The Civic Case for AI

For as long as I have worked in public-sector and community-adjacent technology, the problem has rarely been ideas. The problem has been execution. A neighbourhood knows the intersection is dangerous. A family knows they cannot navigate the system. A region knows the record has been twisted. The knowledge already exists. What has usually been missing is a way to turn that distributed knowledge into working infrastructure without a grant, a committee, or a nonprofit with three years of runway.

AI changes the economics of that last mile. That is why I believe using AI to build civic technology is not a nice-to-have. It is the single most important application of these tools available to us right now.

Atlas Atlantic maintains three community tools as a public good. None of them is a business. Each exists because AI finally made it possible for one person, working with their neighbours, to build what used to require a full institution.

Grocery Saver helps people in a Canadian city find the best real-time deal on the food they actually buy, and assemble a practical shopping plan across stores. It exists because groceries in this country cost more than they used to, and because most of the savings apps in the market are optimized for retailer margin rather than for the household at the checkout. AI is what makes the difference. Ingesting flyers, normalizing wildly inconsistent product data across chains, and personalizing what matters to a specific family's list used to require a full data team. A single operator can now run that stack and keep it free.

Aidvocate is a systems navigator for people in crisis in Nova Scotia: substance use, elder abuse, youth at risk, housing and benefits, domestic safety. It is the highest-stakes thing Atlas Atlantic has built. Everything on the site assumes the user may be unsafe. The landing page has a Leave site button, an I need to be careful on this device path, and a private AI chat that runs without sending anything to a cloud server. That last point is not a feature. It is the product. In a crisis context, privacy is safety. The only reason a local, device-level AI experience like that is available to a person without a housing subsidy is because the underlying models got small enough, cheap enough, and open enough in the last eighteen months to run it that way. If we do not build these tools with that posture, someone else will build them with a posture that harvests data from the worst moment of a person's life.

Reckonize is a public record of accountability on Canadian national issues: housing, land, water, health, and power. It is built around threads like The Rent Cheque or The Boil Water Advisory, each one tying a lived experience (your rent went up again; you do the math) to a number ($1,550 a month on average; twenty-eight advisories still active) and to the decisions that produced them. The hard part of a project like this is not writing any single thread. The hard part is keeping dozens of threads cross-referenced, sourced, updated, and legible to a person who has fifteen seconds. AI makes it possible to assemble and maintain that public record at a pace that can keep up with the news cycle, with a single editor in the loop on every number. Ten years ago, this was a think tank. Today it is a website one person can steward.

Civic technology has always had a higher ethical floor than consumer software. People using it are often in states of stress, vulnerability, or distrust. They cannot be the product. They cannot be A/B tested for engagement. They cannot be nudged toward an upsell. Privacy, honesty, and durability are not differentiators in this category. They are the minimum to be allowed to exist. For a long time, clearing that floor was expensive enough that most good civic ideas did not ship. AI lowers the execution cost of clearing the floor. Local inference makes privacy cheaper than surveillance. Agentic workflows make editorial and data maintenance tractable for small teams. Pattern recognition across public data sets lets a lone operator see what used to require an institution.

There is a version of this argument that stops at caution, and I understand it. The energy and water footprint of the large-model stack is real, and I do not pretend otherwise. But refusing to use AI for civic purposes does not unplug the data centres. It just concentrates the benefit of those data centres in the hands of whoever is using them for something less useful: ad targeting, engagement optimization, platform extraction. Abstention is not a neutral posture. If we leave this surface to default players, the result is a world where the most capable coordination tools humans have ever built are used primarily to compete for our attention rather than to help us take care of each other.

I have written before about why I think Atlantic Canada is an unusually good place to do this work. The slower pace, the neighbourliness, and the tight social fabric here are not obstacles to building technology. They are inputs. Every one of the three tools above came out of a conversation on a sidewalk, a volunteer meeting, or a walk with my son. AI lets those conversations become infrastructure faster than they used to, and it lets that infrastructure be maintained by the same operator who had the conversation in the first place. That is a radically different supply chain for civic tools than the one we inherited.

The question I keep returning to is not whether AI will get used to build civic things. It will. The question is who gets to build them, on what terms, and with what values baked in. If the only teams willing to use these tools for community work are the ones optimizing for clicks or votes, we will get the civic tech we deserve. If operators who care about their neighbours are also willing to use them, we will get something better. That is why Grocery Saver, Aidvocate, and Reckonize are free and will stay free. And that is why I think using AI to build civic tech — carefully, locally, and with the people who will actually use it — is among the most important and most under-claimed applications of this technology available to us right now.