Handling AI chatbots is shifting. Away from occasional experimentation with individual questions, towards a fixed role in everyday work. Two current reports from the t3n environment describe this development from different perspectives: once as personal practice, once as a didactic recommendation from academia.

In the first case, an author describes how he built his own AI assistant and works with it: Claude first thing in the morning, then the emails. The aim is not to have individual texts generated, but to have a sparring partner and coach available. The punchline of the report: a personal assistant is no longer reserved for management level.

In the second case, Stanford professor Jeremy Utley argues that AI chatbots like ChatGPT only unfold their potential as idea generators and creativity amplifiers when you treat them like colleagues. For this, he recommends an exercise that takes around five minutes.

The common thread: attitude before tool

Both sources start less with the model than with the way of working. What matters, then, is not which system runs in the background, but how consistently it is integrated. Anyone who treats an AI like a search engine gets search engine results. Anyone who treats it like a colleague who needs context, is allowed to ask follow-up questions and comments on interim results, receives results at a different level.

This attitude has a practical consequence: context must be provided. A new colleague receives background information, objectives, examples of good results and pointers on what is customary in the company on their first day of work. These are exactly the pieces of information that a chat window lacks, starting from zero with every request.

From single prompt to built assistant

The difference between the two reports lies in the degree of formalisation. The Stanford recommendation is a short exercise that makes it easier to get started and can be implemented without preparation. The experience report goes one step further and describes an assistant that is tailored to one's own way of working and is part of a fixed routine.

These stages can also be distinguished in a corporate context:

  • Occasional use: individual prompts, no reuse, strongly fluctuating result quality.
  • Structured use: recurring tasks are stored with fixed instructions and examples.
  • Built assistant: a permanently available setup with role description, context material and clear processes, which is scheduled into the daily routine.

The third stage is more elaborate because it requires preparatory work. But it is also the only one in which productivity gains can be repeated, instead of depending on the daily form of the individual user.

What teams can derive from this

From both reports, a sober approach can be derived that requires no major investments:

  1. Select a concrete, frequently recurring task, such as structuring requirements, preparing meetings or checking texts for comprehensibility.
  2. Write down the context once, cleanly: goal, target group, framework conditions, typical sources of error.
  3. Include examples of successful and unsuccessful results so that the quality expectation becomes comprehensible.
  4. Expressly allow and request follow-up questions, instead of immediately demanding a finished result.
  5. Record the setup, share it and refine it over several weeks.

The last point decides whether a personal gimmick becomes a team resource. An assistant that exists only in one person's head disappears from the company along with that person.

Limits and open points

Both sources are experience reports or recommendations, not effectiveness studies. They do not provide reliable figures on time savings or quality gains. The described effects are therefore to be read as plausible practical experience, not as measured benefit.

In addition, there are points that need to be clarified for corporate use: which data may go into which system, who checks results before further use, and how it is documented where AI was involved. The reports do not answer these questions; they belong in every operational introduction.

The preliminary stage to prompt-driven development

Anyone who builds their own assistant trains exactly the skill that is increasingly in demand in software development: describing tasks so precisely that a counterpart without prior knowledge can implement them correctly. That is classic requirements work, only with faster feedback.

For web and software projects, this means: the quality of the specification becomes the decisive lever, not the speed of code generation. Teams that already maintain structured prompts and reusable context building blocks today will find it easier later to integrate AI into development processes.

Getting started costs little: a recurring task, a clearly described context and the willingness to sharpen the result several times.

Sources