Self-made software

Earlier this month I was listening to an episode of the BitFlip podcast, which talks about self-hosting. In that episode one of the two hosts was explaining how he had been looking for a good stand-alone self-hostable calendar server. Nothing he had found was as good as Nextcloud Calendar plugin (I agree there) but he didn’t want to run a full Nextcloud instance.

Since, as he told his audience, he lacked any coding skills, what he did was ask an AI to isolate the calendar code from Nextcloud in such a way that it could be run without the rest of the platform. It worked, and it didn’t cost him a lot of time or effort. He was pleased with the result. He admitted to being apprehensive about its security, so he deployed it behind a firewall and used Tailscale to reach it when outside his network.

The hosts reflected on the meaning of this story, comparing it to 3D-printing. When previously you’d need a craftsperson to build physical objects for you, 3D-printers allow you to do it yourself now. The result will be tailor-made for you. The same applies to software. AIs allow people without any (coding) skills to create tailor-made software for a single person to use.

I have another analogy. It’s like he told us he got someone to take the front door from his house and put it in his shed because everybody does that now. It makes no sense, has no value for anyone but himself and is only interesting because of the hype.

Unfinished

The other week, at work, I was given just fifteen minutes to create a report of a certain security aspect of all companies in our supply chain. Since our supply chain is enormous, the use of an AI was implied. In the following meeting, I could not present the results because I hadn’t written the report. I hadn’t had time to even read it. A machine had done the research and created it.

It was okay, I wasn’t meant to, but it still didn’t feel good. It felt, like, unfinished, perhaps? What did I even create? A prompt that resulted in the report, but not the report itself. That was just generated. What was, in fact, missing, was the feeling of accomplishment, the tiny bit of pride on something created.

Cheating

It’s like playing a (computer) game. I can play Portal, Lemmings, or some other puzzle game and look up online how to solve every puzzle. Obviously, I will finish the game without feeling any accomplishment. Also, I love gaming, so what would be the point?

But what if I solve most of the game myself and only look up a few levels that I can’t seem to solve. That is in fact how I finished Portal 2. If I hadn’t looked them up, I wouldn’t have finished the game. That wouldn’t be a problem, but I would also have missed the later levels that I did solve myself. As can be expected, the feeling of accomplishment was still there, but less, and I won’t forget having cheated a few times.

That is also how it would feel for me to have an AI write a scene or a dialogue in a story. I love writing, so there wouldn’t be a point in having an AI write them for me, even if the result would be objectively better. The effort itself is the point, because of the experience and what I learn from that, even it's just something about the characters involved.

Back to the report

Reflecting, I find that that is probably what I felt after generating an entire report at work. Having started on this job only a few months ago, going through the bulk of our supply chain, like the AI did, would have been a good way to gain knowledge that would serve me well in this job. Having realised that, whenever I have time, I re-do some of the research myself. This takes time and effort, but I find it worth it.

Which, in turn, brings me back to the podcast episode from the introduction. The podcast host can ask the same or even another AI to patch his calendar server when Nexcloud contributors have published an update. The same mechanism will enable him to add or remove functionality according to his needs. But without going through the, admittedly, considerable effort to learn to code and understand the codebase, he will forever depend on AIs to keep his calendar server running.

Depending on something that is not sustainable is eventually not a good idea.

After reading this, perhaps you think you’ve never read so many open doors kicked in at once. I certainly feel that way after writing this. Of course the feeling of accomplishment won’t be there when you didn’t create something. Thanks for explaining that! And without putting in the effort you won’t learn anything. Nothing new there.

So why, if this is all so clear, do people keep doing this? Why is this worth telling in a podcast? I want to live in a normal world again with houses with front doors. Can we stop the hype now?

Please?