Designing an operating system for a 1.85-inch screen.

The plan was a watch around ₹5–6,000, roughly a third under the Nothing Watch 3 Pro and the Moto Watch it gets compared to. Memory prices climbed through the year and every price in the category climbed with them. Anarc 2 launched at ₹6,999. Those two now sit at ₹7,999. The gap narrowed. It held.

But the interface was designed against the original number. A memory budget and a battery ceiling set for a cheaper watch don’t loosen when the bill of materials goes up; they get harder to hold. Almost every constraint below was decided before that price changed, and none of them were revisited afterwards.

Anarc 1 sold roughly 25,000 units over its life and has been out of stock for a long time, which is the pressure behind getting generation two out. Generation two keeps the silhouette people recognised and changes nearly everything else: a new operating system, GPS, better sensors, lighter on the wrist.

That last one wasn’t a design choice. Anarc 1 was stainless steel; Anarc 2 is aluminium, and the saving paid for the GPS module and the sensor upgrade. It made the watch lighter as a side effect. If the budget had allowed I’d have kept steel and spent more to hold the original thickness despite the GPS. But the money only goes one place, and we put it in GPS and sensors.

The category also moves quickly enough that more R&D time isn’t free. A better watch a year from now competes against next year’s watches. The date slipped more than once. The pressure never did.

Seymourpowell, in London, delivered a visual direction for Anarc 2: around 210 screens, roughly 35 flows, 24 icons and a basic set of global components: buttons, toggles, radios, playback controls. It gave us a clear starting point and a consistent look to argue with. A colleague helped with early exploration; the watch interface itself was mine to build out.

What shipped is around 460 screens and 50 flows, on a rebuilt component system. All 24 inherited icons were redrawn to the new visual language, four or five replaced entirely, and three new ones added for features that didn’t exist in the handover.

Two of those features are Layers AI and Canvas AI, which weren’t part of the original scope at all.

The direction started earlier, and elsewhere. A survey inside our own community went to Seymourpowell in London, who built the style frame Anarc 2 began from. A style frame is a direction, not a system. It doesn’t know about a firmware memory budget, an always-on power ceiling, or what 460 screens do to each other once they have to live together. Most of it changed as it met those things. That is what a frame is for.

By the time the interface was close to final, I wanted it checked by people who weren’t us. I wrote a survey and put it out to 400 people across the country: Gujarat, Mumbai, Bangalore, Delhi. I screened the responses and called about a hundred of them myself. Eighteen of those came in for a session in person, under NDA: some at our office in Delhi, the rest on site in Vadodara.

Most of it came back confirmed. The clearest thing it changed was the icon set: I brought two versions into those rooms, and people rejected the one we had been leaning towards.

That gap is the single biggest number in the product, and it is decided almost entirely by interface. On AMOLED a black pixel is off and costs nothing. Every lit one is a withdrawal.

It’s also a bet. At this price, we decided a watch that lasts the week beats one with a richer interface you have to charge every other night. So that’s where the budget went.

Every module on the watch has a storage cost, and they’re listed in a table our ODM partner keeps. Reading it tells you what this product is actually for.

The four activity badges (the screens that appear when you close a goal) come to about 33 MB. That’s twice the entire weather app with its twenty animated conditions, and more than heart rate, blood oxygen and stress put together.

A company that thought it was selling a health tracker would never spend its storage that way. We did, because that’s the part people talk about.

Watchfaces run from 240 KB to 17 MB depending on how much they move. One elaborate face can cost more than the entire weather app.

What we did about it: we kept the badges and capped them. They’re the part of the product people show each other, so cutting them to save storage would have been saving the wrong thing. What shipped is the most compressed version that still looked like a celebration, the floor, not a compromise we settled for.

The screens existed. The feature was cut because the sensor couldn’t measure it reliably at the price this watch sells for, and a health number you can’t trust is worse than no number at all.

I argued to keep it. The founder and product head cut it. They were right, and it’s the decision I’d defend hardest to someone outside the company.

This is where most of the time went. The handover gave us a direction; turning it into something that reads on a 1.85-inch AMOLED screen in Indian daylight took months.

To ask Layers AI a question, you originally held your finger on the screen while speaking. It tests perfectly on a laptop. On a wrist, with your arm raised and a long question to ask, it’s uncomfortable, and it gets worse the longer you talk.

I only caught it on an internal build. It’s now tap to start, tap again to stop.

I’d had a hunch it would be awkward, but no way to demonstrate it: the problem doesn’t exist on a screen, only on a body. On-wrist testing moved earlier for the features that followed.

Layers is a lifestyle brand. The health tracking is real and we took it seriously, but the reason someone picks this watch over another one in the same price bracket is how it feels to live with.

The samples that didn’t match. On the last trip, the production samples came back with visible grain the earlier ones didn’t have. I flew out with the original samples, put them side by side, and had the vendor show me what surface finish they could actually hold at volume.

The new sample was close enough to the original that I approved it. If it hadn’t been, that was the trip to find out, not after a hundred thousand units.

The gap in the battery readout. Watchfaces draw the battery percentage from a three-character slot plus a percent sign. At 100% it looks right. At 10% you get a number, a hole where the third digit would be, and then a stranded %. On the second trip we found it, and fixed it in the room: iOS and Android don’t print the percent sign at all; they put the number inside the battery capsule. We did the same, and the battery icon went into the watchface tool. Faces that were already finished got redrawn around it. Over messages that is three days and a translation problem. Standing next to the developer it was an afternoon.

Every day, workout and sleep data from both went back to the team. Sensor tuning improved against that comparison. UI defects went onto a running list and shipped as fixes in the next firmware update.

Other people on the team wore it too. Later, tech reviewers got units. The feedback so far has been positive for this price segment.

There’s no launch data yet, so here is what I’d actually go looking for, and what I already believe, from wearing it.

Most of what shaped this product came from arguing with people who knew something I didn’t: what the panel could render, what the ODM could hold, what the firmware would allow.