Hey Jules: Don’t make it bad!

Take a sad trend, and make it better!

Even if you’re not that interested in technology trends, it was hard to miss some of the innovations announced at Google.io. Literally as I write this, I was sent a hilariously realistic Veo 3 video entitled “Yeti vlog”. I think we can all expect a lot more of that in the weeks and months to come. But I digress… the subject of this article is to be another of Google’s conference announcements: Jules , the AI coding agent (note the tagline in the lower-right corner):

That tagline very much reflected the overall “vibe” (no pun intended) of the product announcements: We’re here to help unleash your creativity, and take care of the pesky details. It’s a really appealing notion. Very empowering for those who do wish to create things, but feel held back by the “barriers to entry” of learning the latest programming language, framework, video editing tool, 3-D modeling software or just computer skills in general. So many of these tools are now multi-modal… you can give them images or audio or even video as input and they can deal with any media as easily as you or I could. We’re not limited to just textual interaction anymore.

I lived with a vision-impaired person for a few months, and helped guide him around the city and read signs and things… I haven’t seen him for many years, but these innovations must be amazing for his quality of life. My personal quality of life has changed a lot, and mostly improved. My “pairing bot” has given me new confidence and allowed me to have more success with a lot of things I struggled with when my only resource was internet searching. There is so much good that can come of this.

And yet, we seem determined to “make it bad”. While the spin is all “creative assistance” to empower us all… why is there an equally pronounced undercurrent of “replacing people”? Perhaps our love of dystopian futurism and killer robots has led us to believe this is an inevitable outcome. The general feeling seems to be that the second our boss, or the government, or heck… the AGI decides we’re disposable, we will suffer our ignominious fate. It makes a good story, I suppose. And lots of good short-form videos, for sure.

So what is Jules, and why can it make things bad (more than other tools)?

Engagement. (…or lack thereof)

A few times in other articles, I’ve slighted another tool (Devin) for encouraging the idea that the AI can substitute for a functional developer who is part of a cohesive team. I was a little worried about Devin, but I am a lot more worried about Jules (mostly because of the Google monopoly).

I’ve referred to these kind of products as a form of “outsourcing” the work of developing code, which hasn’t (in my experience) been a very successful endeavor of modern globalist capitalism. This has nothing to do with other countries, other cultures, other languages or geographic separation (although all of those don’t help). You can outsource your lawn mowing to the kid down the street… remember when that used to be a thing?

Let’s explore that metaphor a bit… how does that go? There’s the schedule, the price negotiation, the quick walk around the yard with instructions. Did he mow this week? It looks a little long. Why do I suddenly have an Old Faithful eruption when the sprinklers turn on? Man, he missed a few spots… I’ll have to remember to tell him. What a pain, so much disappointment. We’ve made it bad.

I contrast this Devin/Jules-like experience with working in the yard with your family… you’re a team, you’re all just out there trying to make things look nice, sharing the load… offering help or suggestions, learning and teaching. If we encounter difficulties, or miss a spot… we own it. We take responsibility, we fix it. No biggie, and we’re in good company, we took a sad lawn and, well… made it better.

I recently injured my foot, and so was unable to mow the lawn. Of course, the solution was to get a robot lawn mower 🤖. You know what, I’m pairing with bots there too… we’re a team. It’s not perfect, but we can work it out. And thus we see the difference… keeping the expectations in check. Working with other people or AI to co-create things and enjoy the journey.

Pairing with bots does not absolve you of the responsibility to engage in the work of software development, and tools that want to “do it all for you” should be considered highly (as the youths like to say…) “sus”.

There are so many new tools, guys… I can’t possibly tell you about all the things I’ve been one-shotting and trying lately. The Zed editor is really promising, but not there yet. Claude 4 is amazing, and amazingly expensive as your “daily driver”. Gemini 2.5 Pro is so good, so fast and reasonably priced. I’m afraid it looks to cement this monopoly. There are some quite exciting open-source models you can run on LM Studio with reasonable hardware (Apple silicon macs or PC’s with really nice gaming/graphics cards). Devstral (for Python), GLM-4 (for web dev) and Phi-4 (for general dev) come to mind, though things are still slow and somewhat problematic for local LLM development. Text diffusion models are completely whack, but so dang fast! It’s easy to get distracted by all this… progress? shiny-object worship? nonsense? What is a developer to do?

Keep pairing with bots

So choose tools that keep you engaged with the driver (maybe switch drivers once in a while), keep navigating safely and professionally to your destination. Take your sad code and make it better.

I realize it’s been a while since I last dropped an article, so next week I’ll try to give some more helpful technical details on what I’ve been working with lately.

Hey Jules: Don’t make it bad! was originally published in Pairing with Bots on Medium, where people are continuing the conversation by highlighting and responding to this story.