AI bootcamp for EU affairs, day 8: from prompts to agents
The next step is a system that can do the work again tomorrow. What an agent is, the four parts every useful one needs, and how to design a monitoring agent you can actually trust.
Sebastián Rodríguez Pérez is a European campaign strategist with direct experience running and advising campaigns for some of Europe's largest pro-European organisations. He founded the European Campaign Playbook to codify practitioner knowledge across the EU's 27 political environments and make it accessible to campaign professionals, advocacy teams, and civic organisations working at the European level. His work spans electoral strategy, EU public affairs, digital campaigning, and the integration of AI tools into political communication workflows.
Disclosure
Sebastián Rodríguez Pérez is the founder of the European Campaign Playbook and Campaign Intelligence Library. He has worked with pro-European organisations across multiple EU member states. He maintains editorial non-partisanship in all published content; where articles relate to campaigns or organisations he has directly worked with, this is disclosed within the article. He holds no current positions in active electoral campaigns.
What it covers
What AI agents are, how they differ from ordinary chats and projects, the four components every useful agent needs, how to decide what an agent should and should not be allowed to do, and how to design a simple public-affairs monitoring agent.
Who it is for
EU public affairs, policy, advocacy and communications professionals who already use AI for individual tasks and want to turn repeated work into controlled workflows. Especially useful for teams that monitor media, track policy developments, prepare regular briefings, or repeat the same research workflow every week.
When to use it
Use agents when a task is repeated, reasonably structured, based on identifiable sources, governed by clear rules, and useful enough to justify running without you starting every step. Understand the workflow before you automate it; an annoying task is not on its own a reason to build an agent.
Key takeaway
An agent needs four things: instructions, knowledge, capabilities and boundaries. Weaken any one of them and the agent becomes less useful, and potentially much riskier.
In brief
- A normal AI conversation responds when you ask it something. An agent pursues a defined objective across several steps and, where authorised, uses tools to act.
- The change that matters is autonomy inside a workflow.
- Every agent needs clear instructions: its role, objective, behaviour, audience and output.
- It needs trusted knowledge: the documents and sources it should rely on.
- It needs capabilities: web research, email, or access to another system.
- And it needs boundaries: what it must never do, when it must stop, and when a human must approve the next step.
- Public affairs monitoring makes a good first agent, because the workflow is repetitive, source-based and easy for a human to inspect.
- Give an agent autonomy because the workflow has earned it, rather than because the platform allows it.
- A bad prompt gives you one bad answer. A badly designed agent repeats the mistake every Monday morning.
Framework
The four parts of an agent
Weaken any one of these and the agent becomes less useful, and potentially much riskier.
| Layer | Label | Description |
|---|---|---|
| What job is this agent doing, and how? | Instructions | The agent’s job description: what it should do, how it should behave, and how it should produce its answers. |
| What information should it trust? | Knowledge | The trusted information the agent can consult when answering: approved documents, policy papers, previous work, the files you stand behind. |
| What is it allowed to do? | Capabilities | The additional things it can do beyond generating text: search selected organisational content, analyse information, send an internal email, update another system. |
| What must it never do? | Boundaries | The rules that define what the agent should not do, assume or claim, and the point at which it must stop and ask a person. |
From asking to assigning
So far in this bootcamp we have used AI one task at a time. You ask. It answers. You review. Then you ask something else.
Agents change that relationship. Instead of saying:
Find the latest news about this issue.
you can start saying:
Every Monday morning, monitor this issue, check these sources, identify anything important, prepare the report in this format and send it to me.
That is the shift from generative AI to agentic AI, and for public affairs teams it may be the most consequential shift in the whole bootcamp.
A bad prompt gives you one bad answer. A badly designed agent repeats the same mistake every Monday morning. That is why agents deserve more thought than prompts, and why today is about design rather than features.
Every prompt from this bootcamp is collected in the AI prompt library.