As my friends, family, coworkers, and probably several people on the internet already know: I am an AI skeptic – and I’m not particularly shy about...
For further actions, you may consider blocking this person and/or reporting abuse
This sounds pretty balanced! As an early AI adopter I find myself agreeing with you more often than not. Especially the writing part. Once you see it enough times, picking out an AI-written piece is trivial. I find AI more useful for writing when I ask it to suggest outlines or do research, then do the writing myself.
Yeppp – even if you feed it samples of your own writing, etc. it still has a distinct (and easily clock-able) tone. I know everyone has their own personal list of "AI tells" but the one that always jumps off the page at me is "It's not just X – it's Y" 🙃
That one's my biggest pet peeve 😂 I read one article where it was literally in every paragraph. Like come on. Saving on context window much??
This is hands down one of the best articles I’ve read about AI, probably because I relate to it in so many ways, I don’t even know where to begin.
I wouldn’t call myself an AI skeptic, but I am still skeptical about the whole vibe coding hype. (I’ve talked and written so much on the topic, I’m pretty sure I’m not every vibe coder’s favorite person 😄)
I completely agree with what you said about writing. Someone once called me out on one of my DEV articles, claiming it was probably written by AI. But the truth is, we (written content creators) enjoy writing, and we don't have to write anything. Writing with AI just isn’t fun, so what’s the point of writing anything if it's neither required nor fun? Besides, you won’t build a real audience without your own unique style. AI can’t truly mimic that, or at least not well enough.
Now, when it comes to images... I'm guilty of overusing AI-generated visuals for my articles. They’re not amazing, but they’re decent and definitely better than nothing.
What really caught my eye was the “Body Doubling” concept. I haven’t tried it yet, but it sounds intriguing. I’m not convinced it’ll boost my productivity, but hey, never say never. You don’t know until you try, right?
Now, back to vibe coding, which is probably the most sensitive and interesting part for me. Writing code with LLMs doesn’t really excite me (probably because I’m a pre-AI era developer). That said, I do think there’s value in it and potential for meaningful use cases. I talk about this more in the podcast episode and one of my articles. Still, I believe the best approach right now is to find that sweet spot, somewhere between going all-in on AI and ignoring it entirely. Personally, I try to keep myself grounded and avoid falling too deep down the rabbit hole.
Glad it hit home with you! It's been really cool for me to see folks reply to this article and talk about some of the ways they're also in the one-foot-in-one-foot-out zone on AI. It feels like a lot of what's getting posted and talked about online is from folks who are ALL IN and have a very "get onboard or get left behind" approach, haha. Nice to find some other in-betweeners :)
As for the body doubling thing, it sounds SO silly but there's just something about having an "accountability buddy" (even if the buddy isn't real) that seems to work for me. I don't do it all the time, but it's nice for stuff that I need a little push to get started on or have been dreading / putting off. I think of it kind of like those livestreams where a group of folks will get together and work on separate tasks – sometimes, there's just something to saying you're going to do something and knowing you'll have to tell someone if you didn't do it that's enough to get the gears turning when you're stuck. And for whatever reason, that still works for me even if the other person isn't a person at all, lol
That "get onboard or get left behind" drives me crazy. It's not that simple, people should understand it.
About body doubling, I don't think it's silly to be honest, I actually want to try it myself. Beats talking to a rubber duck at least, right? 😄
This is great Kathryn. Going to repost this to another dev network (with full credit of course). Such an interesting read. It's refreshing to hear an honest and objective opinion of the pros and cons of AI. Thanks for sharing !!
This is such a balanced take on using AI effectively. I especially agree with treating it as a focused tool rather than a complete solution—your examples around debugging and docs lookups are spot-on. The scheduling use case is a great idea I hadn’t considered before. Really appreciate these practical, no-nonsense insights!
Very solid take. I will say theres a pretty huge gap in using paid models vs the free plan's though. ChatGPT saved me a lot of time with creating prototypes but I don't think I would use even a paid one beyond that.
It is great for that accountability without bugging another person and helping summarise things when your working on more elaborate projects though.
I love your approach! I use AI for both writing code and text, but the key point is that I always manage the process. I communicate with it, ask questions, and guide it in the right direction.
As a result, I get working code and quality text (well, I hope it's quality since English isn't my native language!), but under my control.
AI is a tool,
humans are the architects of the solution.
Agree with these! As someone who's been on and off about learning code, having AI write an app for me is a pretty big hack XD however I like to use it more as a learning experience, ideation, and structure of code.
Great post👍, I do use gemini deep research to get some ideas and links to then do research on my own
I've enjoyed using it for initial research, and I totally understand your sentiment around AIs never saying “I can’t find that". I'm addind a custom instruction in claude to see if it searches more official docs and other sources. If it can't find it it must be honest and say that to the user.
I'm also using more agentic coding to experiment and see the limitations.
Also, glad to see you called out how bad it is for image generation 😄. I simply do not understand anyone that uses AI generated images with grammar issues.....it fails at text, just don't use it in slides or blog post please😅
I am a handyman that have lots of tools to deal with metal, wood, plastic, etc… Also, I do gardening, beekeeping and composting with worms. But happen that also I do some amateur coding in my GNU/Linux. If you believe... I do coding since 1984. Yes! I am almost 60.
Lately, I started to solve challenges from frontendmentor.io using Next.js and React. For that, AI is helpful, for sure. It is a tool, noting more. The wood tools are not very good to handle the bees and to feed to worms there are special instruments to handle the kitchen scrapes.
I use Windsurf inside the IDE, Clause, GPT and Deepseek as I have only free tears and need to distribute the tasks.
With a backhoe we can do a lot more work that with a shovel. But is the shovel obsolete ?
By the way, right now there is an AI plugin in my browser correcting my poor English.
The takeaway: Give to AI what it can help us. All the rest (mostly) will remain ours...
Brains are weird, indeed. But thanks for sharing your own experience and these interesting, useful examples! Some of these would never have occurred to me. Much appreciated.
Re: "I didn’t remember how I had structured or named things (because I hadn’t actually structured or named them)", I find that GitHub Copilot, integrated into my JetBrains IDEA IntelliJ IDE, does a great job of identifying, "learning", and utilizing my own naming conventions in the code it suggests.
Just in case you're interested. I love that you're figuring out how to fit AI into your processes and comfort zone. I believe strongly that augmentation and partnership with our AI tools -- using them where we could use a little help -- is a far more healthy and realistic future than being replaced by them.
Keep your standards up, and keep learning. All the best!
AI as a scheduler sounds decent, but I still think that having a good note taking app is just better, like Obsidian. So it's probably bias here but I think its better, as you're not prompting the AI what tasks are scheduled/completed (and not using up tokens), you just straight up mark it as complete and your notes are always available.
Plus you're able to pick up other habits like note-taking and journaling.
The image generation stuff is just poor cognitive offloading mostly for people that are impatient with their current creative talent
Oh, I'm still a very dedicated user of Notion and my paper Moleskine weekly planner, haha. I like to have the AI divide up / time block the tasks for me, but then its job is pretty much done. I'm a big fan of journaling as well, although I like to keep that as a personal thing and not a professional / workspace tool.
And 100% agreed on the image generation 🙃
Loved this, So as a developer using gen ai at first i was ashamed of it but lets face the reality this is the gen ai era no point of hiding it but the fact is to start using less replace ai work with ur own the more u learn right?
Yepppp – it's an interesting tool and good in some limited capacities, but at the end of the day you just have to do stuff yourself (often over and over) in order to actually get good at it. No good way to automate the benefits of practice, I fear.
This is great - well-written, and (of course) not by AI ! ;-)
Can't say I don't feel the same way!
Great thoughts Kathryn! As someone who is "all in on AI" its nice to hear from someone in the other camp.
Great post
For an AI skeptic, you’ve got quite a list! I mostly use ChatGPT to brainstorm structure ideas + a bunch of outreach campaign tools with built-in AI (seriously, which ones don’t have AI these days?)
Quite an honest and balanced take on AI ...
Great post, thanks a lot!
ChatGPT for the chat based stuff (as a free user), Copilot for code (paid for by my employer). In general, I don't mess with changing the models so I use whatever they default to.
That is fantastic
Really enjoyed reading your post🙌.
Welcome aboard! It sounds like you’ve already figured out where AI really shines. Here’s another use case most folks overlook: the o3-o4 models are insanely good at pattern recognition — especially if you give them targeted logs or DB dumps (and bonus points if you add a repo or two into the mix). Granted, it still took me a good three days to hunt down memory leaks and rogue zombie threads clogging up the JVM, but AI was a lifesaver combing through hundreds of layer traces across the DAO when the batch jobs were running at 3x capacity. 😇
I hopped on the AI train early this year with GitHub Copilot — so if you’ve read anything I’ve written lately, yeah... there's a theme. My initial goal was to push Copilot as close to full-on autonomous as possible, while I kicked back and played director. Did it go smoothly? Not a chance! Early days were pure, unfiltered chaos 😆 (think: edit mode, file A, “write this function for me”—then a flood of ALLCAPS commentary and review cycles until I gave up and switched on the voice feature, which made things even weirder).
But after a month? I had a scrappy POC v0.5, a functional-enough test suite, more docs than I have sense, and a much better sense of how to wrangle AI tools without letting them yeet your entire codebase into the sun (unintentionally tested that too). Now, I’m actively teaching, mentoring, and still testing the boundaries, because that’s where the real wins are.
Here’s what irks me: CS colleges are outright banning AI? I get it for first-years, but by junior year, devs should be leveraging AI the right way to boost their skills, not just “hit a button, get a test case.” It’s amazing that so many people expect AI to just work straight out of the box. It doesn’t! The human is always the linchpin. Your code, your responsibility — no matter what helped you write it.
I use ChatGPT for all my writing (guilty), but it’s never a straight copy-paste job. I’ve spent months tuning custom instructions so it doesn't lose "me", and I still type or dictate every post before handing it over for edits. And yeah, I use AI art for my blog too — Leonardo was a recent treat-yourself splurge after hitting a personal milestone.
Bottom line: AI is a tool, and like any tool, it’s only as good as the human driving it. Just because you launch a container in K8s doesn’t mean it’ll scale. Same with AI: tossing it at a problem doesn’t guarantee the answer you want. It takes skill, patience, and a little bit of chaos to make the magic happen. 🔮✨