Diplomacy in the Age of AI Agents

A diplomat opens an artificial intelligence (AI) chatbot, pastes a draft resolution, asks for a summary, and closes the tab. The next day, the same routine. The AI waits to be asked. It never monitors developments. It never delivers a briefing without being prompted. It responds when spoken to, then goes silent.

This is AI as a passive tool. The diplomat must remember to ask, must know what to ask, and must do all the work of deciding what needs attention. The AI just answers.

Meanwhile, the same technology can monitor developments across six committees simultaneously, track which delegations shifted their language in yesterday’s debate, and deliver analysis before your morning coffee without being asked. It can work while you sleep. It can surface what you did not know you needed to know.

The gap between what diplomats do with AI and what AI can do for diplomats has never been wider. We are using AI like a taxi service when we have access to a fully autonomous self-driving fleet.

The real opportunity lies in AI agents: persistent, proactive assistants that operate as part of daily diplomatic practice. The shift is from AI that waits for prompts to AI that takes initiative, anticipating needs, monitoring developments, and handling preparatory work so diplomats can focus on political judgment, relationships, and negotiation.

© Unsplash/Alexander Shatov

What AI Agents Can Do in Diplomatic Practice

AI agents can replicate a diplomatic team structure. A policy analyst reads publications, tracks statements, scans media in multiple languages, then delivers a morning briefing tailored to your portfolio at 7 a.m., every working day, without being asked. A meeting analyst produces interpretive summaries, identifying main themes, attributing positions, and offering political assessments. A drafting counterpart understands the nuances of diplomatic language (“took note of” is neutral, “welcomed” implies endorsement, “condemned” is the strongest rebuke) and produces first drafts respecting these conventions. A coordinator ensures these assistants work as a system.

The critical point is that these assistants work together at a scale and speed that human teams cannot match.

Three Use Cases That Could Change How Diplomacy Works

1. Continuous Intelligence

The diplomat drafts a position paper. An AI agent surfaces three precedents, flags that the proposed language contradicts a statement from two months ago, and suggests alternative phrasing. The diplomat remains in charge and now has an advisor who has read everything, remembers everything, and never takes a day off. Instead of occasional assistance, the diplomat enjoys the continuous accompaniment of AI agents, drastically amplifying capacity without replacing judgment.

This solves a structural problem: when diplomatic posts rotate every three to four years, accumulated knowledge usually leaves with the person. With an AI agent, the incoming diplomat can ask, “What was our position on the last draft?” and receive an answer drawn from the actual record. The institution develops a memory and a partner.

2. Sleepless Intelligence

You sit in one meeting. While you are there, four things happen: a parallel session convenes, a revised draft circulates, your capital requests urgent analysis, and a think tank publishes a contradictory report.

In the current reality, you make a choice. You attend your meeting. The rest will wait, or you will miss it.

With AI agents, this changes. During the break, you open your phone. Waiting: a summary of the parallel session; the revised text annotated with changes affecting your stance; a first draft of the capital’s requested analysis, structured and sourced; an alert about the report with relevant pages extracted.

None of these outputs is final. Each requires your judgment. But the impossible day has become manageable. You are choosing what to refine, not what to ignore.

This is intelligence that never sleeps: analytical capacity distributed across the entire agenda. A delegation of three diplomats with well-configured AI agents can cover ground that used to require fifteen.

3. Anticipatory Intelligence

Before committing publicly, diplomats need to anticipate reactions. This currently happens through informal consultations and accumulated experience.

AI agents can create a strategic rehearsal space: a diplomatic sandbox where you test positions before committing. Multiple agents, trained on historical positions and voting records, simulate how delegations might respond to proposed text.

Before proposing new language, you run the draft through a simulation. You see which phrases trigger objections, where there is flexibility, and which delegations might be persuaded. You refine based on the simulation and enter the negotiation better prepared.

This is not a prediction, but it shows what delegations have consistently done in similar situations, allowing you to explore options before committing political capital.

Where the Diplomat Remains Irreplaceable

AI agents can perform diplomatic tasks, yet core aspects of diplomacy remain irreplaceable.

Political judgment: Knowing when to push or concede, when to speak or stay silent, and when ambiguity serves the negotiation. An AI agent can tell you what every delegation said. It cannot tell you what they meant, or what they left unsaid.

Relationship building: Trust is built in hallways, over dinners, through years of knowing someone. AI has nothing to offer here.

Accountability: When a diplomat sends a cable, they put their name on it. Institutions must clearly state who is responsible when an AI-assisted analysis proves wrong.

Confidential work: Any system that processes diplomatic information raises security questions. The question of how to build secure AI tools for diplomatic use is serious.

The Risks

Over-reliance could flatten the nuance that distinguishes good diplomacy from adequate diplomacy. If assistance becomes dependence, the risk is erosion of the diplomat’s own capacity to write, analyze, and think under pressure.

If every delegation uses similar tools, the diversity of perspectives could narrow. Multilateral diplomacy thrives on the friction of different worldviews. If AI systems converge on similar framings and language, we risk losing the diversity that makes negotiations productive.

A Call to Action

The shift from chatbot to agent is happening now. The technology is available and accessible. What is required is institutional imagination and care.

Governments should establish pilot programs to test AI agent architectures in low-stakes environments before deploying them. These pilots should involve diplomats and technologists and be designed to fail safely: to surface risks, identify limitations, and build institutional knowledge.

I have been exploring AI agents applied to diplomacy over recent months. I built virtual embassy simulations, modeled international institutions, and ran cross-linguistic benchmarks across 100 languages. What would have taken months of manual work was completed in weeks. These were early and imperfect efforts, but sufficient to witness the transformative potential of AI agents to replicate the structure and dynamics of diplomatic practice.

The diplomatic community must actively shape the deployment of agentic AI. The opportunity is significant. The risks are real. The decision should not be made by default.but

Tono Pérez-Hernández is a diplomat covering AI governance and digital affairs. He served as advisor to the co-facilitators of the General Assembly resolution establishing the International Independent Scientific Panel on AI and the Global Dialogue on AI Governance. The views expressed are his own.