I’m seeing so much about LLMs in my feed today.

[splendorr@mastodon.social]

@jensimmons Feels terrible, and all of my colleagues who admit to using them are doing worse and worse work while feeling like they’re on top of the world

[jasonnickel@nrw.social]

@jensimmons I‘ve been avoiding it because I didn’t like how much energy it needs and how dangerous it can be for society.

But I also need to think of myself and how my coworkers are doing their work much faster now. I will not be able to compete anymore without AI. So I gave up and started using it. It does help already but there is still much to learn. The ecosystem is already so advanced that it’s hard to get into it now.

We need to solve the fundamental problems on the way.

[developerjustin@mastodon.social]

@jensimmons I find great use when asking these tools to “do” things and not to “know” things.

Transform a bunch of HTML that would require me to do it by hand, rework a specific conditional to work a different way, etc.

Any task that requires access to prior knowledge is hit or miss at best for me but honest to goodness “grunt work” saves me a TON of time in the aggregate.

[dutchcelt@mastodon.social]

@jensimmons I’m using LLMs for some things more regularly. It helps with incrementally documenting my work. I'm sometimes surprised how well it does. But it still requires a lot of checking and redaction. So, not a win for productivity.
Also, they are still awful at working on existing CSS code bases. Not sure what's going on there.

[jaredwhite@indieweb.social]

@jensimmons I do not use LLMs on principle, and my work as a software developer and creator continues to hum along just fine.

I also consider slopcoding to be incompatible with free/libre software, so I am very concerned both by (a) increasing slop additions to ostensibly FLOSS codebases, and (b) impending ecosystem collapse due to open software's endless scraping & license stripping by LLM vendors.

[paulrosen@hachyderm.io]

@jensimmons A coworker uses it and comes up with something that looks like it works pretty quick. But he misses a bunch of cases, messes up error checking, and produces new components that are nearly identical to existing ones. The more I start to straighten it out, the more it looks bad. I could add the feature quicker if I didn't have to understand the slop first.

[torb@hachyderm.io]

@jensimmons I don’t use it and the pressure have convinced me that I need to invest further in understanding the systems I work with.

[gundersen@mastodon.social]

@jensimmons I expect LLMs will get better and the cost lower, which means simple webapp will become disposable. You don't need to plan or project an app for a department, they can just have one made and then throw it away when they are done, or generate a new one when the specs change. LLMs are much better at small projects (and always will, because of the cost of context) so it's cheaper to generate many disposable apps than maintain one large complex one.

[gundersen@mastodon.social]

@jensimmons beyond that it's also possible to produce bespoke apps. Instead of everyone using the same calendar or task management app, with hundreds of features that nobody uses, everyone can have their own custom generated app. Atlassian can generate a jira clone not for each of their customers but for each of their customers users. Millions of uniquely generated jira boards, each one tailored to its user.

[sknob@mamot.fr]

@jensimmons you’re the last major browser without LLM crap crammed into it. Make it a marketing argument. There is no ethical use of GenAI.

[gundersen@mastodon.social]

@jensimmons in the long term I'm thinking of just in time generated code from LLMs. If an LLM can generate the code fast enough, with some future hardware, then it doesn't need to write the code before it runs, it can generate it on the fly. Then you don't have to ship code around, you just need to ship the prompt. This obviously assumes a 0% error rate, so might not be realistic, but it's fun to speculate. Most of the world runs on JiTed js code, in the future you might download the prompt only

[dandean@indieweb.social]

RE: https://indieweb.social/@dandean/116053988050640356

@jensimmons

[jonikorpi@mastodon.gamedev.place]

@jensimmons Feeling disrespect towards the people building it, and sadness towards the people using it.

[marcusanthonycyganiak@mastodon.social]

@jensimmons

I’m currently using Gemini 3 Pro in Google Antigravity for WordPress theme development. It’s incredible.

[alahmnat@woof.tech]

@jensimmons I hate it, it's garbage, and when possible, I intend to avoid using applications that are written with it.

God I cannot wait for this damn bubble to pop already.

[mnmlst@indieweb.social]

@jensimmons RESIST

[lukesutton@mastodon.au]

@jensimmons can’t wait for the bubble to pop and for us to — potentially — salvage the really valuable stuff after. Lots of hype, little evidence.

I’m generally disappointed with contemporary web dev. There is a lot of JS slop all ready — slow, bloated, brittle — and I only see LLM accelerating that.

The next few years are gonna be rough.

[mvsde@mastodon.social]

@jensimmons Even though I’m not using LLMs at all, the constant onslaught of news and tech bro boosters has beaten me into a state, where I’m loosing interest in web development altogether.

Right now I’m hoping that I can just survive a little bit longer than the AI hype bubble. But the fun is already not there anymore. Doing my job feels like slowly getting sucked down a quicksand hole.

[beeps@social.beeps.gay]

@jensimmons Not a fan of the plagiarism machine, no. I can only imagine it’s already stolen things I’ve made before.

A lot of experienced developers I know are tempted to just quit the industry. They’re sick of organisations trying to force genAI tools on them, in turn devaluing their work and taking away the joy derived from making things and solving problems yourself.

I’ve used it for non-dev tasks before, mainly collating things or transforming data formats, but every bit of code I’ve gotten out of an LLM has been buggy or broken in some form, even really basic shit. I don’t trust it to write code at all.

[idiomatic@mastodon.social]

@jensimmons There are plenty of good reasons not to incorporate LLM code gen into one's workflow, but one I haven't seen mentioned here is that I fear that relying on such tools will, over time, make me a worse developer.

Like, I crochet as a hobby. If I used a magic machine that did most of the crocheting for me, my skills would plateau and inevitably atrophy.

Programming is a craft like any other, requiring deep, consistent practice to get good and stay good. LLMs hinder that practice (imo).