https://news.abnasia.org/blog/posts/en-microsoft-scales-back-ai-goals-because-almost-nobody-is-using-copilot-2732
This headline nailed it! Turns out, Microsoft just learned the hardest lesson in AI - distribution doesn’t beat usefulness 😳
Microsoft’s AI Copilot was supposed to be everywhere.
In Windows. In Office. In your workflow.
Turns out it’s mostly ignored.
Recent reports say Microsoft quietly cut internal Copilot sales targets by up to 50%.
Not because of vibes. Because of math.
→ Copilot ~14% market share → ChatGPT ~61% → Gemini sprinting into 2nd place
And this is with Microsoft’s insane advantage:
Windows + Office + Azure + OpenAI access 🤯
If that stack can’t force adoption, maybe the problem isn’t distribution. It’s value.
Enterprises tried Copilot. Piloted it. Demoed it. Bought licenses.
Then, employees opened ChatGPT in another tab.
Because most of today’s “AI agents” are confident interns with no context.
So when Microsoft says“70% of Fortune 500 have adopted Copilot”, what it really means is this:
Procurement bought it. Employees didn’t.
Most importantly, forcing AI into everything didn’t help.
People didn’t ask for:
→ AI in Paint → AI watching their documents → AI narrating PowerPoint like a hostage video
They asked for one thing: AI that actually saves time, or does something humans couldn’t do before.
Right now, Copilot does neither.
Some extra link:
https://www.youtube.com/watch?v=QF4VccxdNEg
Test Confirms Copilot Can’t Do What Microsoft’s Ad Shows - https://propakistani.pk/2025/12/20/test-confirms-copilot-cant-do-what-microsofts-ad-shows/
AI search engines fail accuracy test, study finds 60% error rate - https://www.techspot.com/news/107101-new-study-finds-ai-search-tools-60-percent.html
It's buggy and not as fast as manual coding
I mean you can state that, but most disagree. We're very in as lemmy bubble here.
Manual coding is buggy too. If your non ai assisted code was buggy, so still will be your assisted code. I think the idea that its inherently a bug exponentializer sounds more like cope than grounded reality.
More than that, code focused llms can be much more efficient with the targeted focus and if someone desires, can be based on permissively licensed code.
Wasn’t there a recent METR study that found 20% decreased productivity with ai coding tools? Oddly enough, the people using the tools thought they were 20% faster.
https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
From my own experience, they can be useful until they aren’t… and if you don’t know what you’re doing they can output convincing but flawed or downright dangerous code or suggestions. I’m not sure if it saves me time or not. Im not doing front end web development anymore so maybe the stuff I’m working on now is too obscure for the current tools?
The "not as fast" thing is confirmed by a study, which the other reply to your comment links to:
Also vibe coding is unsutable for junior devs because junior devs don't have the skill level needed to debug AI code.
The study you linked that specifically says it cannot be used to confirm exactly what you are saying it confirms?