One of the current top 100 posts relates to western religion. It’s easy to avoid if uninterested. I enjoy that every now and then we have an ancient history, archeology, theology, literature, futurism or etc. post make the front page.
And don't forget the safety in getting to say "our systems are down because of [X TRUSTED SOFTWARE FROM LARGE KNOWN BRAND] and we're just waiting for them to fix it" instead of "our shitty internal tooling is broken and no one knows how to fix it"
This feels like it goes along the lines of "people's vibe code is cluttering up our PR's, people still need to review" -- it misses the boat: models are already capable of getting you up to speed on how the code is organized and works, in as much as you want to or need to be up to speed. They are already helping me cut down review time because I don't need to aimlessly hop around, I have a good starting point that I can scrutinize and dialogue about. Same thing here: employee leaves company -- about 3 years ago you would be right, now the company is left with an unmaintainable mess of legacy code and tech debt. TODAY this just doesn't matter. No one really needs to read that code too closely, it's already easy for agents to digest and explain and modify.
Doing this today, in production, with full trust, is clearly not wise, but the writing is clearly on the wall that this is going to be the norm more and more over the coming years. The times they are a-changin.
Can you be clear what you mean? What are you saying has not worked once?
And "it will be the norm" is a clear corollary of, absent any significant and unforeseen roadblock, even with the current highly imperfect agentic sausage-making factories we have today, what capabilities will be in like 6-12 months time.
When the effectiveness of vibe coding an internal workflow was measured, only 5% of efforts worked. That's pretty rare and bad. And those are the early adopters who tend to be more adaptable and effective with new technology. That's a big part of why people don't believe the (your) hype. Its great at simple things with a low cost of failure. Problem is, its rare to pay a good wage for simple things with a low cost of failure. Also, writing new code isn't a big part of most engineers jobs. To put it more simply, vibe coding optimizes the wrong things about engineering.
I think you can avoid the pain by thoughtfully designing it to avoid lock-in. You want it so that if needed, a dev can vibe-code a migration tool to the equivalent SaaS offering. AI lowers the barrier for creating these in-house replacements, but it also lowers the barrier for scrapping them too.
The thing about lower barriers is that it makes it easier for e.g. Salesforce to raise the level of expectations. And that's the moving target. New employees will come from elsewhere and wonder how a company is operating using tools from 2020 when X, Y, Z are becoming industry standard.
The key here is that the moving target will _never_ be "what can 1-2 people vibe code without any expectation of being the best at what it does?"
(Also: training people on bespoke tools takes much longer than training on configurations of standard tools. Imagine if you had to learn a new source control system at every job, like in the '80s.)
4. Employee continue in the company and request help (or the managers see the need):
4.1 They hire more, but if all are vibe-coders too
4.1.1 The product gets more complicated (no more complex, that good developers can manage!)
4.1.2 Bugs creep in, feature request pile up.
4.1.3 People start to get desperate, not worries! now:
4.1.3.1 Somebody vibe-code a new alternative that solves
the immediate problem
4.1.3.2 Bugs creep in, feature request pile up.
4.1.3.3 Needs to sync with the other tools
4.1.3.3.1 Somebody vibe-code the sync that solves the immediate problem
(the saga continue)
In parallel:
4.2 Eventually is obvious that need external help
THEN:
4.2.1 They ask for consultors for build tool, of course, from a company that has embraced the IA!
4.2.2 They build new shinny tool!
4.2.3 Bugs creep in, feature request pile up.
4.2.4 Needs to sync with the other tools
....
AND:
4.3.1 They ask for consultors, to teach them what to do, of course, from a company that has embraced the IA!
4.3.2 New shinny theory of how do the thing with IA is now being implemented!
4.3.3 It require a rewrite of not only past solutions but, a change of how the company behave!
4.3.3.1 Needs to sync with the other tools
.......
4.3.4 And it spark beautiful office/political debates around some philosophical whatever that also trigger changes in the structure, hiring or whatever, alienating the people that has been working there, that after months, has started getting the handle of it!
4.3.5 Employees either leaves the company or moves on to another project.
4.3.6 New employees arrive, with a wild new IA tool and different vibes that vibe-coding!
... the saga continues
5. Is now clear that it need to buy a product form a well stablished software provider
If I understand correctly many organizations will not develop original stuff internally, because nobody internally wants to be the one is shouted at if something goes wrong.
That's a huge part of it. But also you presumably hired a full-time programmer for a reason, and in almost every case that reason was not to have somebody to write and maintain your CRM system. So any system they build and maintain is not just another thing for you to worry about, it's a huge chunk of time that the developer isn't doing what you hired them for.
They're in series A right now, so most likely yes they will fail like most new businesses do. Internal chat is a hugely competitive field with multiple large players (and Teams has had Copilot for a while now). So, I wish them the best, but this is pretty much exactly the kind of "existing product + AI" mashups that will have a tendency to burn out over the next couple of years, just like "our old business but with .com at the end" did back in 1999-2000.
this is the argument i continue to have with people. first mover isnt always an advantage - i think openai will be sold or pennies on these dollars someday (next 5 years after they run out of funding).
Google has data, TPUs, and a shitload of cash to burn
I'm not sure because google was by far the best search engine for a long time in the early 2000s and there are a lot of models close to what openai has right now.
Name recognition only gets you so far. "Just Google it" happened because Google was better than Hotbot/Altavista/Yahoo! etc by orders of magnitude. Nobody even bothered to launch a competing search engine in the 2000s because of this (until Microsoft w/ Bing in 2009). There is no such parallel with ChatGPT; Google, Bing, even DuckDuckGo has AI search.
First mover advantage matters only if it has long-lasting network effects. American schools are run on Chromebooks and Google Docs/Slides, but these have no penetration in enterprise, as college students have been discovering when they enter their first jobs.
reply