GenAI is like a chocolate bar

I said that genAI is like 3D printing. Now I think it is more like a chocolate bar.

Stay with me for a second.

When 3D printing came along, there were a lot of voices it will revolutionize manufacturing. Had its mini-boom, and by today it is normalized along its natural constraints. It did revolutionize a very specific part: prototyping. But scale remains part of other, specialized processes and manufacturing develops further.

See the parallel? I think that is where genAI eventually really end up: prototyping digital products has been revolutionized. However its constraints still unknown at this point.

What is a chocolate bar? A temporary good feeling, maybe even a pinch of energy and a tickle to your brain’s happiness spots. But when you consumed it, it is gone, and costly to replenish for next round. The more you eat, the less one piece feels satisfying. And the hidden costs appear way down the line. I think now it is a perfect parallel for genAI.

It feels good to use in the moment, but the hidden costs kick like a horse way later:

• Subsidized market: current consumer prices are still undercutting real costs and opaque business deals make it hard to measure up real costs and where current balance should lie.
• Loss of reasoning: you and your team unable to own the output as they never did the hard work of coming up with in the first place.
• Mental atrophy: losing necessary thinking skills by outsourcing (brain is like a muscle, oversimplified but catchy comparison).
• Value misattribution: we had a copyright system for a good reason, the original thinkers need to be valued appropriately to keep a market reward so they invest time and energy (I am a writer myself, and cheapening a craft makes it demoralizing at the least).
• Attack surface explosion: non-engineers coding any of their own tools/problems create enormous IT security surface area we don’t have tools or processes yet to handle. At least not stably and not economically.

I frequently use genAI myself for coding and prototyping, it can be great but I already feel the negatives on my own skin every day. I don’t think the world was ever constrained by how much code we can write.

It is indeed constrained by how costly the code is. Just code doesn’t have a linear or simple cost structure. Most of the cost is in the understanding of a problem, designing a dozen or more solution, and try them until one fits. You can automatically test if code works but can’t always test correctness.

And making code cheap now just creates a world where everyone asks only the “can I do it?”, rather we need to ask as always: “should we do it?”