A colleague's mother opened her kitchen cabinet to show us the water purifier behind it.
I want to stay on that, because it took me a long time to notice what was odd about it. She was not proud of the machine but happy that it gave the family clean water. She opened the cabinet door the way you'd open a cupboard to show a guest your good crockery. And the machine was inside a cabinet with the door shut.
This was a customer visit in Bangalore, 2022.
We asked if we could fill our bottles and she said yes and took mine off me before I could do it myself, which is what happens in an Indian kitchen if you are a guest and you try to do anything. It was a 750ml bottle. She held it under the spout with her right hand, and somewhere around halfway she moved it to her left, and then back to her right, and she did not appear to notice that she had done any of that. Then she tilted her head sideways to look at the gap between the mouth of the bottle and the spout, because the spout sat too low for the bottle to go under it straight, so she was holding it at an angle and reading the water level from the side. When it was nearly full she pulled it away and wiped the base with a cloth that was folded on the counter next to the machine. The cloth was there for that.
I have been in enough kitchens since to know that none of that was unusual, and that she wouldn't have been able to tell you she was doing any of it.
You can tell a lot about a kitchen by watching someone fill a bottle. Which hand takes the vessel and which one hovers. Whether they bend their neck or their knees. How long they wait before they look up to check, and whether they're looking at the water or at the gap. I had no language for this at the time. I just kept seeing it in house after house, and at some point I started photographing other people's counters, which you can do on a customer visit without anyone finding it strange, because everyone assumes you're photographing the machine for data collection purposes.
You see, water purifiers stick out. That's the first thing. They're taller than they need to be, or wider, and they sit at the end of a counter where nothing else is, or on a stool, or on a bracket on the wall with the pipes visible and a bit of pipe looping down. And then people cover them. A cloth over the top. A stitched plastic cover, the kind you get for a mixer grinder, cut down. Inside a cabinet, door shut.
Which is strange, because the machine is doing more work than anything else in that kitchen. It touches chai, rice, dal, soaking, pooja, the glass of water you drink before you sleep. As long as it works nobody mentions it. The day it stops, the whole house is in a mild panic, and someone is on the phone within the hour.
We'd been servicing these machines at Urban Company for years by then, so we weren't guessing about the failure modes. We had the service requests. Filters going every four to six months routinely. Customers who had spent close to the price of the machine on parts and service visits inside eighteen months and were still paying, because the next filter was already due.
At some point somebody asked what would happen if instead of servicing these, we built one. I don't remember at what point this started.
Nobody in the room had designed hardware. Urban Company is a services platform, we send people to your house, that's the business. But we'd spent enough time standing in enough kitchens by then to be asking a different set of questions, so through 2021-22 we ordered every purifier we could get hold of and then we opened all of them.
Some brands were packaged better than others and felt better. But it was the same story when we opened them. Filters were wedged into whatever space was left after everything else was placed. Pipes with kinks in them, actual kinks, bent past their radius and left that way. RO membranes smaller than they should be for the claimed output. A generic carbon or a taste enhancer with no documentation anywhere on the machine or the box about what was in it or what it was supposed to be doing.
And a printed circuit board held down with a zip tie fastener.
I keep coming back to that one. This is an appliance that is plugged in and running in a kitchen, near a sink, twenty-four hours a day, for six or seven years. And the board that controls the pump is held to the chassis with a zip tie, because somebody decided a boss and a screw were not worth the cost of the boss and the screw. Most of the warranties on these machines, incidentally, were tied to you buying servicing.
Anyway, so we started from zero, and the brief was one line. Build a purifier that doesn't need an annual maintenance contract.
None of us knew how to do that. We knew what we wanted though. We built it asking what Apple would do with it, which sounds like the sort of thing you say afterwards to make a decision look well thought through, but it was genuinely the closest to good products example we used in the room. It has to look good and not like anything else on the store shelves. It has to be worth the money that customers would pay.
I drew it in Figma, with rectangles
I drew the first concept of the machine in Figma using rectangles.
Engineering drawing was my favourite subject in college, and I was good at it, and then I spent eleven years not doing it. The last time before Native that I'd drawn anything with a physical thickness was copying concept vehicles out of Scott Robertson's books, badly, for fun. I'm still fascinated by concept vehicles and spacecraft and hybrids, basically anything mech, and none of that was any use to me, but it meant I wasn't frightened of the drawing part.
So I took a couple of fundamentals courses on Skillshare and Udemy, mostly for the nomenclature. How ribs work and why they're there. How injection moulding actually happens, as a physical process, with a mould that opens. Which material belongs in which part of a device and why nobody makes a tank out of ABS. I wasn't learning to design. I was learning enough to sit in a room with a design agency and a manufacturer and understand what they were telling me, which turned out to be the entire point, because for the first few months I couldn't tell the difference between a suggestion and a constraint. The more I knew, the better my decisions got. That's the whole of it.
I used to think joints were trivial. Capacitive touch, trivial. The load bearing capacity of the little channel a tray slides in and out on, obviously trivial. I don't now.
Water, and the six things you do to it
A purifier does five or six things to water depending on how you count, and every single one of them can be done badly, and in the machines we opened, usually was.
Start with sediment. Nearly every machine used one sediment filter, a single cartridge taking the entire particulate load, all the dirt and rust and grit in the line. Which means it clogs. It clogs fastest exactly where you'd least want it to, in cities with hard water or old iron pipes, which is a lot of the country. So we stopped using one and stacked several, an external sediment filter with different micron levels layered in sequence, so the big stuff stops at the top layer and the finer stuff carries on down to a layer sized for it. Then an internal sediment filter after that, for whatever got through the first one. It's not a clever idea. It just costs more, which is presumably why nobody was doing it.
Then the carbon filter. Brands were using granular carbon, or compressed blocks that weren't compressed enough, and we were measuring more chlorine in municipal water than we had expected to find. That matters more than it sounds, because chlorine attacks the RO membrane, so the cheap carbon block upstream is quietly shortening the life of the most expensive part downstream. We went to an OEM and worked on a high surface area block that holds up past ~25,000 litres.
The membrane had its own problem, which is that these machines were not flushing it. Salt accumulates on the surface, performance drops, the membrane degrades, you replace it. We put in an automatic flush that runs after every fill and again every few hours whether anyone's home or not. Of everything we did to the water path, that one change moved membrane life the most, and it's the least interesting thing to talk about.
The mineral cartridge was a mystery mix in almost every brand. Nobody would tell us what was in theirs. So we partnered on a block with a known release rate and then tested it for years years, which is a long time to spend confirming that something does what the spec sheet says it does.
Now, tanks grow microbes because water sits still inside a plastic box in a warm kitchen and that's what happens. Ours has a UV-C LED that activates every few hours rather than only when you dispense, so if you've been away four days the water is still clean when you get back, which is a scenario nobody designs for and everybody lives.
And then the PCB, the board that runs the logic. This was the most overlooked part in every machine we took apart. Almost all of them were relay-based, which means the machine has no real idea what it's doing, it just switches things on and off. We built ours from scratch. It counts filter life by how much water has actually gone through, not by the calendar, which is the difference between replacing a filter because it's finished and replacing it because it's April. It runs the flush, the pump, the UV, the TDS reading. It's on Wi-Fi. It'll tell you what's actually wrong with it, and it means someone books a service visit when there is something to service.
Form
The agency's first sketches came in and I didn't like them.
I should be fair about this: they were competent, and they were what we had asked for, which was a premium water purifier. Curved fascia. Tap mounted on the top. Hollow bezels. Chrome. It was ornamental design doing an impression of expensive, and it looked like every other machine in the category, and I sat with it for a day trying to work out whether I was reacting to the drawings or to the fact that I hadn't made them. Then I decided it was the drawings. I would not have put that object on my own counter.
So I started sketching myself, and spent a run of nights on industrial sketching tutorials so that I could say what I meant about a form and have it land.
Audi, not Toyota. That was the shorthand I kept using. No organic shapes, no rounded faces, no surface that existed only so the thing would look different from the last one. A rectangular body with the edges softened, clean vertical symmetry, buttons flush with the face, and a tray that read as part of the machine instead of a thing clipped onto the front of it.
Here's the underlying problem with the category. You cannot tell one Indian purifier from another. Take the logos off and put six of them in a row and you will not be able to sort them by brand. They look like they were designed by twenty different product managers across two different years, because to a first approximation they were. So the silhouette had to do the work a logo usually does: a vertically stretched rounded rectangle, symmetric, planted, sitting flat. I wanted somebody to see ours with the badge covered and ask what it was.
I referred to Braun for restraint, the MacBook line for how a surface should feel under a hand, Polestar for geometry (my favourite, and I know it shows), Copenhagen 's Bauhaus facades, for the specific trick of a flat front that still doesn't feel blank. And then a list of things we weren't going to do, which was almost as useful: no chrome, no gloss, no backlit icons, no transparent covers, and no stickers on the front of the machine listing its own features.
The agency took the sketches and the moodboards and turned them into CAD-ready form, which is the part I couldn't do, and we ran four or five directions and killed all of them but one. Foam board first, because it's fast and you can cut it at your desk. Then 3D printed shells from the manufacturer, to check fitment and assembly and how the thing actually felt when you put your hand on it.
The shadow
About ten foam boards in, it still looked bulky, and I couldn't see a way out of it.
The reason it looked bulky is that it was bulky. Our filters were physically larger than the standard OEM parts, deliberately, because that's where the filter life was coming from. So the body had to be deeper than the competition. Every good decision we'd made about the water was making the object worse to look at, and I could either give the depth back and lose the filter life, or ship something that looked fat next to a machine that was worse.
I sat with a foam board on a counter for a long time, on and off across days, not really getting anywhere.
What eventually landed is embarrassingly simple. It was a moment of epiphany for me. Kitchen light comes from above. Almost always, in almost every kitchen I'd been standing in for the past year, the light is a tube or a panel in the ceiling or under a cabinet, and it comes down. Which means the shadow a machine casts goes behind it, not in front. You don't see the shadow. And if you don't see the shadow, you don't see the depth that's making the shadow.
So we curved the rear surface, hard, much more than you'd curve it for any structural reason. The back of the machine falls away from you. The depth is still there, it's just in the part your eye has no information about. It reads about two inches shallower than it is.
That's the trick I'm most pleased with as well.
The prototypes were not good
I want to be straight about this part, because the version where you go from foam board to shipping product in a paragraph is not what happened.
The buttons didn't work properly. Not electrically, they registered fine. But the feedback was inconsistent, so you'd press and you wouldn't know whether you'd pressed, and you'd press again, and now you've dispensed twice. There is no way to describe how bad that feels in a spec, and no way to catch it except by putting your finger on it a hundred times.
The tray was too shallow. Which we knew, and had a reason for, and were wrong about.
And the LED ring didn't glow evenly. It had blind spots. Standing straight in front of the machine it was fine and slightly to the side it wasn't, and once you've seen that on a prototype you cannot stop seeing it.
We redesigned the tray, with venting, which we should have had from the start. Added optical light guides with diffusion behind the ring so the light spreads instead of pooling where the LEDs are. Roughly fifty other fixes around those three, none of which I remember individually and all of which took a week (or more) each.
We started design for manufacturing while the CAD was still moving, which our manufacturing partner in Pune had a number of opinions about, all of them correct and some of them expensive to act on. The front fascia went to a snap-fit so that no screw is visible on the machine anywhere. The tap cavity got redesigned because there wasn't clearance to get a tool in there. The button structure got tightened up to stop a wobble you could feel but not see.
The awkward one was touch. For capacitive touch to work at all, the board and the LEDs have to be on the front fascia, which means a wiring harness running from there across the body to everything else. That's more parts, more assembly time, and one more thing to go wrong in the field. We took it anyway, and then made the front panel removable and mounted the board onto the panel, so a technician takes the face off and the electronics come with it. It made the machine harder to build over months but it made it much easier to fix.
What it's made of
We picked every surface for the kitchen it was actually going to live in. In an Indian kitchen, oil will be in the air, steel utensils would be going down next to it. A damp cloth over it few times a month. Turmeric, which gets into everything and stains because Indian food.
We went ahead with matte ABS with a satin texture for the body. See, gloss photographs beautifully but then spends the rest of its shelf life reflecting a tube light straight back at you and collecting scratches you can see from far. Food grade PP for the tank. Polycarbonate for the tap lever, because it needs to be stiff and it gets pulled hard.
We launched in black, which honestly came off the Space Grey MacBook because I was 'CMF' illiterate back then. Black hides dust, takes fingerprints without showing them, doesn't go yellow after two years, and doesn't shift colour under the sort of light Indian kitchens actually have.
The logo we argued about for weeks, of course. Not the mark, the size of it on the device. We ended up making it small.
The name
About fifty names, over months. Some clever, some safe, a few genuinely bad.
The one we liked early was Better. It's short, assertive, clear, yes but it's comparative, and that's the problem with it. Better than what? Better than whose? You can't say it without implying the sentence continues, and we couldn't get past that.
Then one evening, over two Slack DMs, Native.
Every object looks right in the place it belongs, and this one was going to live on a counter in a kitchen, so I wanted it to look like it had been there since the kitchen was built. That's all the name is doing. There's nothing clever in it, which is why it felt right.
Two things nobody asked for
One-touch preset dispensing, and the tray. Neither came out of a research deck and neither came out of a feature list. Both came out of watching people fill bottles, and specifically out of the fact that people stand there holding a vessel and waiting, and that they put the vessel down somewhere wet.
Nobody had asked for either. Nobody asks for either. They're the two things people mention today.
We also put the Urban Company logo on the tray, where you'll only see it when you take the tray out to wash it, and a small light at the dispensing tap.
October 2023
We launched Native M1 and M2 in October 2023. Our own app first, then Amazon.
Then I, along with others who built it, stood on a stage in front of a hall full of journalists and talked about it, which I had never done, about a physical product! I don't remember it especially clearly. I remember the specific fear that somebody was going to ask me a manufacturing question I couldn't answer in front of the room.
I bought three right after. One for my in-laws in Udaipur, one for my parents in West Bengal, one for us in Bengaluru.
My parents saw it on the app first, because I sent them the link, and then in person when the installation happened. My father looked at it for a while and then said, "Oh, so you make ROs!!? This looks futuristic!"
For over a decade neither of them had been able to explain to anyone what I did for a living. Now my mother fills a bottle from it every day and doesn't tilt her head.
Then it needed a team
One machine is not a company, and one launch is not a roadmap, and I did not fully understand that until about a month after we shipped.
Shreyas was the first design hire, as an industrial design intern. He had full-time offers from other places, better ones, and came to us instead because he wanted to work on smart hardware from scratch and we were the only people offering that.
Krishna came second, a senior industrial designer with a mechanical engineering background. His portfolio was great and what actually got him hired was that in the interview he'd sketch a form and then immediately start talking about the fixture it would need, without being asked, as though the two were the same thought.
We had no special perks and no fancy office built for hardware creation. We didn't have a track record in hardware at all. What I had was a story, and I told every candidate the same version of it: imagine being one of the first fifty people at Apple. Native could be that for India, and you could be in that fifty.
Some of them believed it.
If you're early at a hardware company your fingerprints end up on the button feel, the hinge geometry, the mould angles, and which parts a technician can physically reach. Not many places can offer that.
What we didn't have was mechanical engineering depth. That's the thing I'd change if I could go back. We had industrial design, and product thinking, and a group of people who would push a date to get something right. But tolerance stackups and screw boss design need somebody who has done it before and can tell you in the meeting rather than in the second tooling revision, and we were learning it live, on a product going to market. We had good vendors and a good manufacturer covering for us. An internal design engineering team from the beginning would have saved us months, and I knew that at the time and didn't act on it fast enough.
So we started building industrial design and design engineering in house, because the alternative is pitching a change and waiting three weeks for the agency's CAD revision, and you cannot run a physical product that way. I wanted the person sketching in the same room as the person doing component R&D, and both of them within earshot of whoever is reading service tickets that morning.
Hiring too was slow and mostly disappointing. The portfolios were either generic or over-rendered, page after page of beautiful things that had never been made. So we started asking for the unglamorous files instead: exploded views, parting lines, tolerance stacks, BOM breakdowns. It worked.
Digital to physical
Before Native I'd spent over a decade on apps and interfaces and flows, and written something like a hundred PRDs, and none of it prepared me for this at all.
I didn't know what a core was. Or a cavity. Or what draft angle two mould halves need to come apart cleanly, or how ABS holds up in sunlight, or how much a PC+ABS part shrinks after it's moulded, which it turns out is enough to matter. The first time someone put a mould flow simulation up on a screen for me I sat and looked at a coloured image of my own product and had no idea what I was being shown, and said something random, and then went home and looked up what mould flow was.
So I went at it like a college student, which at thirty-five is a slightly odd way to spend your evenings. YouTube, books, ten-odd and still going, teardown videos, Skillshare, Udemy again. I sketched constantly, I took photographs of other people's appliances in their kitchens. I spent a genuinely unreasonable amount of time staring at remotes and mixer knobs trying to work out which way the mould had opened, and got it right maybe half the time.
Two things from digital design did carry over, and only two. Watching how people actually use a thing, which works the same on a tray as it does on a screen and is mostly a matter of shutting up and not helping them. And tracing a bad feeling back to its cause instead of fixing the symptom, which in hardware is just asking what job a part is really doing and whether it's the part that should be doing it.
The thing that made it manageable was giving up on understanding hardware as one job. Industrial design is how it looks and feels. Mechanical engineering is how it works and holds together. Manufacturing is how it gets made ten thousand times. I didn't need to be able to do all three, which was a relief. I needed to be able to tell which of the three a problem belonged to, so I could put it in front of the right person.
Hardware is slow and every mistake ships into somebody's kitchen and stays there for six to eight years.
Since then
A few people with a following called it the most beautiful water purifier in India.
The M1 and M2 gave us enough confidence to hire industrial designers properly and start on the rest of it what would then start becoming Native. I still think about doing a part-time industrial design & mechanical engineering course, something I could get through after work or over a weekend. I've looked, on and off, for three years.
My parents put me in drawing class when I was a kid. That turned out to matter more than anything else on this list.
I still don't understand hardware the way the industrial designers, mechanical engineers, and R&D on my team do. Feels like 2010 again, the year I started my professional career.