2026-06-22 · Is AI ruining our skills? Early results are in and they’re not good

Transcript

Transcript. Paraphrased; sources in notes.md.

HostWelcome everyone. It is June 22nd, 2026. I am Daniel, and sitting across from me, nursing what I suspect is her second beer of the afternoon, is Karen. Today we are talking about the slow erosion of human skill in the age of AI, the death of the collaborative pull request, and some new guidelines on how to handle all of this without burning the house down. Karen, you look like you have been waiting for this exact conversation.

GuestWaiting? Daniel, I have been living this for two years. The industry is currently trapped in a giant feedback loop of mediocrity, and everyone is acting like it is an efficiency miracle. I just spent three hours fixing a bug in a codebase that someone AI-generated, and it was the most infuriating game of whack-a-mole I have ever played. But sure, let us talk about the future of work.

HostLet us start there then. There is a report in Nature about medical professionals and their diagnostic skills. It claims that once people start relying on these AI diagnostic tools, their own ability to spot issues begins to decline. Over on Lobsters, the discussion is quite sharp. One commenter named lcamtuf basically argues that while we are used to skills becoming obsolete, like knowing assembly language, this is different. It hits our ability to actually evaluate the machine, which is a terrifying place to be when the machine is wrong.

GuestIt is not just terrifying, it is lazy. We are trading our intuition for a black box. Look, if you are a doctor and you stop looking for the growth yourself because the computer says it is clear, you are no longer a doctor, you are just a clerk for an algorithm. And the same thing is happening to junior devs. They do not know how to debug anymore because they never had to learn why the code worked in the first place. They just prompt it until the red text goes away.

Hostlcamtuf’s point about the sameness multiplier is interesting too. He thinks the content consumers are happy with this, but it is destroying the creative process for everyone else.

GuestExactly. You end up with a recursive loop of AI training on AI content, and eventually, the whole thing loses its soul. If your output is just an average of everything that came before it, you are never going to have a breakthrough. You are just going to have, at best, a really smooth, boring version of the past.

HostSpeaking of output, we have a story from Miguel Grinberg who has decided he is done with unsolicited pull requests on his projects. He calls it the reverse centaur problem. He is tired of reviewing code that was spit out by a machine, and he wants his project history to reflect human intent.

GuestI read Miguel’s piece this morning and I almost stood up and applauded. He is hitting the nail on the head. Open source used to be about sharing a journey, about seeing how someone else solved a problem. You read the code to learn the logic. Now, you are just auditing a machine. I find myself doing this with my own internal tooling too, where I just reject anything that looks like a boilerplate auto-complete dump. If you cannot explain the logic, I am not merging it. It is not about being a gatekeeper, it is about maintaining a coherent codebase that is actually readable by a human being.

HostA Lobsters user called lake echoed that sentiment, noting that the act of sharing code used to be about sharing a process, and that this feels deeply anti-intellectual now.

GuestBecause it is. When I am working on a complex state machine for a client, I am not just throwing darts at a wall. I am crafting a narrative with the code. If I start accepting machine-generated pull requests that rewrite my architecture because some LLM thought it looked cleaner, I lose control of the system. I become the person who is just pushing buttons, and that is not the job I signed up for.

HostFinally, the Software Freedom Conservancy has released recommendations for using AI in FOSS. It is a practical guide for how to navigate these proprietary tools while still trying to protect the integrity of open source.

GuestIt is a nice gesture. The Conservancy is trying to give people a map for a landscape that is already halfway underwater. The reality is that people are going to use these tools because their bosses tell them to, or because they are lazy, or because they are over-leveraged. But I appreciate that they are at least acknowledging the conflict. It is not going to solve the problem, but it might give a few people the vocabulary they need to say, hey, this license is a minefield and maybe we should think about what we are actually checking into the repo.

HostIt is interesting to see the tension between the legal reality and the practical reality. One commenter on Lobsters, ema-pe, just dropped the link for anyone who wants to dive into those guidelines.

GuestI will probably read them, but honestly? It feels like we are trying to put a screen door on a submarine. We have collectively decided that speed is more important than craft, and no set of guidelines is going to reverse that until the industry hits a wall so hard that it finally wakes up.

HostWell, on that cheerful note, I think we have hit our time for the day. Karen, thank you for being honest, even when the truth is a bit bleak.

GuestAnytime, Daniel. It beats the alternative. I am going to go finish this beer and try to forget that I have to merge a PR tomorrow that is probably going to be sixty percent hallucinated nonsense.

HostYou have been listening to us discuss stories found on Lobsters. I am Daniel, and I will be back tomorrow with more. Have a great evening.