An AI talk should talk about real work

An AI talk can easily become a guided tour of today’s tools. That is useful for about ten minutes. Then the audience goes back to work with the same question as before: what does this actually change for us?

I prefer to start from another unit of analysis: real work.

AI does not enter abstract tasks. It enters organizations, jobs, teams, rules, responsibilities, existing tools, deadlines, exceptions, tensions and tradeoffs.

A useful talk should not only ask what AI can do. It should ask what AI redistributes.

What AI redistributes

When an assistant writes, someone still has to verify. When an agent acts, someone has to understand what it did. When a model summarizes, someone has to decide whether the summary is faithful. When a tool accelerates a task, someone has to check whether the workload has simply moved elsewhere.

The important topics are often less flashy than the demos:

• responsibility;
• quality;
• evidence;
• human recovery;
• cooperation;
• cognitive load;
• new dependencies on tools;
• the limits of automation.

These are the topics leaders, managers, domain teams and public organizations need when they want to understand AI without falling into fascination or rejection.

A talk as a decision space

For me, an AI talk should help an organization make better decisions.

It should give people a shared language. It should distinguish tools, assistants, agents and work systems. It should show why successful automation is not just a completed task, but an activity that becomes more controllable.

This is also why I connect my talks to ergonomics. Ergonomics studies the gap between prescribed work and real work. AI makes that gap even more important, because it can make things look simple precisely when supervision becomes harder.

The format I defend

A good format is not a flood of trends. It is a progression:

1. what generative AI and agents can already do;
2. what they cannot guarantee;
3. what they change in work;
4. the risk of invisible displacement;
5. the conditions for testing properly;
6. the decisions to make before deployment.

The talk page is here: AI keynote, real work and organizations.

For organizations that want to move from the talk to action, the next topic is situated AI integration: choosing a use case, limiting permissions, keeping evidence, organizing validation and planning human recovery.

AI is not only a technical topic. It is a work topic.