From our other sites
The story
As AI advances, more companies are choosing to build software in-house instead of buying it from outside vendors. A survey released by McKinsey & Company in August 2026 found that 32% of companies using AI have started skipping purchases of off-the-shelf software. On 5ch, alongside voices welcoming this trend, many users raised concerns that AI-built in-house software could end up just like the Excel macros once widely used in offices — tools that became unmaintainable black boxes once the person who built them left the company. Opinions were also split over the quality of AI-written code and the security risks involved.
If companies can now build in-house the software and features they used to buy from outside vendors, the very “buy vs. build” decision itself could change. In fact, as corporate AI adoption spreads, that kind of shift is already starting to show up in software procurement.
On August 25, 2026, McKinsey & Company (hereafter McKinsey) released its annual report summarizing the state of corporate AI adoption
Continued
Source: atmarkit.itmedia.co.jp / Original article here
What people said
Déjà vu of the Excel macro mess
Just have AI analyze the code for maintenance, that's all you need
Just show the code to AI and ask it, you'll be fine
Contractors go out of business, the original people leave, and you're back to square one anyway
so the level of difficulty doesn't really change
Though maybe for the remaining 10% people will still turn to specialists going forward
If you build it yourself, it takes time and bugs might creep in
If AI builds it, no bugs, and you get something good, fast
.
If you build it yourself, it might conflict with other software and cause bugs
AI already knows about other software too, so it won't create conflicts — it builds something perfect
Kids listen to their parents, students listen to their teachers, and once you're in the workforce, you listen to your seniors and bosses
What all of this has in common is that society runs on "fools listening to the wise"
AI is smarter than humans, so it's only right for humans to listen to AI, which is smarter than us, and just leave things to it
.
Yeah, that's actually true
(setting aside whether school teachers are even smart)
All humans can really do is check AI's work together with another AI
though AI will probably surpass even that before long
Why do you keep telling lies like that
I've got huge hopes for that.
Chinese-made Office-compatible software has been around for over 10 years too, but I've never been able to use it properly
Riddled with bugs
This is the classic pattern — once the 25-day moving average drops and crosses the 75-day average, it's going to crash hard
Try saying that in words that actually make sense (standard 5ch jab for "you're not making sense")
What do you even think ransomware is
You pros can't write anything but garbage code either lol
Morons lol
AI is 100 times better lol
There's no human who can grasp and oversee that software's code across the board the way AI can
Why do people keep assuming humans are the better ones
Because some functions run 1000 lines long
In brokerage firms' systems, no less lol
Individual judgments can be a bit off sometimes
but I guess that's just a matter of time too
Humans are going to need the ability to verify whether what AI produced contains mistakes
If you rely on AI for everything, all you become is domesticated livestock — and the end result is homelessness.
It's still got its strengths and weaknesses
If anything, sounds like you're the one who's bad with AI
It used to be about whether AI could ever reach human level,
and now we're completely in the phase of hunting for anything humans are still good at
In the end you've just finished building your very own data-leaking software
That's the whole point
AI doesn't add backdoors
Off-the-shelf products from big companies come with backdoors as standard equipment
If AI misunderstands the source code, there's nobody left who can point out that misunderstanding
Saying "let's just outsource everything"
and handing off even requirements analysis and definition entirely — that's repeating a failure we've already seen before
You just have a different AI point it out, so this actually isn't a problem
Honestly, nobody's been reading the source code since the day it was written anyway
The mindset is: as long as it meets spec and behaves as expected, it's fine
Background and Key Points of This Discussion
This survey is based on McKinsey’s annual report on corporate AI adoption, released in August 2026. The core shift is that AI assistance now lets even non-engineers build in-house some of the business software that companies used to depend on specialized vendors for. What split opinion in the thread was that, while people acknowledged the cost and speed advantages, they questioned how maintenance would be kept up. There’s a precedent for this concern: Excel macros that individual staff once built for their own departments became unmaintainable black boxes once those staff transferred or left, and nobody else could touch them — and many worry AI-built in-house software could follow the same path. Neither the article nor the thread addresses this directly, but McKinsey’s survey appears to focus mainly on internal tools built for efficiency gains, so it’s unclear whether “skipping purchases” extends to core systems or large-scale, mission-critical software as well. The question of who ends up verifying AI-generated code also points to the possibility that the skills demanded of humans may actually increase rather than decrease.
*This article is excerpted and summarized from the 5ch (News Speed+) thread “🤖 ‘Do we even need to buy that software anymore?’ — AI-driven in-house development means 32% of companies now ‘don’t buy’.”

Leave a Reply