I mostly approach what we’re calling AI with skepticism because of who controls it and the impact it has on power dynamics and the planet.

I also realize it’s transforming how developers work, and it was Jeff’s reading list that got me thinking more about it and recently working with AI coding agents like Sketch, Claude, and Cursor CLI.

I’ve already been tinkering with Ollama for a few months and exclusively limiting my AI chat experience to local models. I couldn’t manage to get any of them working well for coding. Or at least not nearly as well as popular commercial ones.

I haven’t asked an agent to build me an app from scratch yet. I’m not sure that I would, because even reviewing minor refactoring and improvements has taken a fair amount of time.

It is novel and impressive to watch an agent evaluate a code base, make a clear todo list, suggest improvements in a short amount of time, run tests, and fix things all on its own. Directing it just like a person is natural. I find that I am polite and use proper punctuation which may be insane.

For me, the time to review with care is still significant. I’m not sure that it’s a massive time saver compared to thinking and coding my way through a problem with only my old brain.

However.

What I really, really like is getting suggestions I wouldn’t have come up with. It could be an architectural improvement or utilizing a framework feature I wasn’t aware of. One such example was using a scoped query in a Laravel model. Not a huge deal, but it looked weird so I consulted the docs and learned that it’s a thing! (And this isn’t a given; Ollama confidently told me I could use some Craft CMS features that don’t exist!) It took Cursor CLI (ChatGPT 5) less time to implement query scoping than it did for me to learn about it and realize the syntax differed from the docs because I unwittingly took myself to the wrong documentation version. Cursor got it right and I made the error.

I agree with something David Crawshaw pointed out, which is that working with an agent provides a boost to momentum that can be powerful. This happens with pair programming too, in my experience: tackling a problem together has its own sort of energy that can be the difference between letting something languish and actually getting it done. My experiments had me making several improvements to different projects—I felt a sense of accomplishment and probably wouldn’t have ticked off the same items on a normal organic day.

The idea I keep grappling with most is this:

Does an intern cost $20/month? Because that’s what Cursor.ai costs.

My first experience working with coding agents has been like working with really smart interns. They do tedious work for me, offer some great ideas, and still make mistakes I have to be careful to either call out or fix myself. This is easily worth more than $20/month.

But what about the intern starting out in the field gaining experience?

What happens to lower-wage developers that had been earning a decent enough living to support themselves?

What happens when we all adopt these tools and moving faster with agents becomes the new norm? (And do we end up with better software, or more complex messes to fix with more complex tools?)

Working with AI coding agents has so far been easy to wade into, and there’s an overwhelming amount of activity in this rapidly-changing space. Knowing how to use these tools well seems like a skill of its own.

Let’s also acknowledge the hype. I know many of you are very excited, but there are a lot of self-congratulatory and breathless videos and we could all calm down just a bit. My hype allergy really flares up around this stuff.

If you’re a seasoned tech bro building software for business, the immediate cost is trivial and the benefit to you is potentially massive. If you’re a hobbyist or a newcomer to the field or not doing work for a massive company, is this good for you too?

Maybe power consumption and model accessibility are things that improve over time. Maybe these tools can truly help everyone and lead to better work and greater understanding and reduce some of the tedium that nobody needs.

I’ll continue to be wary as I see a benefit and wonder who controls the technology and what my use of the thing means for other people.

A quote in Minsuk Kang’s daily sentences struck me:

I’ve found only one metaphor that encapsulates the nature of what these AI power players are: empires.

Karen Hao, Empire of AI

Coding agents are not that expensive for what they can do, but I can’t get excited about feeding money and energy into increasingly-large machines that literally strain power grids and change power dynamics.

The older I get, the more wariness I have for massive companies and the impact they have on human beings. I like whatever puts people ahead of business or scale. I can’t tell yet whether that puts me at odds with The New Way™, whether I just have to find my own balance with it, or whether I should get the hell out of here and build things with my hands.