Designing for humans and machines
For a long time, interface design meant a person in front of a screen. That person is still here. But they now share your product with search engines, AI assistants, voice interfaces and agents acting on someone's behalf.
Designing for machines means building a product that software can read correctly and use without guessing. The work shifts from pixels to meaning: how information is structured, what things are called, what state something is in, and what each action actually does.
At Pink Forest we design for both audiences at the same time. In our experience, when a product makes sense to a machine, people find it easier to use too.
Why now?
More and more people meet your product through a machine first. They ask an assistant instead of searching. They hand a task to an agent instead of clicking through it themselves. If that assistant can't make sense of your product, it won't recommend it, and an agent certainly won't use it.
Three shifts happening at once
Discovery is moving to AI
When someone asks which service to choose, the answer depends on how clearly you've described what you do.
Work is moving to agents
Agents compare prices, book appointments and fill in forms. A service whose logic only lives in its visuals is a dead end for them.
Machines act on what's written down
Vague labels and hidden states lead to wrong actions, and a wrong action reflects on your brand, not on the agent.
Screens still matter. They're just not the only interface anymore.
Five layers of UI
One product, several kinds of usersWe use a simple model: an interface has five layers. Most products touch several of them, often without anyone deciding so. We prefer to design them on purpose.
Human to human
People talking to people through a product: messaging, comments, shared workspaces. The question here is less about how it looks and more about whether people understand each other.
Human to machine
The interface everyone knows. A person tells software what to do through screens, forms, voice or chat. Good feedback and forgiving error handling still make or break it.
Machine to human
Software reporting back: notifications, recommendations, AI-generated answers, status updates. People need to know what happened, why, and how sure the system is about it.
Machine to machine
Systems talking through APIs, feeds and structured data. Here the naming, the data model and the documentation are the interface. A badly designed API is a badly designed product. It just fails more quietly.
Agent to agent
AI agents working with other agents on someone's behalf: delegating tasks, exchanging data, handing work back to a person. They need to know what each side can do, what it's allowed to do and where things stand. It has the fewest established patterns, which is exactly why it needs designers.
What we've learned so far
Six principlesStart with meaning, not looks
Get the information and data model right first, and every interface built on top of it gets easier, visual or not.
Be explicit
Machines don't read between the lines. Use the same name for the same thing everywhere, label actions plainly, and show states: waiting, done, failed, needs approval.
Describe what's possible
An agent has to know what your product can do, what it needs and where its limits are. That description deserves the same care as any other product copy.
Keep people in the loop
Plan the moments where someone checks, approves or undoes what a machine did.
Let the machine explain itself
When software acts or answers, show where its information came from and how confident it is, in plain words.
One source, many surfaces
The same content should work on a screen, in a voice reply, in an AI summary and through an API. Design it once, then adapt it.
How we can help
We work across strategy, structure and front-end, so the thinking doesn't get lost somewhere between a slide deck and the code. Read more about what we do.
- Reviewing how AI assistants, search engines and agents see your product today, and where they get stuck
- Information and data modelling: entities, relationships, naming and states
- Designing how AI features and agents behave around people: what they do on their own, when they ask, and how they explain themselves
- Designing APIs, structured data and documentation with the same care as screens
- Front-end that is clear for people and well structured for machines
- Advisory for design teams adapting their roles and ways of working
Questions we often get
Short answersWhat does designing for machines mean?
It means designing a product so that software, such as AI assistants, search engines, APIs and autonomous agents, can understand it and act on it correctly. The focus moves from visual layout to information structure, data, naming and clearly described actions.
What is agent-to-agent UI?
It's how AI agents communicate and cooperate with each other on behalf of people: how they describe what they can do, ask for actions, share data and hand tasks back to a human. Like any interface, it can be designed well or badly.
How do you design a product an AI agent can use?
Start with a clear data model and consistent naming. Make every action and state explicit, describe what the product can and can't do, and set clear permissions. Then design the moments where a person reviews what the agent did.
Does this replace UX and UI design?
No. People are still the ones it's all for, and visual interfaces still matter. The job has simply grown: one product now has several kinds of users, and good structure makes the human experience clearer as well.
How is this different from SEO?
SEO helps search engines find and rank your pages. Designing for machines goes further. It makes the whole product understandable and usable by software, including the assistants that recommend it and the agents that operate it.
Who needs this?
Any organisation whose customers, partners or teams will reach it through AI tools or automated systems. Today that's most digital services, especially in public services, SaaS, fintech, healthcare and e-commerce.
Curious how your product looks to a machine?
We're happy to take a look with you.