters
AI is context-bound. It knows what it has learned to find through training, and through the history stored. Do your customers a favor and make sure you set them up for success with the prompts that will help them get the information they need.
Want to optimize a flow? What information am I looking at? How might we anticipate what someone wants to do next?
How do we let someone know information is present and what kind of information it is?
LLMs require scaffolding for customers to understand what information they can ask for.
Do your llm and your users a favor by providing the context, the regulatory rules, the cultural nuance, and your brand voice.
* By the way - I have a canvas for making a brand voice for chatbots that helps to set the stage for this.
Make Meaning with the big picture
Make Meaning with the big picture
In a world where AI can competently create a screen or a discrete user flow, what AI cannot yet do effectively is understand the gestalt of any given experience.
Use the interface design principle of progressive disclosure to allow users to visualize and make meaning of the data, and then help them drill down to get to the details.
Dashboard, heads-up displays, and landing spots serve the purpose of orienting a user to what is available to them and to help make meaning, give context, and most importantly, surface what features are available to them.
Transparency & Trust are table-stakes
Just say no to suspect data, deliver the minimum trustworthy product
Unclear or un-actionable results will not do. Show your work, allow customers to tweak the knobs, and add and remove variables. Give the agency to your users.
Most llms do this by showing the thinking as information is being gathered. In other places, you can use loaders and other devices to show what is happening under the hood.
Ideally, you can also have places in your product where customers can have the opportunity to change certain variables to get results more tailored to them.
And always show confidence levels if you aren’t 100% sure. Better to be honest than be wrong and lose trust.
Respect the model
Mind the model - not all are created equal
What kind of model are you using? LLM (text)? Image generation/recognition? Sound? A multi-modal combo? How was your data trained? How large or small is the data set? Are you pulling in data from secondary sources? Which ones? What is your company's “secret recipe” and how do you make it ownable?
Knowing this information will help shape what experiences you can deliver with a high degree of confidence and avoid the possibility of data hallucination. The better the data, the better the output.
A picture is worth a thousand words
Visual Design is critical information design - so don’t skimp
A symbol can tell a thousand lines of text. Use visual design to inform and guide users to things they should pay attention to. This is not in the service of being pretty; it is the method of conveying information to your customers and making your product necessary to them. You can use the size of text and placement to make something more important than others. Typeface is not just for style its for legibility, do your zeros like “o”s? Is a one confused with a lowercase letter “l” (like this)? Thats is type not doing its job - particularly when you are using data with mixed case and numerals.
Optimize for the majority, innovate on the edge
Optimize for the 80%
What do 80% of your customers need to know? AI systems thrive on repeatable workflows, common questions, and repeated patterns. When data can be returned with a high degree of confidence, you will be the hero to your customers. It’s okay to start small, make sure you do that, and do it well, and build from a strong foundation. Improves perceived quality, reduces friction, and makes the system feel smarter and more trustworthy.
But… make sure you leave room for feedback on the edge cases. These are the potential opportunity areas.
Mind the gap
We are not our users.
This one is hard; no one gets up in the morning and says, “I want to frustrate, alientate and confuse other people, and make them feel small”. I believe we all want to be good people. BUT.. it is human to be biased towards what we know.
For example, if you stare at terminal command-based interfaces all day, then of course, that’s easy for you to use.
Are you a person who learns best with written words? Then reading a wall of text is fine and dandy.
Are you like me and are visual-spatial - then you lean towards information with diagrams.
But consider this:
The average type size for data tables is roughly the equivalent of 12 point size. Does information get lost in the shuffle? What about someone who doesn’t have 20/20 vision or has tired eyes from staring at a screen? What about someone who needs data on the go- on a tiny smart device?
We must always remember that the person buying the product has different experiences, hardware, physical and cognitive constraints, priorities, aesthetics, comfort level with technology, and areas of expertise than the people making or designing products. Contextual, human-centered design matters more than ever in an AI world.
Don’t be the person who unintentionally leaves someone else behind.
People and planet first, last and always
The core principles of good design are good for business. When you start from a place of caring about people and the planet, you get better products.
Ask yourself, do I have a way to run the data and sense make without making a data center call? Is this something I would run on-prem (using my own servers)? This could work if you have the resources. , have predictable workloads, or there are strict regulatory or latency requirements.
Consider creating default prompts that are most likely to get the job for the desired outcome to prevent user frustration and unneeded data calls.
Thinking at the scale of the micro and the macro is the practice that will save the space we all share and get us to a cleaner, greener future for ourselves, our children, and generations ahead.
The things that transcend the latest fad and are timeless, necessary, and feel effortless and inevitable.
Let’s begin
To wrap this up
These are only ten things I have learned from the front lines using this medium for the last two years and from working with machine learning on and off for over a decade.
Technologies are evolving at a rapid pace, but what doesn’t change is how people understand information. Design principles are the fundamental foundation for how to approach something - before it gets built.
That said, I am always curious and in learning mode, and love to hear about other things they have learned. And I am always willing to talk about these things because they shape the future we all have a stake in.
By building on a solid foundation, you are enabled to not reinvent the wheel and instead concentrate on intent, focus, and feasibility, which are the ingredients that are specific to your business and your customers.