Intent, Operate, Check: a model for organising work with agents

René Deist was CIO and CDO at several multinational groups and brings the perspective from Shanghai, Paris and San Francisco with him. The episode's opening question: will the collaboration of human and AI be a symbiosis or rather parasitic.

His picture of the engine room of life hits the core: AI has long been working invisibly in the background, in the noise suppression on the podcast audio, in the automatic translation of a website. Now it is moving closer to value creation.

The gap between wanting and being able

The McKinsey figure is the most telling one in the episode: 92 per cent want to increase their investments, one per cent considers itself ready.

That is not a figure about technology, it is one about organisation. Being ready means: data is available and accessible, responsibilities are clarified, staff are trained, and there is a way to check a result and answer for it.

Anyone investing without clarifying these points is buying tools into an organisation that cannot deploy them. That is exactly what the gap between 92 and 1 describes.

The most uncomfortable finding in the episode concerns people starting out. Classic entry-level jobs are disappearing because agents take over precisely the research and support tasks on which people used to learn.

That is a structural problem for which there is no answer yet. If the tasks on which experience is built fall away, the experienced people will be missing in ten years. A company can solve this connection for itself by deliberately preserving entry-level tasks. The market as a whole does not.

Why prompt engineering becomes more important

Against the widespread expectation that better models make precise formulation superfluous, this episode puts a counter-argument.

On top of that comes the view of agent-to-agent communication. When agents delegate tasks to one another via model cards, humans increasingly lose insight into what is actually being negotiated between the systems. That leads straight to explainable AI, meaning the question of how to make a result comprehensible when nobody observed how it came about.

The model

The core of the episode is Deist's organising model: Intent, Operate, Check. Humans define the strategy and check the result, agents take over the operational middle part.

That sounds sober and has consequences for the organisational structure. If the middle part is automated, the roles shift to the edges: who formulates intents, and who answers for the check.

For IT departments that means a new task. In future they provide agent platforms and have to assess their learning progress. That is something other than operating systems, because an agent changes while it runs.

Note that the model was developed further in later episodes. Operate became Agent Performance, Check became Human Check, because it turned out that the middle part no longer has a fixed sequence.

Trust

At the end the conversation arrives at the question that runs through the whole series. Deist's formulation: trust is the last interface challenge that has to be solved before people rely on autonomous systems, whether at the operating table or in corporate procurement.

The sentence carries because it locates the problem in the right place. Trust does not arise from accuracy alone. It arises from comprehensibility, from reliable behaviour in edge cases and from someone standing behind it when something goes wrong.

Conclusion

This episode supplies the basic model that is drawn on repeatedly later, and three immediately checkable points.

Clarify whether your organisation belongs to the 92 per cent or to the one per cent. The question is not decided by tools, but by data access, responsibilities and paths for checking.

Deliberately preserve entry-level tasks, even when an agent completes them faster. Otherwise the level above them will be missing in a few years.

And for every automated process, define who formulates the intent and who answers for the result. That is the short version of the model and the one commitment without which none of the others carries.