The other day, my Django Chat co-host Carlton Gibson and I were having yet another discussion about AI: how people are using it, whether it’s going to replace us all, and the ethical and economic implications of it. And he remarked in passing, “I miss tabs versus spaces.”

I felt that. Deeply.

Tabs vs Spaces Silicon Valley Gif

So many of the developer debates these days are not simply about classes versus functions, object-oriented programming versus functional programming, or other small and not-so-relevant programming and technical considerations. Instead, it’s about AI. What even is it? How is it working for you? Is it okay to use, given all the economic, environmental, and political factors? Will it replace us all? And what about companies now using it as an excuse to lay off workers they overhired during COVID? It can feel like a heat shield for management in general. And then there are the very strong echoes of colonialism in how leading AI companies operate, as documented by Karen Hao in the excellent book, Empire of AI.

But also, there’s really something there about Carlton’s comment. Every developer I know is using AI or trying to. AI–and agents in particular like Claude Code–can feel like magic at times, even to people who should know better (myself included). I know that an LLM is just statistical autocomplete. But sometimes, if you’re using an agent and you add in your entire code base and ask it the right question and other alchemy, you can get insights and ideas that don’t seem like they should be possible from just statistical autocomplete. And the same prompt and output—even though you can get wild hallucinations and things that make no sense—can sometimes produce something useful.

If you push back against the models–argue with them essentially–they will immediately agree and take a 180-degree view of what they just said. This sycophantic behavior is well known, and more recent models appear to be more neutral, but it’s still the case that they are trying to predict the next token to present; they’re not really here for a deep debate about any specific topic.

Then there are newer techniques like agent rules, where you essentially pass in a style sheet and tell the agent exactly how you want your code formatted. See here for some Django Forum discussions on this point. I see a lot of value in having consistent prompts set each time with reclass, yet it takes a while to find out what you like. And as the models change, so too do your prompts need to change to get the most out of these models.

Coming out of DjangoCon US and numerous hallway conversations, there is still a broad spectrum of AI usage. Some people are still dipping their toes in the water, using it for search or small code snippets. Others sit in between, using chat functionality for entire files or multiple files. And then there are the cowboys and girls who are fully on the agentic train: go CLI or go home. Use all the tokens and get all the benefits.

I will admit I find myself unsure which port to anchor in amidst this sea of uncertainty.

A final thought: many companies have economic incentives to hype and exaggerate what these models can do. Many people, when they post online, adopt extreme positions for or against AI, but when you talk to them in person, they’re much more measured and considered. That’s partly because sometimes introverted people are more comfortable acting out online or in digital capacities. However, I think it’s also because many people feel that in order to post something publicly, they have to have something interesting to say. And interesting can often delve into aggressive, condescending, or disrespectful tones.

So back to Carlton’s initial offhand remark: tabs versus spaces carries a whiff of nostalgia for past tech debates and perhaps something slightly more lighthearted. But the tools we have today are in another realm of what they can do and what they might do. The stakes feel greater than ever because they really are. LLMs are not the AI technology that replaces us all, yet, but they are a giant leap in that direction, and the genie is out of the bottle. However, this current AI trend plays out in a few years’ time, LLMs are already a core part of the modern world–almost 1 billion people use ChatGPT alone each week–and we’re all still coming to terms with what it all portends.

Genie Out of the Bottle