Typology 2: The Gatekeeper
(Low steering + in-depth understanding)
The Gatekeeper is situated in the quadrant of low steering and in-depth understanding. It is a position in which professionals do possess strong domain and process knowledge, but actively manage AI processes to a relatively limited extent. The Gatekeeper relies heavily on the system’s capabilities and tends to intervene reactively, particularly when the output does not sufficiently align with the professional’s intentions or when problems become apparent.
This position highlights that in-depth understanding does not, in itself, automatically lead to a high steering score. Professionals may have a very strong substantive grounding, whilst at the same time allowing AI to function relatively freely during the work process.
What characterises the Gatekeeper?
The Gatekeeper uses AI as a relatively autonomous tool within the process. The professional does possess sufficient domain and process knowledge to critically assess the output and detect problems, but provides little or no active guidance to the AI during the process. Initial instructions are limited, and the AI is given considerable scope to generate its own proposals, variants or solutions. Active intervention mainly occurs retrospectively: when the output does not sufficiently align with the professional’s intention, proves to be inconsistent or causes quality issues.
This position therefore differs fundamentally from the Director. Whereas the Director sets the framework for the AI in advance, makes iterative adjustments and consciously provides direction, the Gatekeeper primarily monitors the end result. The core of this typology therefore lies not in active process control, but in reactive quality control based on sufficient professional understanding.
The fundamental difference from the Dependent lies not in the degree of active steering, but in the level of understanding present. The Gatekeeper does indeed possess sufficient domain and process knowledge to recognise problems, analyse errors and intervene in a targeted manner where necessary. The underlying understanding therefore remains present, even when it is not constantly explicitly visible in day-to-day actions.
At the same time, this position also entails clear limitations. Because AI is relatively poorly defined at the start of the process, part of the steering role shifts towards post-hoc correction. This creates the risk that the output will only later prove to be insufficiently aligned with the professional objective. Professionals then often waste time on additional prompting, iterative corrections or manual reworking. What initially appears efficient can thus evolve into a reactive and time-consuming way of working.
How does this manifest in both sessions?
Within the IT session, the Gatekeeper manifests primarily in the tension between functional AI output and the limited use of steering during the development process. Participants describe how developers regularly use code that functions operationally, but whose underlying logic is not always fully and explicitly explored during the development process itself. Problems often only become apparent when systems need to be integrated, adapted or debugged.
As a result, the developer’s role shifts partly from active construction to reactive analysis after the event. Whilst the professional does possess sufficient understanding to ultimately analyse problems, the steering of AI is not made sufficiently explicit in the early stages of the process. Debugging therefore becomes a kind of stress test that reveals whether the underlying understanding of the system is still truly present.
Lecturer Niels Jacobs’ practical experience illustrates this clearly. He describes how a production problem just before a release meant that automation suddenly proved inadequate, and everything had to fall back on manual checks and system knowledge. In stable situations, AI appears to take over a great deal of work, but as soon as systems deviate or unexpected problems arise, in-depth understanding becomes crucial once again.
Within the CMM session, the same logic emerges in a more creative context. Participants describe how AI is used for brainstorming, variant development and rapid content generation, whereby systems are allowed to generate output relatively freely before being actively selected or fine-tuned.
At the same time, participants explicitly highlight the risk of superficiality and ‘one-size-fits-all’ solutions when AI output is not sufficiently guided by a clear vision of the content. Here, too, a tension arises whereby professionals do possess subject-matter expertise, but initially allow AI to operate with relative freedom, meaning that quality issues or deviations in content only become apparent at a later stage.
What makes this situation particularly complex?
The Gatekeeper places high demands on latent understanding: knowledge and skills that are not constantly being actively utilised, but which must remain permanently available.
This presents a challenge both cognitively and pedagogically. Anyone who relies on AI support over the long term without continuing to actively analyse, structure or provide explicit guidance runs the risk that crucial expertise will gradually be used less actively. However, it is precisely at moments of deviation or system failure that the continued necessity of that underlying understanding becomes apparent.
Furthermore, a low steering score often creates a reactive working dynamic. Problems are not always prevented in advance through clear framing or targeted guidance, but only emerge later in the process when the output does not sufficiently align with the intended outcome. Professionals are then forced to make intensive corrections, re-prompt or manually rework the output. This makes this position less efficient than it appears at first glance.
The Gatekeeper therefore requires not only a deep understanding, but also a highly developed ability to recognise when a more active steering becomes necessary.
Relationship to the other typologies
The Gatekeeper and the Director share the same in-depth understanding, but differ fundamentally in the degree of active steering.
Whereas the Director explicitly frames the AI in advance through objectives, context, instructions and quality criteria, the Gatekeeper initially allows the AI to function with relative freedom and intervenes primarily reactively when deviations become apparent. As a result, the Gatekeeper more often involves an iterative corrective process after the event, whilst the Director attempts to steer the process more at the outset.
Furthermore, the distinction between the Gatekeeper and the Dependent is not always immediately apparent. Both provide relatively limited active guidance during day-to-day operations. The fundamental difference, however, lies in the level of understanding. When systems go off track or output proves insufficient, the Gatekeeper does have the capacity to analyse problems in depth and intervene in a targeted manner. The Dependent largely lacks this capacity.
Implications for the curriculum
The Gatekeeper role has significant implications for curriculum development. Both sessions make it clear that in-depth understanding alone is insufficient if students do not learn sufficiently how to actively and explicitly steer AI systems.
Programmes must therefore not only teach students to analyse, assess and understand, but also teach them how prior framing and goal-oriented AI-steering influence the quality of processes.
In addition, learning environments must explicitly confront students with situations in which AI output does not adequately align with professional intent. It is precisely these moments that reveal the amount of time and complexity involved when active steering is carried out too late and correction only takes place reactively.
Within IT, this may involve debugging, integration issues or AI-generated code that is difficult to maintain. Within CMM, it is more a matter of generic output, superficial content or creative output that is insufficiently aligned with the context, target audience or brand identity.
It is precisely these situations that highlight how in-depth understanding remains essential, yet is insufficient without sufficient active steering in AI processes.