Albania’s AI ‘Minister’: What It Means for the Future of Government Work

Why does Diella’s appointment matter?

Volodymyr: What’s interesting is that this is one of the first highly visible examples where an AI agent is in a genuinely important national role. Diella started doing basic government service tasks online and within about six months was moved into ministry level position. It’s also worth noting they built the system with input from people who are currently or previously worked at OpenAI, removing a number of questions around how well tuned/reliable the agent actually is.

Beyond that though look at what the agent is actually doing. Procurement evaluation usually has quantitative underlying comparison (is one bid cheaper than another, is one delivery faster etc). That’s why the risk profile here is lower than people might think, most of the decision making is basically deterministic anyway (especially when clear criteria are already defined which is how it’s done in most public tenders anyway) and where subjective judgment is needed I’d expect the agent to be working from tightly defined instructions with built-in reasoning steps so that process can be explained and auditable.

Are there things that still concern you even in a use case like this?

Volodymyr: The thing I would flag is how the final score gets weighted. If you’re combining several factors (e.g. quality of materials, price and delivery time) into one composite score then someone still has to decide how much each factor should count for. How to measure/set the right weights or impact of different metric parts, like materials quality vs price vs time to complete? Still even with that caveat this is a good first step. It really has the potential to improve procurement decision making and reduce human bias in those processes.

What happens if this kind of AI involvement becomes the norm across governments?

Volodymyr: If half the world’s governments start handing real functions (maybe ministries) off to AI agents that’s almost science fiction territory, agents from different countries interacting with each other, competition over which agent is better tuned or instructed and then harder question of how errors and their severity affect public trust over time? You can easily end up in a feedback loop: more trust = more agents deployed but more errors erode that trust and pull adoption back the other way.