The Cities Engine is friendly on the surface, but there's real power a step below it. This piece is for the users who want to get the most out of the platform: turning the assistant into an analyst, searching by meaning, and — for the team maintaining the data — growing and governing the directory with AI.
João as an analyst copilot
Most people meet João as a helpful guide. Used well, it's closer to an analyst who never gets tired. Because João can call the same ranking and analysis tools the platform runs internally, you can ask genuinely analytical questions — “rank high-capacity tropical cities by recycling rate”, “compare Nairobi and Accra across environment and waste”, “show the waste composition for this archetype” — and get back a live leaderboard, a side-by-side radar, or an inline chart rather than a paragraph of hedged prose.
Two design choices make this trustworthy. First, João never guesses at numbers: when a question depends on data, it resolves the city, runs a defined tool, and reports what the tool returned. Its analytical queries are deliberately bounded — grouped summaries over an allow-list of dimensions, not free-form access to the database. Second, anything that changes data is human-in-the-loop: João can propose an edit, a new task, or an approval, but the action is shown to you for confirmation and re-checked on the server before it runs.
What João can do also depends on who you are. A community member can explore everything and submit feedback but cannot alter platform data; a staff member can propose edits and create tasks that flow into an approval queue; an administrator can make direct changes. The assistant's personality is the same for everyone — the permissions behind it are not.
Search that understands meaning
Ordinary search matches words. The Cities Engine also searches by meaning. Every organisation in the directory is converted into a mathematical representation — an “embedding” — that captures what the group is about. When you search semantically, your query is converted the same way and matched by closeness, so a search for “groups helping informal waste pickers” can surface the right organisations even when they never use those exact words. If you'd rather find an exact name, the platform falls back to conventional text matching. (Today this meaning-based search is tuned for organisations; extending the same idea to find look-alike cities is on the roadmap, not yet shipped.)
The AI discovery console
A directory is only as good as it is complete and current — so the team maintaining it has an AI Control Panel with five areas: City Discovery, Enrichment, Embeddings, Settings and Debugging.
The most powerful is discovery. Point it at a city and an automated pipeline (running off-platform for scale) drafts locale-appropriate search queries, searches the live web, removes duplicates against what's already known, vets each candidate with an AI model, enriches the ones worth keeping, and follows partner links one hop outward to find related groups — repeating until it stops turning up anything new. Everything it finds lands in a review queue: a person decides what actually enters the directory. The console also manages enrichment (filling in missing details on existing records), regenerates the embeddings that power semantic search, and exposes provider settings, spend controls and an error log for debugging.
Keeping track of what changed
For teams that want to plug the platform into their own tools, a small, read-only interface exposes exactly three things: a monthly “what changed” summary, bounded analytical queries, and population evidence. It's intentionally narrow — it can report, but it can't alter anything — which makes it safe to wire into dashboards and monthly intelligence briefings.
The throughline
The advanced features share one principle: put real analytical power in front of people, while keeping the guardrails honest. The assistant is capable but permissioned; the automation is ambitious but always ends in human review; the integrations are open but read-only. That's how the Cities Engine stays both powerful and trustworthy.
If you skipped ahead to this one, the overview and the feature tour cover the ground beneath it — and how we built the Cities Engine explains the machinery underneath.