Sometimes the simplest reset works best. When things start to feel stuck, stepping away can help bring everything back into perspective.
That thought came back to me recently after a walk from Stephen’s Green Shopping Centre, and it connected with something I have been noticing more and more while building AI Dubliners.
For nearly a year, I have been creating content and visuals for AI Dubliners. To represent the companies and products I feature as accurately as possible, I often compare outputs across multiple AI models. More recently, I have also been building the AI Dubliners website myself, partly to learn and partly to experience more directly what it means to work alongside AI every day.
In that process, I have been collaborating constantly with what I now think of as my three AI collaborators: ChatGPT, Claude, and Google Gemini.
And over time, I noticed something important.
AI does not exactly forget in the way people do. But it can absolutely lose focus.
You can be moving smoothly through research, writing, design, or website work, and then suddenly the direction starts to shift. Even when something is clearly important, and even when you have already said so, you often need to repeat the goal again and again. Otherwise, the output can drift away from what you were actually trying to do.
That is why the idea of goal drift feels so useful.
The issue is not only whether the model remembers previous context. The deeper issue is whether it keeps the right objective in attention as the task grows, expands, and becomes more complex.
That changes how I think about working with AI.
The answer is not to assume the model will hold the thread on its own. The answer is to keep re-anchoring the work. Rewrite the goal. Bring it back into focus. Restate what matters. Make the direction explicit again.
Because the real problem is often not memory. It is attention.
If you are building a project with AI, that has practical implications. Take notes. Re-state the objective. Keep the structure visible. And most importantly, remember that control should stay with you.
Otherwise, projects can drift quietly. What looks like a quick task can become messy, frustrating, and unexpectedly draining.
This is one of the most important lessons I have learned while building AI Dubliners: working well with AI is not only about prompting. It is about maintaining direction.
So I am curious:
What do you do when you notice this kind of drift while working with AI?
