As a .NET developer who actually likes some Microsoft products, I can say this: the Copilot series is the worst thing they've shipped since Internet Explorer—and honestly, it might overtake it. The sad part is they had a huge head start before competitors gained access to powerful models, yet this is what we got.
If you haven’t seen how bad it is, here’s one example: Copilot Terminal. In theory, it should help you with terminal commands. Sounds great. In practice, it installs a chat panel on the right side of your terminal that has zero integration with the terminal itself.
It can’t read what’s written, it can’t send commands, it has no context, and the model response time is awful. What’s the point of a “terminal assistant” that can’t actually assist the terminal?
This lack of real integration is basically the core design of most Copilot products. If you’ve been lucky enough to avoid them, good for you. If your company forces you to use them because they’re bundled with a Microsoft license, I genuinely feel your pain.
Yes, but...does it? How do you know it isn't missing key points raised in the meeting? Not collecting actionable items?
Summarising seems like the absolute lamest thing an LLM could do for me. Like I want a Reader's Digest of my life, written but a word guessing machine.
This is arguably the only use case where a chat interface (or just text) actually works, because it's an information compression task. But building a $50 billion business strategy just on the fact that AI can recap a meeting is risky. If this is the only feature that works well, then the business ROI is questionable
> lack of real integration is basically the core design of most Copilot products
I would wager a month's wages that this is the doing of some internal Security Review, wherein a bunch of out-of-touchers decided that the customers will want to prefer to be Safe and Secure instead of getting some actual value from integrating copilot into shell workflows.
Meanwhile people are yolo'ing it with various janky DIY wires and duct-tape githobbits that mash together whatever open weights model and user-level access to the system (or worse).
> a bunch of out-of-touchers decided that the customers will want to prefer to be Safe
You mean the other way around, right? Because what could possibly go wrong when we let a language model hallucinate its way through which terminal command rhymes best with your prompt according to that SO comment from training data.
i mentioned this upthread but an LLM with enough access to be fully integrated into all apps/services/files in an enterprise managed workstation sounds like privilege escalation attacks just waiting to happen.
You'd be hearing a few more hair raising failure stories if they hadn't done that. And possibly a few big customers institute Win11 or Copilot site wide bans. Or just straight up going out of business. In businesses other than software, it's possible to be legally liable for mistakes.
> This lack of real integration is basically the core design of most Copilot products.
I think they're scared of the very real security issues with LLMs that may be unsolvable. It's not wise to give an llm free reign, at best maybe across your local computer but to be fully integrated into every application and every file it needs root. That would be the front door to many privilege escalation incidents in an enterprise managed laptop/desktop.
Try to think about it like a bunch of merchants running to line up in front of you as fast as possible with anything that resembles a product you might buy. The only thing they were successful at thus far was recognizing that something needs to be in your terminal, and they ran as fast as possible to your terminal with anything to beat others to the front of the line. I suppose they are doing this with all their efforts. A universe where a C-exec said in no uncertain terms, "get anything out there", is the very universe we're in.
Apple is the only merchant not running to line up with anything at the moment.
IMHO, one company needs to make the bold move and make a fork of their OS that is AI native with AI native apps/workflow and phase out the old paradigm. It'll have to be two product lines, but I think the new OS will have uptake like we've never seen before.
AI works. It's actually useful. Since GPT-4, tool calling capability is good enough. It's trivial to do a better job than Copilot on any task using any current model of the major LLM providers. I'm not talking API, even with basic chat frontend, regular users easily beat Copilot by simply copy-pasting between Word/Excel and the chat frontend.
If a twelve year old can one-shot a better product for any given use case than Microsoft Copilot, then it's not just "merchants running to line up in front of you", something more basic must be broken.
When was the last time Microsoft had a unified vision that was focused on building an amazing line of products that integrated well with each other?
I can only think of short snippets in history where they moved in that direction for maybe a year or two & then went scatterbrain.
Microsoft has benefited from a monopoly in the enterprise and has never been forced to innovate from a product perspective. See Slack/Teams as a case study of how they have operated when even slightly pushed.
* Edit - .NET, C#, TypeScript teams are an exception to the above. Highly underrated. Amazing talent there. Not sure who all gets credit. Anders & Mads for sure though.
I'd say the late '80s - early '90s when Microsoft was building the early versions of Microsoft Office. Integration among productivity apps was one of the key points of competition among all of the office suites of that era.
There were other huge coordinated efforts like the TwC initiative and the Windows 10 refactoring but those were invisible to end users.
You are right if talking about the efforts improving C# and CLR, taking the lessons out of Midori, Blazor and Aspire.
Regarding F#, VB, and C++/CLI, the poor souls are on a lifeline that makes CLR nowadays stand out for C# Language Runtime, instead of the original Common.
And the chaos that reigns on Windows Forms, WPF, WinUI, MAUI certainly isn't helping.
Finally they are constrained in what they can put into C# DevKit as means to not canibalize Visual Studio and Windows sales, it can only be good enough, not great and feature parity.
Also regardless of the reasoning behind it, having TypeScript rewriten into C# instead of Go would have been a great opportunity to make C# more relevant outside Microsoft shops, instead anyone looking to contribute to TypeScript compiler will be learning Go instead.
The crazy thing is Microsoft was so early with an AI product but was burned pretty badly when it instantly turned into a Nazi. Funny how the constant complaint on Twitter is how the modern AI agents are too "woke" and how Elon has to keep fighting his own AI model to conform with his viewpoints.
If you haven’t seen how bad it is, here’s one example: Copilot Terminal. In theory, it should help you with terminal commands. Sounds great. In practice, it installs a chat panel on the right side of your terminal that has zero integration with the terminal itself. It can’t read what’s written, it can’t send commands, it has no context, and the model response time is awful. What’s the point of a “terminal assistant” that can’t actually assist the terminal?
This lack of real integration is basically the core design of most Copilot products. If you’ve been lucky enough to avoid them, good for you. If your company forces you to use them because they’re bundled with a Microsoft license, I genuinely feel your pain.