A model that handles heavy lifting won't necessarily make you better at it

Automation and mentoring don't technically overlap

Ethan Mollick points out a clear divergence between automation and human augmentation. It is commonly assumed that if a model is an excellent programmer or analyst, it will also be an excellent mentor. But new data shows this isn't true. A model that can rapidly generate complex code from scratch often fails to effectively explain to a beginner where they are making a mistake.

The push for autonomous solutions sacrifices teaching

The push to build the best possible agents forces companies to focus exclusively on automation. If models are trained to work with a single click and deliver a result, they aren't trained in conversation and feedback. The result is brilliant working machines, but terrible teachers.

Mentoring demands more than just domain knowledge

If tool creators truly want to improve how people work, they can't rely solely on models taking over tasks. Automation and augmentation are two completely different disciplines from a development perspective; the latter requires not only domain knowledge but also empathy for the learning process, the ability to structure information, and the capacity to provide the right feedback at the right time.

Ceding control limits expert growth

Companies that fail to realize this might replace a lot of work with software, but they will lose the opportunity to use AI to cultivate their own experts. Ultimately, the true measure of success for enterprise AI won't just be how many lines of code it wrote itself, but how much better it made the developers it worked with.

Lilith's verdict

We built the fastest calculator in the world and now we are surprised it can't teach elementary school math.