Comment by msy
21 hours ago
We suddenly have a proliferation of new internal tools and resources, nearly all of which are barely functional and largely useless with no discernible impact on the overall business trajectory but sure do seem to help come promo time.
Barely an hour goes by without a new 4-page document about something that that everyone is apparently ment to read, digest and respond to, despite its 'author' having done none of those steps, it's starting to feel actively adversarial.
Without good management AI is just a new way to make terrible work in unprecedented quantities.
With good management you will get great work faster.
The distinguishing feature between organisations competing in the AI era is process. AI can automate a lot of the work but the human side owns process. If it’s no good everything collapses. Functional companies become hyper functional while dysfunctional companies will collapse.
Bad ideas used to be warded off by workers who in some shape or form of malicious compliance just would slow down and redirect the work while advocating for better solutions.
That can’t happen as much anymore as your manager or CEO can vibe code stuff and throw it down the pipeline for the workers to fix.
If you have bad processes your company will die, or shrivel or stagnate at best. Companies with good process will beat you.
[flagged]
We had a coworker vibecode an internal tool, do a bunch of marketing to the company at how incredible it is. Then got hired somewhere else.
I just went and deleted it because it's completely broken at every edge case and half of the happy paths too.
My team has also adopted this - it's much easier to add another layer than to refine or simplify what exists. We have AI skills to help us debug microservices that call microservices that have circular dependencies.
This was possible before but someone would maybe notice the insane spaghetti. Now it's just "we'll fix it with another layer of noodles".
That's so interesting because where I work, the push was to "add one more API" to existing services, turning them into near monoliths for the sake of deployment and access. Still a mess of util and helper functions recursively calling each other, but at least it's one binary in one container.
Unfortunately I saw this pre-AI with microservices, where while empowering developers with their beloved microservices, we create intense complexity and deployment headaches. AI will fix the slop with an obscuring layer of complexity on top.
Are you concerned this will just lead to coupling everywhere like microservices tend to do?
My main use of vibecoding is creating dozens of internal tools that have sped up tasks, or made tasks possible that were previously not. These tools would have taken weeks of time to build manually and would have been hard to justify, rather than just struggling with manual processes every now and again. AI has been life-changing in creating these kinda janky tools with janky UI that do everything they're supposed to perfectly, but are ugly as hell.
Are you able to describe any of those internal tools in more detail? How important are they on average? (For example, at a prior job I spent a bit of time creating a slackbot command "/wtf acronym" which would query our company's giant glossary of acronyms and return the definition. It wasn't very popular (read: not very useful/important) but it saved myself some time at least looking things up (saving more time than it took to create I'm sure). I'd expect modern LLMs to be able to recreate it within a few minutes as a one-shot task.)
It's almost always a CRUD app or dashboard that no one uses while being extremely overkill for their use case.
edit: LOL called it, a bunch of useless garbage that no one really cares about but used to justify corporate jobs programs.
4 replies →
I have one that serves a few functions- Tracks certificates and licenses (you can export certs in any of the majorly requested formats), a dashboard that tells you when licenses and certs are close to expiring, a user count, a notification system for alerts (otherwise it's a mostly buried Teams channel most people miss), a Downtime Tracker that doesn't require people to input easily calculatable fields, a way for teams to reset their service account password and manage permissions, as well as add, remove, switch which project is sponsoring which person, edit points of contact, verify project statuses, and a lot more. It even has some quick charts that pull from our Jira helpdesk queue- charts that people used to run once a week for a meeting are just live now in one place. It also has application statuses and links, and a lot more.
I'd been fighting to make this for two years and kept getting told no. I got claude to make a PoC in a day, then got management support to continue for a couple weeks. It's super beneficial, and targets so many of our pain points that really bog us down.
6 replies →
The ones I can mention.. one that watches a specific web site until an offer that is listed expires and then clicks renew (happens about once a day, but there is no automated way in the system to do it and having the app do it saves it being unlisted for hours and saves someone logging in to do it). Several that download specific combinations of documents from several different portals, where the user would just suck it up previously and right-click on each one to save it (this has a bunch of heuristics because it really required a human before to determine which links to click and in what order, but Claude was able to determine a solid algo for it). Another one that opens PDFs and pulls the titles and dates from the first page of the documents, which again was just done manually before, but now sends the docs via Gemma4 free API on Google to extract the data (the docs are a mess of thousands of different layouts).
1 reply →
We’re seeing the exact same where I work. Our main Slack channels have become inundated with “new tool announcements!”, multiple per day, often solving duplicate problems or problems that don’t exist. We’ve had to stop using those channels for any real conversation because most people are muting them due to the slop noise.
And what’s worse is that when someone does build a decent tool, you can’t help but be skeptical because of all the absolute slop that has come out. And everyone thinks their slop doesn’t stink, so you can’t take them at their word when they say it doesn’t. Even in this thread, how are you to know who is talking about building something useful vs something they think is useful?
A lot of people that have always wanted to be developers but didn’t have the skills are now empowered to go and build… things. But AI hasn’t equipped them with the skill of understanding if it actually makes sense to build a thing, or how to maintain it, or how to evolve it, or how to integrate it with other tools. And then they get upset when you tell them their tool isn’t the best thing since sliced bread. It’s exhausting, and I think we’ve yet to see the true consequences of the slop firehose.
I'm sorry to hear that you have people abusing their new superpowers.
I run a team and am spending my time/tokens on serious pain points.
Such as?
I'll throw this out as something where it has saved literally weeks of work: debugging pathological behaviour in third-party code. Prompt example: "Today, when I did U, V, and W. I ended up with X happening. I fixed it by doing Y. The second time I tried, Z happened instead (which was the expected behaviour). Can you work out a plausible explanation for why X happened the first time and why Y fixed it? Please keep track of the specific lines of code where the behaviour difference shows up."
This is in a real-time stateful system, not a system where I'd necessarily expect the exact same thing to happen every time. I just wanted to understand why it behaved differently because there wasn't any obvious reason, to me, why it would.
The explanation it came back with was pretty wild. It essentially boiled down to a module not being adequately initialized before it was used the first time and then it maintained its state from then on out. The narrative touched a lot of code, and the source references it provided did an excellent job of walking me through the narrative. I independently validated the explanation using some telemetry data that the LLM didn't have access to. It was correct. This would have taken me a very long time to work out by hand.
Edit: I have done this multiple times and have been blown away each time.
3 replies →
I answered this in a different comment below, but a lot of the friction is around the amount of time it takes to test/review/submit etc, and a lot of this is centered around tooling that no one has had the time to improve, perf problems in clunky processes that have been around longer than anyone individual, and other things of this nature. Addressing these issues is now approachable and doable in one's "spare time".
1 reply →
For me/my team, I use it to fix DevProd pain points that I would otherwise never get the investment to go solve. Just removed Webpack for Rspack, for example. Could easily do it myself, which is why I can prompt it correctly and review the output properly, but I can let it run while I’m in meetings over more important product or architectural decisions
Creating stakeholder value
5 replies →
>Such as?
it's crazy that the experiences are still so wildly varying that we get people that use this strategy as a 'valid' gotcha.
AI works for the vast majority of nowhere-near-the-edge CS work -- you know, all the stuff the majority of people have to do every day.
I don't touch any kind of SQL manually anymore. I don't touch iptables or UFW. I don't touch polkit, dbus, or any other human-hostile IPC anymore. I don't write cron jobs, or system unit files. I query for documentation rather than slogging through a stupid web wiki or equivalent. a decent LLM model does it all with fairly easy 5-10 word prompts.
ever do real work with a mic and speech-to-text? It's 50x'd by LLM support. Gone are the days of saying "H T T P COLON FORWARD SLASH FORWARD SLASH W W W".
this isn't some untested frontier land anymore. People that embrace it find it really empowering except on the edges, and even those state-of-the-art edge people are using it to do the crap work.
This whole "Yeah, well let me see the proof!" ostrich-head-in-the-sand thing works about as long as it takes for everyone to make you eat their dust.
6 replies →
> but sure do seem to help come promo time.
I personally noticed this. The speed at which development was happening at one gig I had was impossible to keep up with without agentic development, and serious review wasn't really possibile because there wasn't really even time to learn the codebase. Had a huge stack of rules and MCPs to leverage that kinda kept things on the rails and apps were coming out but like, for why? It was like we were all just abandoning the idea of good code and caring about the user and just trying to close tickets and keep management/the client happy, I'm not sure if anyone anywhere on the line was measuring real world outcomes. Apparently the client was thrilled.
It felt like... You know that story where two economists pass each other fifty bucks back and forth and in doing so skyrocket the local GDP? Felt like that.
>Barely an hour goes by without a new 4-page document about something that that everyone is apparently ment to read, digest and respond to, despite its 'author' having done none of those steps, it's starting to feel actively adversarial.
well, isn't that what AI can be used effectively for - to generate [auto]response to the AI generated content.
What a delightful world we're building.
That's not on Claude, that's on the authors.
Claude is a tool. It can be abused, or used in a sloppy way. But it can also be used rigorously.
I've been beating my team to be more papercut-free in the tooling they develop and it's been rough mostly because of the velocity.
But overall it's a huge net positive.
Sounds like a workplace wide DDoS.
Im convinced none of these people have any training in corporate finance. For if they did they'd realise they were wasting money.
I guess you gotta look busy. But the stick will come when the shareholders look at the income statement and ask... So I see an increase in operating expenses. Let me go calculate the ROIC. Hm its lower, what to do? Oh I know, lets fire the people who caused this (it wont be the C-Suite or management who takes the fall) lmao.
Do you really think companies have started spending millions on tokens and no one from finance has been involved?
You could argue that all the spending is wasted (doubtless some is), but insisting that the decision is being made in complete ignorance of financial concerns reeks of that “everyone’s dumb but me” energy.
There is a difference to just noticing and attributing it to and recognizing negative financial outcomes. Right now for most companies they are still adjusting to declining inflation. Their bottom lines are doing quite well because consumer price inflation is much stickier than supply inflation. We are coming off of one of the quickest and largest supply lead inflationary cycles. It may not be immediately apparent for many companies that new expenditures are a drag on profitability.
The real thing to look at is whether or not the future outlook for company AI spend is heading up or down?
What a finance team allocates on spend has nothing to do with what the tokens actually get used for.
Are they peeking over the shoulder of each team and individual? Of course not.
It can be the case that the spend is absolutely wasteful. Numbers don’t lie.
> Do you really think companies have started spending millions on tokens and no one from finance has been involved?
Oh, they were involved all right. They ran their analyses and realized that the increase in Acme Corp's share price from becoming "AI-enabled" will pay for the tokens several times over. For today. They plan to be retired before tomorrow.
6 replies →
More that there is a poor incentive structure. Just like how PE can make money by leveraged buyouts and running businesses into the ground. Many of the financial instruments that make both that and the current AI bubble possible were legal then made illegal within the lifetimes of the last 16 presidents.
Round-tripping used to be regulated. SPVs used to be regulated. If you need a loan you used to have to go to something called a bank, now it comes from ???? who knows drug cartels, child traffickers, blackstone, russians & chinese oligarchs. Even assuming it doesn't collapse tommorow why should they make double digit returns on AI datacenters built on the backs of Americans?
1 reply →
I'm sorry to hear you have such poor leadership.