In autumn 2025 I started a project on my own initiative. No client, no brief and no fee, which meant the subject was mine to choose.

The first question I asked myself wasn't about the sector but about value. Over twenty-five years I've built a lot of products, and some of them were yet another app of the kind there are a million of, or a site for a company that wants to earn a bit more and is prepared to use dark patterns to do it. I wanted a problem where the benefit is real and the count runs not in thousands of users but in a country.

Hence the choice of field. Agriculture is food, the most basic need there is. There's more to it than food, of course — coffee, exports, revenue. But at the bottom of it lies what feeds people, and it's hard to think of an area where the benefit is more obvious.

That was the first design decision. Not "I'll build a system for agriculture" but "I'll take a problem where the scale and the benefit are real", and agriculture came out of that.

What followed was six months of work, a concept for a national agricultural infrastructure in El Salvador, and a meeting with representatives of the FAO, the Food and Agriculture Organization of the United Nations. Below is how it was done: which decisions were made and why, where the time went, and where the work ran up against things that can't be settled remotely.

Why El Salvador

I know the region first-hand. We have family in three Central American countries — Guatemala, El Salvador and Nicaragua — and I've lived in each of them several times. It isn't a tourist acquaintance, and it saves a great deal of time at the start: you understand how things work day to day before you open the first report.

Of the three countries I chose El Salvador for these reasons.

It's small. Decisions there are taken faster than in a large country, and the path from idea to implementation is shorter. For a concept that could in principle be built, that matters more than market size.

And it's more technologically forward than its neighbours. In recent years the country has shown it's willing to make fairly bold digital moves, which means a conversation about national infrastructure won't stall on the premise itself.

There's a third reason, a personal one. I find it more interesting to work for countries that have room to grow than for those where everything already exists and the only question is money.

Three months before the first screen

The first three months went into studying the ground.

I started with the agency websites: what they publish, how it's presented, what can be seen through the eyes of someone on the outside. There turned out to be little in the open. Some services are closed, some require registration with local documents, so how they work on the inside I don't know and won't claim to. Of what can be worked with without access, the most useful was the digital soil map.

Alongside that I looked at other Salvadoran public services, including the Chivo wallet, and paid attention to their house style — that dark blue. A concept has to look like part of the state system rather than a product from outside, and its visual language has to be recognisable.

A separate large piece was other people's agricultural products. Apps for farmers, drone systems, crop-monitoring platforms, produce marketplaces, services for cooperatives. Dozens of interfaces, and I wasn't looking at them for inspiration but to understand three things: what already exists in this field, what recurs everywhere, and where the generic filler begins that helps nobody.

The most useful findings were the recurring solutions. When the same thing turns up in five different products from different countries, it usually means the problem is real. And when an elegant idea appears exactly once, that's a reason to look twice: it may be new, or it may have been tried and gone nowhere.

And the dullest but most useful part: the structure of the agencies. Who does what, who answers for what, how they relate to one another — down to specific names and roles. Without that you can't say who in the system a given action is addressed to.

I wrote all of it down and structured it as I went. What accumulated was a working file and a body of material several times larger than the case study on my site: wireframes, three branches of interfaces — for the producer, for the state and the public one — a separate block of descriptions for each section, and a reference library sorted by theme: colour and style, water, drones and markets, field monitoring, dark interfaces, energy.

Some of that belongs to the current concept, some was gathered for later, for sections the work hasn't reached yet. If the project goes ahead tomorrow, all of that groundwork is already in place.

And every place where information is missing and where someone on the ground needs to be spoken to is marked separately. There are a lot of those marks. They turned out to matter more than I expected at the start.

Every system lives on its own

The first thing that stood out: soils live on their own, climate on its own, markets on their own, applications on their own. Each system does its own job and stops there. A place where the data of one turns into an action in another is something I didn't find.

So the central idea of the project formed early and barely changed afterwards. Not an app for farmers and not another dashboard, but an ecosystem where data on the agricultural sector, food security, climate and risk come together.

The practical consequence is simple, and it was my main argument. When the data sits in one circuit, the cabinet doesn't have to commission reports from separate agencies and wait for them to be reconciled. The picture assembles itself, and an analytical layer on top points at what to look at.

Agencies are rarely connected to each other — that's a property of how states are built, not of El Salvador. Which is why a system that joins up what matters has value in its own right, whatever the country.

I'm not an agronomist, but I grew up working a plot

Formally I'm not an agronomist. But the land is familiar to me from more than reports.

I grew up working a vegetable plot. Potatoes, cabbage, tomatoes, cucumbers, courgettes, apples, currants, onions, dill. I built the beds myself, and from about the age of ten I planted and hoed some five hundred square metres of potatoes on my own. It was part of life, not a weekend at the dacha.

The coffee plantations of El Salvador are a different scale, but land behaves the same way. The harvest depends on what you did three weeks ago. A missed moment doesn't come back. Advice in general terms is useless until it's applied to your own piece of ground. When you know that with your hands, it's easier to tell an indicator someone will act on from an indicator that merely looks good on a screen.

The digital side of the sector I worked through from scratch. That's easier today than it was ten years ago: I put questions to AI, then checked the answers against sources and across several different models. It went reasonably deep.

How much ground a single agricultural drone actually covers, and when you need two. What types of agricultural drones exist and how they differ in practice. How coffee rust is detected from imagery. Whether the result of a flight can be shown on a phone, and what is visible that way and what isn't.

Separately I went into soil moisture: how it's measured, what counts as normal, at what point the figure becomes grounds for action. And into food security indicators — which of them mean something at cabinet level and which merely look substantial.

That is a significant part of an architect's work. Not assembling an elegant diagram, but acquiring enough subject knowledge for the diagram to mean anything at all.

Some questions stayed open. Above all those around the offline mode: how exactly it should behave in real conditions isn't something you can establish remotely.

And I don't think it's an accident that I was drawn this way. Someone with five hundred square metres has no use for a national dashboard, but a way to understand what's happening with the land and what to do this week would have made life a great deal easier. I may well be building the tool I lacked myself thirty years ago.

How long the invisible part takes

Now about the timings, because this is where people on the outside usually have the most mistaken picture.

Design is assumed to be work on screens, with most of the time going there. For me it came out like this.

About three months on research, gathering information and the design proper: how the data is organised, what the layers are, who works with what.

About three months on the screens: how it all looks, the dashboard, the producer's app, the public part.

Plus around a month on putting together the detailed case study, though that's packaging rather than design.

The figures are approximate: I wasn't rushing and worked at a steady pace, alongside paid work.

But the ratio is telling. Half the time went into what doesn't show up on a single slide. Screens are how a decision is presented; the decision itself is taken earlier, and costs exactly as much.

Where to stop

A concept has no floor. The deeper you go, the more detail appears and the more interesting it gets: you could add a marketplace, you could expand the drone side, you could work through another dozen scenarios. Far more ideas accumulated than made it into the final piece.

So I set the boundary in advance: the core analysis plus a defined number of basic screens. The plan had more of them, but that would have pushed the work towards a year.

Six months for a concept is already a lot. I could have gone faster, but chose not to push: a project without a client is good precisely because you can think rather than deliver.

Everything worked through stayed in the materials. But much of the groundwork and analysis didn't make it into the case study, because checking it requires conversations with people on the ground.

The meeting

For the meeting I put together a separate presentation, tighter and more concentrated than the detailed case study on my site. People are busy, and the less of their time goes on looking at things the better. That's a design decision too: material for a site and material for a conversation are built differently, and they work better assembled separately. A site someone goes through alone and at their own pace; in a meeting the first minutes decide everything.

I presented in Spanish. I was nervous, naturally: my Spanish is weaker than my English and the audience was a serious one. So I prepared the talk in advance and read most of it from the text. After that the conversation ran mainly in English, with the leads on the FAO and MERIAN side.

What I took from that meeting, beyond the fact of it happening: screens occupy little space there. There were questions about interfaces, but time was short and the bulk of it went elsewhere.

What dominated was reach. According to the CENTA leadership, the service currently reaches around twenty-five thousand producers out of more than four hundred thousand, roughly six per cent. It works through specialists who travel to farms and advise on site. Year after year they retire, hiring new ones gets harder, and a digital channel remains CENTA's only way of substantially widening that reach. Seen in that light, the interest in the concept made sense to me: not as an elegant idea, but as an answer to a specific piece of arithmetic.

The second thing said was almost word for word where my work had started. The country has many systems and a great deal of information in them, but they aren't assembled around the tasks of the person using them. There's a soil database with twenty years of metadata, an award-winning one, and in day-to-day work it goes unused. A producer doesn't know where to go for weather data or how to read it.

I had reached that conclusion from the outside, through my own experience and analysis of open sources. They live with it from the inside. That the diagnosis matched was the most valuable result of the meeting for me.

After that we talked about timings, the technical side and the name. The last was unexpected: Chivo Agro is a working name for the concept, but in the country Chivo reads as the wallet, and that agenda is in the past. Which means the name is unlikely to fit what the government is communicating now. I'd thought about the brand as a designer rather than as someone looking at the state's agenda as a whole.

We also discussed pace. Things move slowly in international organisations, and an answer, if it comes, won't arrive before the end of the year and more likely in the next one.

Where the concept ends and the project begins

All of this work was done remotely, from open sources. I haven't spoken to producers, to CENTA staff or to anyone in the ministry. That is precisely the line where the concept ends.

It does its job all the same: it shows how the system could be built and verifies that the structure holds together. What begins next is gathering data on the ground, and that changes the detail. Some decisions will be confirmed, some will fall away, and tasks will appear that aren't visible now.

Which is why the materials mark out separately where information is missing and who needs to be spoken to about it. In effect it's a ready brief for the first stage of real work: not a blank page, but a specific list of questions and people.

How accurately work done from the outside lands in reality was shown by the meeting. The same problem and the same figures I heard from people who deal with it every day. None of it is on the surface; it emerges only through experience and sustained analysis.

And a project on one's own initiative gave what work to order doesn't: the right to choose the problem, to think about it as long as it takes, and to decide for myself where enough is enough. For a concept of this scale that turned out to be a condition rather than a luxury.

Projects of this kind are my field, and I'm open to proposals.