Working on artificial intelligence with an administration requires broadening the question. It’s not enough to know if a technology works. It is necessary to understand who serves, as justified, what collectives can affect, who takes responsibility and how capacity will be maintained over time.
Work linked to the Plan Director of AI Terrassa It has reinforced several learnings that are useful beyond a single project.
Listening Before Prioritizing
A public strategy cannot be built only from technology. It needs to collect the needs, concerns and proposals of people with different experiences: municipal teams, economic agents, expert knowledge and citizenship.
Participation is not a decorative validation of an already written plan. It can reveal problems that had not entered the map, change priorities, and show where confidence or understanding is lacking.
Separate opportunities, proposals and commitments
In open processes many ideas appear. It is important to explain in what state each one is: contribution, proposal pending prioritization, line of work or approved project.
This accuracy protects credibility. Presenting a proposal as if it were already an implementation generates expectations that the institution has not yet assumed.
Governance is part of the infrastructure
Governance is not a final document. It should appear in the selection of use cases, data, purchases, permissions, human controls, evaluation and public communication.
A governed system lets you know who decides, with what evidence, and what happens when the outcome is wrong. This traceability is necessary in any organization, but in the public sector it has additional democratic value.
Build internal capacity
An administration cannot depend indefinitely on a provider to interpret each decision. It needs people who are able to understand the possibilities, formulate projects, review results and maintain common criteria.
The training must combine general AI culture with practice linked to real roles. It also needs coordination spaces so that learning in an area is not isolated.
Use cases should be valued for public value.
Efficiency matters, but it’s not the only measure. It is also necessary to observe accessibility, quality of service, fairness, transparency, trust, working conditions and environmental impact.
Some projects can save time; others can expand access to information or help detect needs. Prioritization must explain these criteria.
A useful strategy is a decision-making capacity.
A AI Master Plan You shouldn’t freeze a list of tools. You should leave a way to decide: principles, roles, criteria, lines of work, mechanisms of participation and review.
Technology will change. What must endure is the institutional capacity to ask why, for whom and under what conditions it is used.