← All episodes

2026-07-12 · Working With AI: A Concrete Example cover art

2026-07-12 · Working With AI: A Concrete Example

Show notes

BRINE — 2026-07-12 · show notes

Guest: the tooling optimist (a fictional archetype).

Claims are paraphrased and attributed; nothing is read verbatim. Where a thread disagreed with the article, the show surfaces the disagreement.

Segments

  1. Working With AI: A Concrete Example
  • Source: https://htmx.org/essays/working-with-ai/
  • Discussion: https://lobste.rs/s/my94an
  • Topic: AI-Assisted Debugging · interest 85
  • Carson Gross details his experience using LLMs to diagnose a regression in the hyperscript parser. He demonstrates the 'Sorcerer's Apprentice' phenomenon: while the AI effectively identified the cause of the parsing conflict, its proposed fixes were either hacky or unnecessarily complex, requiring the author to apply domain expertise to find a cleaner solution.
  1. The Joy and Power of Understanding
  • Source: https://binaryigor.com/the-joy-and-power-of-understanding.html
  • Discussion: https://lobste.rs/s/6vsofh
  • Topic: software craftsmanship · interest 75
  • The article argues that deeper understanding of software systems is a source of both professional power and personal joy, warning that over-reliance on LLMs creates a 'deskilling' loop. It provides a nuanced framework for when to prioritize understanding versus quick, AI-generated solutions. The discussion highlights the tension between using AI for rapid prototyping versus the risk of losing the intuition necessary for long-term maintenance and debugging.
  1. Are We Harold Bloom?
  • Source: https://abner.page/post/are-we-harold-bloom/
  • Discussion: https://lobste.rs/s/tgzdhf
  • Topic: culture · interest 75
  • Abner Coimbre reflects on the 'Harold Bloom' archetype within developer communities, arguing that elitist dismissal of AI is less effective than leveraging data about its poor ROI and economic waste. The discussion features substantive reflections on the incentives for intellectual radicalization and the potential for a shift toward 'de-hype' as compute costs impact the bottom line.

Transcript

Transcript. Paraphrased; sources in notes.md.

HostIt is July 12th, 2026, and welcome back to the show. I am Daniel, and sitting across from me is Samantha. We have got a packed agenda today, Samantha. We are looking at the realities of AI-assisted debugging, the enduring value of software craftsmanship, and a look into the culture of our own tech circles, all drawn from the community over at Lobsters.

GuestIt is good to be here, Daniel. I have been thinking about that AI debugging topic all morning. I actually spent half of yesterday fighting an LLM that kept hallucinating a perfectly valid looking, but totally non-existent, library function. It is definitely a tool, but one that demands a very firm hand.

HostThat perfectly sets the stage for our first story. Carson Gross, the creator of htmx, wrote about his experience using AI to track down a regression in the hyperscript parser. Hyperscript, for those who might not know, is a small scripting language designed for the web that lets you attach behavior directly to HTML elements. Carson found that while the AI was decent at spotting where things were going wrong, the actual solutions it proposed were messy or overly complex. He calls this the Sorcerer's Apprentice problem, where the tool handles the task but does not really understand the implications of the fix.

GuestOh, I love how Carson frames this. It is the classic trap. You ask the bot for a fix, it gives you a sledgehammer, and if you just hit enter, you have effectively turned your codebase into a pile of technical debt. Over on Lobsters, the community really latched onto the humor in the piece. A user called eduard even compared Carson’s writing style to Douglas Adams. But regarding the technical side, a commenter named viraptor made a great point that even if the AI suggests garbage, the process of filtering those ideas can be a way to find real bugs. You just have to be the one holding the leash.

HostIt is interesting that even when the AI fails, it acts as a sounding board. But does that reliance on the board make us worse developers in the long run?

GuestThat leads us right into our next topic. BinaryIgor wrote a piece on the joy and power of really understanding your systems. The core argument here is that knowing your code deeply is not just a pragmatic advantage, it is genuinely satisfying. The author warns that our natural human laziness makes us reach for AI-generated shortcuts, which effectively deskills us.

HostIt is a strong warning about the trade-off between speed and agency. If we stop understanding the *how*, we lose the ability to innovate on the *why*.

GuestExactly. And the commenters were not holding back. A user named ugur mentioned how tired they are of the trend where people build tools without knowing what is happening under the hood, as if that lack of knowledge is a badge of honor. Then there is hgrsd, who pointed out that there is a real economic incentive for AI labs to push this deskilling, just to get us hooked on their infrastructure. I think the key is finding that middle ground. I use AI to write my boilerplate, sure, but I am in the terminal reading the diffs and stepping through the execution every single time. If you are not looking at the code, you are not building, you are just clicking buttons.

HostYou are certainly not alone in that sentiment. That kind of frustration with the current state of things brings us to our final story today. Abner Coimbre wrote a thoughtful piece about the Harold Bloom archetype in our tech circles. Harold Bloom was a literary critic who was famous for being a bit of a curmudgeon, and Abner suggests that we have a lot of developers acting like Bloom today, dismissing AI-generated work as mere slop.

GuestThis was such a fun read. It is so easy to fall into that elitist trap, right? It feels good to be the person who says, no, that is not real code, or that is not real art.

HostAbner argues that while it is easy to be a critic, it is more effective to focus on the data, the return on investment, and the actual economic waste rather than just elitist posturing.

GuestThat is the crux of it. A Lobsters user called lcamtuf had a really sharp take on this. They pointed out that taking a crisp, extreme position gets you way more attention than saying something nuanced like, well, there is some merit and some hype here. You end up with a personal brand based on being the loudest critic in the room. If we want to actually change how we use these tools, we need to stop the performative outrage and start talking about the actual utility.

HostIt is a fair critique of our industry's tendency toward radicalization. I think I am going to try and catch a bit more nuance in my own intake this week.

GuestGood luck with that! Though, given how many people on the internet seem to thrive on being the grumpy professor, I think you might be fighting an uphill battle. I am actually looking forward to a weekend of completely analog gardening, where the only thing I have to debug is why my tomatoes are turning yellow. No LLMs required for that.

HostThat sounds like a perfect reset. Thanks for stopping by today, Samantha. And thanks to everyone for listening. All the stories we covered can be found over at Lobsters. We will see you all again tomorrow.