Writing
Empower, not overpower
The rule I hold every AI feature to, from Copilot in Word to a lamp that asks students to keep their distance.
My design ethos is simple to say and hard to keep: technology should empower, not overpower. It should enhance what a person is doing without overshadowing them. I wrote that line for the about page of this site, and it has turned out to be the test I apply to every agentic feature I work on.
Here is why it matters more now than it did before. An agent can do a great deal on its own. It can draft, summarize, rewrite and edit a document. It can create, analyze and transform content in a loop, reasoning through several steps before it shows you anything. Capability is no longer the scarce thing. Restraint is. The question that decides whether people keep using an AI feature is not what it can do. It is whether it leaves them feeling like the author of their own work.
When I worked on Copilot in Word, the aim was to help people draft, summarize, rewrite and edit with a conversational agent, and to do it inside the Word environment they already knew. The suggestions were tailored to the document in front of them, and some text transformations were proactive, offered before the person asked. That is exactly the kind of feature that can tip from helpful into overbearing. The principle I kept coming back to was that the person stays in charge of the page. The agent proposes. The person decides. Where a suggestion appears and how easily it can be set aside both have to reinforce that the document belongs to the writer.
I learned the same lesson in a very different setting. At the University of Edinburgh my team built an expressive light lamp for shared study spaces during the pandemic. Its job was to help students keep a physical distance from each other. The obvious design would have been an alarm or a display with a number on it. We deliberately went the other way. The lamp was non-authoritative, non-invasive and subtle. It sensed proximity and answered with a change in colour. We wanted something that would blend into the room and never resemble a surveillance tool. Usability testing showed it did change behaviour. It also showed habituation: after a while people stopped noticing it. That trade-off is the whole design problem in miniature. Push harder and you become a nag. Stay quiet and you risk being ignored. The right answer is a calibration, not a slogan, and you only find it by watching real people over time.
Fidget Digits was another version of this. It was a handheld self-tracking device that counted the spins students made during online lectures, so they could see their own attention patterns afterwards. Our aim was a device that does not distract users but helps them become more self-aware. Students told us they preferred a handheld tracker to a wearable, which was itself a signal about how much presence they wanted the device to have. We revised the first prototype, which favoured function over ergonomics, until it could sit in the hand without demanding attention.
Makerarm taught me the positive side of the same rule. It was a desktop fabrication system that could 3D print, laser cut, carve and more, and the software had to make all of that accessible to makers without stripping away professional capability. The interface adapted itself to whatever tool head was attached, and a teaching mode let people train the arm by demonstration. Complex technical capability delivered through a simple interface is empowerment. Capability that requires the person to become an expert first is not.
So when I look at an agentic feature now, I ask a short set of questions. Does it keep the person in the familiar tool, or pull them somewhere new? Does it propose, or does it act without asking? Can the person see what it did and undo it? Does its presence fade when it is not needed? And when we test it with real users over weeks rather than minutes, do they still feel like it is theirs?
None of these questions are about model quality. They are about the relationship between a person and a system that can now do a lot on their behalf. Getting that relationship right is the work. The models will keep getting better. The job of the designer is to make sure that as they do, the person on the other side of the screen feels more capable, not less.
More in the writing section, or see the design process these pieces come from.