1st of Aug. 2026 by Michael

Tech-Blog

Resumen

Estoy hablando de algo que todos hemos experimentado: tienes una pequeña cosa en tu proyecto que no quieres dejar atrás, bloqueándote. Puede ser una idea favorita o una funcionalidad en la que invertiste mucho tiempo. Una de las ventajas reales de la IA podría ser que permite (re)crear ese feature favorito con facilidad on demand, haciendo que sea más sencillo dejarlo atrás y dejar de darle vueltas.

Dead Cheries

I’m sure anyone who has written software is guilty of this: a nifty feature, often not even essential, has taken the sufficient amount of thought or development time for it to become endearing to you. You carry it along the commits past its time, simply because you like the idea behind that feature, or because you don’t want to discard all the effort you spent on implementing it.

I refer to these as dead cheries because although you want to drag them along somewhat obsessively, they are already some kind of zombie you should bury. If you don’t, they will block fresh ideas and can mess with your conception of what you want to achieve.

AI for “tactical” tasks, not strategy

By my experience, LLMs are at their best when used with progressively refined one-shot prompts for coding specific “tactical” features for you. You then have a clear, verifiable goal state and a clear idea of its implementation.

For example, asking the AI to create a complex HTML table with an associated stylesheet. You can get sidetracked by details doing that on your own, and spend an entire afternoon tweaking. Once you do get that perfect table, it can become a Dead Cherie and you will try and fit all your data to it, to get a “return on investment” from it.

Agility as antidote against Dead Cheries

But on the other hand, if you work descriptively and formulate an LLM prompt for this, it allows you to stay more agile, because the exact look of the feature can be changed and verified very quickly just by inspection, if you know your HTML and CSS. There will be much less reason to be protective about the implementation of a particular feature.