“What if you were told you couldn’t hand-write code anymore?”

I listen to the Nextlander podcast pretty religiously. This week they brought up a recent story that Epic Games has a new launcher and that it was entirely (?) vibe-coded, and asked for people to weigh in on the question:

“What if you were told you couldn’t hand-write code anymore?”

I sent them the following email, which I then immediately realised I should have expanded on a bit, of course.

I’m not saying I agree with Tim Sweeney’s approach here, but know this: Almost everything you are using now will have had Claude Code et al be a major contributor to it, whether that is disclosed or not.

Coding is not like building furniture: I don’t believe there is any such thing as an ‘artisan web app’, and the people who do believe that are about to have a very rude awakening. The users, who in most cases ultimately pay our wages, do not care if that if statement was hand-written, nor should they!

It goes without saying though that I think there should be a quality threshold in the product you’re shipping met before it gets released, something that many execs seem to have forgotten about, but I’d also expect a level of correction here over time.

I kept it short in the hope that it gets read out on the podcast (🤞), but I kept on thinking about it.

I’m not morally against a world where Claude (I’m going to use ‘Claude’ here as a stand-in for any current or future coding model/harness) writes all code. In fact, I think that is a semi-likely thing that could well happen.

Whether it does happen, to my mind depends on only one factor: cost.

If you give an LLM a tight enough spec right now, the odds are it will be able to implement your request. This is a big change from even a year ago, where the rough edges would have to be smoothed out, but if you’re working in small enough increments with good enough descriptions and skills at your disposal, Claude will get it done.

Where I and most engineers seem to be spending our time now is on architecture: defining systems with an understanding of how products may scale or change in the future. This planning is often based on requirements that Claude just has no idea about, because it wasn’t in the meeting where that was discussed.

If the costs are low enough though, does that future-looking still matter? If you point Claude at a system and say: ‘Keep it optimised around these parameters’, with enough time and enough environments to deploy and test things in, could it not do it? Would it not continue to do that as product requests change? Could it not also begin defining its own metrics to hit, turning the whole process into a feedback loop?

I don’t believe we’re quite there yet, but given the rate of change in the past 18 months, I don’t think we’re far off it either.

So what if I was told, no more hand-written code? Right now it might matter still, but I don’t think it matters for long.