35 talks I'm planning to watch: AI agents, civic tech, and digital democracy
· 7 min read · updated · Hrönir rank #38/38
I have a habit of accumulating queues because it feels like the cheapest way to simulate progress.
Every weekend in Porto Velho, when the humidity climbs and the energy grid begins to sag under the weight of three thousand air conditioners, I open a new browser window and start bookmarking talks. It is a protective reflex. I work in a state attorney’s office where the backlog behaves like a physical liquid, filling every room, and the temptation is to believe that if I just watch Evan You explain Vite or Geoffrey Hinton discuss the limits of human intelligence, I will somehow be closer to building the tools that might save me.
The browser window now has forty-seven open tabs. It is a monument to intentions I haven’t earned yet.
Most of these are not recommendations. To recommend a talk, you have to have watched it, and to have watched it, you have to have surrendered forty minutes to someone else’s rhythm. What sits below is the queue — the annotated list of things I have decided to care about before the compiler tells me I’ve spent another weekend performing the act of research.
Two clusters. The first is software engineering and AI—mostly from 2025, when the word ‘agent’ went from a conceptual framework to a CLI tool you could actually run. The second is civic tech and govtech, which I’ve been calling ‘civil tech’ in my drafts until Canivete pointed out the term doesn’t exist. The name is civic tech, even if the tools often feel like they belong in a museum of good intentions.
Where the code meets the latent space
The AI engineering cluster runs from very practical (how to build reliable agents) to philosophical (what are we actually doing here). Dex Horthy appears twice because his 12-factor agents framing keeps coming up in conversations and I want to read the source, not the second-hand summaries.
What civic tech smells like when it’s real
This is the cluster I’m more interested in right now. I work in public administration in Rondônia and I’ve been thinking about what it would take for accountability data to actually reach the people who should be reading it. The Brazilian videos are where I start. The international ones are where I go for mental models that Brazil hasn’t built yet.
Brazil
Querido Diário: hoje eu tornei um Diário Oficial acessível para todo mundo — entry point to the project: municipal official gazettes made searchable and usable. This is the Open Knowledge Brasil project for extracting and centralizing official municipal gazette data with a public API.
Navegando na API do Querido Diário — the more useful one if I actually want to build something on top of the data.
Navegando no site do Querido Diário — less technical, good for seeing the public-facing UX before diving into the API.
Brasil.IO: Dados abertos para mais democracia — Álvaro Justen — Brasil.IO is a repository of Brazilian public data in accessible formats. A classic of Brazilian open-data infrastructure.
Ativismo digital e tecnologia cívica — Operação Serenata de Amor — older, but important. Serenata de Amor / Rosie used AI to analyze reimbursed parliamentary spending under CEAP and flag suspicious cases. One of the canonical Brazilian civic-tech stories.
DadosJusBr — I’m following the project more than looking for a single video. Highly relevant: structured remuneration and transparency data for the Judiciary and Ministério Público. The kind of thing that should be standard and isn’t.
Abroad
AI Studio: Solving government’s PDF problem with AI — Code for America Summit 2025 — PDFs as the graveyard of public information. Relevant everywhere, but especially in Brazilian public administration where the PDF is load-bearing infrastructure.
Putting policy to work: Guiding AI use inside government — Code for America Summit 2025 — procurement, guardrails, implementation. The better civic-tech version of “AI in government,” without the demo magic.
How Summer EBT is delivering on a new promise for families — Code for America Summit 2025 — concrete benefits-delivery civic tech. GetCalFresh helped 6.2 million people get food benefits from 2017–2025. That is the unit of measurement I want to be using.
vTaiwan: new experiments in digital democracy — Taiwan is probably the world’s most interesting case for civic tech that actually touched governance. The g0v ecosystem is decentralized, transparent, and has shipped things that worked.
How digital democracy can heal polarisation — Audrey Tang — the philosophical version: not just better forms, but different institutional affordances.
Decidim secret sauce: Building an international community — Decidim is a free/open platform for citizen participation. Infrastructure for participatory processes and assemblies.
Scaling Deliberation: Polis and the Computational Democracy Project — one of the most interesting “AI + democracy” lines: Polis tries to find consensus structure rather than maximize engagement. Opposite of social media.
The Department of Government Improvement: Civic Tech and Why It Matters — 18F/USDS framing: government improvement as craft, not ideology.
2025 State of Digital Public Infrastructure Report Launch — the broader “digital public infrastructure” frame: identity, payments, data exchange, public rails. The Digital Public Goods Alliance framing is useful here.
graph TD
PDF["The PDF (Graveyard of Data)"] -->|1. OKBR / Querido Diário| API["Searchable Public API"]
API -->|2. Brasil.IO / DadosJusBr| Structured["Structured Public Rails"]
Structured -->|3. GetCalFresh / Summer EBT| Service["Real Service Delivery"]
Structured -->|4. vTaiwan / Decidim / Polis| Deliberation["Deliberative Public Assembly"]The path through the thicket
The queue is not a database; it is a trajectory.
If I watch these in isolation, they are just thirty-five people talking in well-lit rooms. The sequence has to argue with itself. I’m starting with the civic tech cluster because that is where the friction lives for me right now. The watch order I’m following is a path from data extraction to actual governance:
- Querido Diário API — Brazilian data liberation, close to home
- Brasil.IO: Dados abertos para mais democracia — the infrastructure that makes the data usable
- AI Studio: government’s PDF problem — where AI meets public administration concretely
- Summer EBT / GetCalFresh — benefits delivery as the unit of real civic-tech success
- vTaiwan — the most complete case study of civic tech that actually changed governance
- Polis / Computational Democracy — the mathematical infrastructure for deliberation at scale
- Decidim — the platform layer for participatory processes
That sequence moves from Brazilian data liberation → public-service delivery → deliberative and democratic infrastructure. It ends at questions I don’t know how to answer yet, which is usually a sign the queue is worth the time.
The illusion of the queue is that the list itself is an achievement. You annotated it, you sorted it, you formatted the Markdown. It looks like a pipeline.
But a pipeline that doesn’t run is just steel sitting in the dirt. I know this because causaganha has been sitting in my repository for three years, and the moment I stop writing code, the official gazette returns to being a pile of unindexed PDFs. The queue will wait. The browser will eventually crash and release the memory.
The list is clean. The code is unwritten. Only one of them survives the weekend.
Related posts
GitHub: a tour of the repositories I maintain (and why they exist)
Sixty-four public repos, grouped by theme. A short tour for whoever lands lost.
Three Hammers Walk Into a Bar
On three professional postures, four alignment properties, and the one property that had to come from elsewhere.
The Agent That Doesn't Invent Verbs
On Cucumber, content-addressing, and an alignment technique that turns out to be older than alignment.
Comments
Comments not configured yet.