An AI agent in a crisis team without a human is a risk, not a saving

In short

• What the agent doesn't know: He does not know all the facts that people in the company know, and the media have not yet learned them. He does not assess the legal consequences and relationships with partners. He does not feel the tone of the moment.
• Where the agent really helps: He monitors mentions and warns of a sudden increase in activity. He collects facts and compiles a map of the parties involved. He prepares draft statements for improvement. He keeps an overview of the tasks of the crisis team.
• The principle we apply: The agent proposes, the person decides. In sensitive matters such as publishing, financial obligations, and legal matters, we require approval from a human being, and in the most serious cases, from two people.
• What does this mean for your company: Before you deploy an agent in crisis communication, record three things: what they are allowed to do on their own, what they are only allowed to suggest, and who approves them.

Agent AI is a program that performs tasks step by step itself: it searches for mentions, compiles the answers, and proposes actions. In a crisis, it is tempting to give it more: let it write the answers itself, publish them itself, and turn off comments itself.

What the agent doesn't know

• He does not know all the facts that the people in the company know, and the media do not yet know them.
• It does not assess the legal consequences and relationships with partners.
• One does not feel the tone of the moment. A correct sentence on an ordinary day can be tactless on a day of tragedy.
• He is not responsible for the decision. The decision is the responsibility of the president or the spokesperson.

Where the agent really helps

• It watches for mentions and warns of a sudden increase in traffic.
• It collects facts and organizes a map of the interested parties.
• He prepares draft statements for improvement.
• He keeps an eye on the tasks list of the crisis team.

The principle we apply

The agent proposes, the person decides. In sensitive operations such as publishing, financial commitments, and legal matters, we require human approval, and in the most serious cases, two people. Each decision leaves a trace: who, when, and on what basis.

What does this mean for your company?

Before you deploy an agent in crisis communication, record three things: what they are allowed to do on their own, what they are only allowed to suggest, and who approves. If it’s not written down on paper, it will be improvised in the crisis.

Do you want to prepare such a division of roles? Free call: [email protected], tel. 665 379 071.

Three levels of agent autonomy

For each task that you want to assign to an agent, select a level:

1. Just a preview. The agent collects and organizes the information; he doesn’t send anything. This is a good start.
2. Proposal for approval. The agent prepares a draft of the response or a plan of action, and the person accepts it, modifies it, or rejects it.
3. Self-made. The agent acts alone. Leave this level for internal tasks that are reversible and without legal or reputational consequences.

Publication, financial obligations, statements on behalf of the company, and everything related to law are at the second level. In the most serious cases, two people approve.

Example of division of roles for the first 24 hours

A hypothetical division that you can tailor to your company. Agent: monitoring mentions, compiling facts from sources, a draft statement, a list of questions the media may ask. Human: assessing the veracity of the facts, deciding on publication, contacting the media, speaking with employees and partners. Together: reviewing the stakeholder map and planning the conversations for the next day.

What to write down before the crisis happens

• The list of tasks that an agent can perform alone, and the list of those that they cannot.
• The name and surname of the person who approves each type of communication, and their deputy.
• Method of recording the decision: by whom, when, on what basis.
• The condition for the agent to be deactivated is when it starts to operate in a way that is inconsistent with its intended purpose.

The question „what do we do when an agent makes a mistake?” is worth asking before the crisis, not during it.

Jak wyglądają nasze prace, zobaczysz w projektach Commplace®.

See also

Related articles: AI agent in marketing and PR PR strategy in a crisis. If you need support, take a look at the offer: crisis management Commwork®AI.

Editorial position of Commplace. Material prepared with the participation of AI and verified by a human.