Did they get it right? I'm geniunly curious because from my point of view the only company to get AI right so far is Microsoft. This is from the point of view of an EU energy company married to Microsoft services mind you, but Microsoft is the only AI company which actually sells AI to us in a manner that can be considered useful. 7 months ago I thought Copilot (the m365 one, not the github one) was useless. Now I think the chat part is as good as what you'll get on the smaller subscriptions to ChatGPT or Claude, and it's Cowork is very good.

Obviously it's not like having Claude Code or similar, but those would be very hard to run legally and safely in EU regulated industries like us in energy production. You could run Codex but it would require your IT department to have yet another place to manage security and software integrations. Which is a fairly big issue when you have around 1 IT staff member per 300 employees, and an organisation which will buy five different sharepoint solutions for the same problem.

Microsoft sells something that seamlessly integrates into everything else you buy from them. It's easy to manage credit spending and when I say that I don't mean that it's easy to set it up. It's easy to track and budget so the right departments pay for it. It can read your emails, your onedrive, your loop, your sharepoint, your powerbi/fabric and anything else you have access to. Sure it's annoying to be confronted with a point you've made 7 months ago (like how you think Copilot is utter trash) and since changed opinions on it's frightingly good at finding relevant stuff, even in some obscoure legal document on your intranet.

I assume that Anthropic and OpenAI makes money of this somehow, but for the most parts our users wouldn't care if it was the GPT and Claude models or Mistral or Deepseek. Which is why I asked if they had really gotten it right.

As culi also touches on I think it depends on what is meant by frameworks. As someone who belives abstraction are horrible practice unless they serve a really good purpose though, I never bought the "this is easier" part of the big CSS frameworks. I fully accept that people who work on user facing web applications more than I do, which is rarely, had a much easier time using something like bootstrap/tailwind/whatever. I always found having to look up documentation on something wich was infinitely worse than the CSS documentation on developer.mozilla.org to be a a weird experience. I do fully accept that a lot of teams experiences would prove my personal opinions wrong on this.

On the flip-side I always found javascript frameworks to have a purpose. Though to be fair I don't think I've worked with user facing javascript without a framework since jquery came about. I guess you could call that a library, but still.

This is true but a lot of that history and culture has not been capable of producing software which has better quality than what came before. Around here banks are spending hundreds of millions trying to exit mainframe systems, and they've been failing for the better part of a decade. This despite having hundreds of people working on "modern" software paradigms in "modern" languages.

I do think the capabilities of AI depends a lot on how you work. If you're doing very low abstraction, fail fast defensive programming in something like Rust then it's fairly good at writing code with high quality. If you're working in some OOP language with tons of freedom, then my experience is that it's usually not better (or worse) than most human programmers, but certainly much worse than some.

I've got decades of experience in enterprise development and I've also been an external examiner for CS students. We have people with no coding back ground vibe coding internal tools in Python which are terrible, but not really worse than what we've previously bought from some of the cheaper coding houses around here. At the same time we have AI writing internal SDK's and libraries in Rust which will frankly write better code than I do on a thursday afternoon after too many meetings and too little sleep because the children kept me awake all week.

Good code can mean a lot of things though. If you're deep into clean SOLID DRY OOP programming I suspect that an AI will produce unreadable horrible code, much worse than most humans. On the flip-side the AI probably knows the difference between a generator and looping over a List. It probably also knows more about JVM optimization than most of those banking programmers trying to replace COBOL with JAVA.'

I doubt Antropic is a good sample though. Bun is written in Rust and it's a single purpose software, which seem to be an optimal environment for LLM's.

I think it was more a question of what that niche was meant to be. Bun does alright in the "we don't want external dependencies" space of Node development. We switched a lot of our Go projects to it because it allows us to use nothing but Microsoft packages for Azure and Bun, similar to how we were using Go, but now we have a language less. Bun is largely a dropin for node though, and it would be rather easy to switch back.

You don't need all those things, and for the most part you likely never did. Today I think most of it is legacy. A member of one of my teams recently wrote a node template for azure function apps, and when I reviewed it I asked them why they were using 20 dependencies when they needed 3. There was a lot of reasons, but the main reason was "that's what we've always done". One of these prevented updating to typescript 7 which is why I was reviewing it to begin with. A day later they're down to 4 dependencies that are actually being used for something useful.

SPA's and their frameworks still have their use cases. I don't think you can really get full "simplicity" when you have a lot of people working on the same code that goes into enterprise production. I do think that a lot of the more complicated stacks mainly exist because "that's how we've always done it". Which is frankly also a fair consideration if you're not in a position where changing it would generate a lot of value.

I find that it always comes down to people. The fewer people you are, the easier it is to make rapid changes. With a lot of people, and maybe people with very different backgrounds and skill-sets things changes. Change management isn't easy, and often the actual tech stack will matter a lot less than building a good team around it. You obviously still want to work toward changes and approvements, but switching a 100 people from their safe and overly complicated tech stack to something else will take a while. You'll also have to be damn sure it's worth it.

One intereting part about installed fonts is that many enterprise companies buy specific fonts. Then they are rolled out to all their devices, and sometimes at least one of them will have the company name as part of it's name.

Isn't this just the ebitengine doing all the work, with someone having used AI to vibe code TA functionality on top of it?

I've programmed my work keyboard to swap the places of the left windows key and left ctrl. I'll never go back to pressing ctrl with my little finger instead of my thumb again!

I've even been rebellious enough to swap the left ctrl to a command button.

> With Bun I think it was basically a marketing ploy.

I don't know, the vast majority of daily Microsoft Defender offenders I have to wade through, as well as the majority of the intune remidy requests are related to Node being used in various Microsoft applications. The Azure CLI is one of them, but the one which really ranks high on the security risk levels is always something "Agent" or "AI" related folder which I assume are what runs the Copilot and Cowork clients.

I think I had around 800 CVE warnings in relations to Axios in side a Microsoft tool related to AI running node this week, from just 8 devices.

Bun having it's own batteries included is probably a nice benefit to avoid that.

Now that Microsoft allows you to do /cost for individual tasks (or whatever they call them today). So I tasked Sol, Astra and Fable in cowork with exactly the same vibe coding task on the exact same zip file containing a code project I needed an update for. Astra used 20x and Fable used 15x of what Sol did.

Fable changed a lot of things I had explicitly told it not to change. Arguably a lot of them would've been correct if you didn't work in a place where abstractions are directly against the core principles, but what it produced was basically unusable. I'm not sure if Sol or Astra did best, they produced rather similar code outputs. Astra's was better, but Sol didn't do so bad. It forgot to clean up a few places after it's refactor and it made two bugs I had to correct but other than that it was fine. Astra on the flip-side might have produced code that didn't need changes but it also rewrote every piece of documentation so that it became horrible.

As far as the "experiment" goes, it just shows you that the credit consumption is basically pure magic. You'd think that the Microsoft AI admin tools and the Agent365 FOMO DLC license they sell might give you some sort of reporting, but it doesn't. What you can see is how many tokens a user consumes and the total number of tasks they've initiated as well as whatever running agents they have. You can't see what models they use or which tasks are expensive, which makes it very hard to help them. Early on we had an employee who hit their limit in an hour, and it turned out they had basically uploaded a lot of information and run it in a single long task that kept going over it again and again. We told them it might be a good idea to only give it what it needed and to create more tasks, and even though it's been three months, they have yet to consume as many credits as they did that first hour.

But that's how you support and track it. You see a user spend a lot, then you go to their computer and now that you can actually do the /cost thing, you go through their tasks and try and figure out where they're spending money...

It's obviously improving. A month ago /cost wasn't there and they just released a new dashboard for cowork, but it's still black magic that is impossible to govern.

I'm assuming you're using Microsoft Cowork or Copilot or something. I suspect it's system prompt and implementation (tooling/harness) isn't the same as that of OpenAI and Anthropic even if the underlying model is supposedly the same. The answers provided can be wildly different and often for the worst.

It's cowork. I'm in enterprise in the EU, so we're more restricted in what we use. My issue is mainly with how impossible it is to do any form of reporting on this. Obviously this isn't the most popular opinion among the people subject to it, but for most things we do in the Microsoft enterprise setup we can basically monitor every thing a device does. With Cowork you can't even see which individual task is eating a users credits. At least not yet.

Which is an issue when you need to get department managers to manage their budgets around the amounts of credits their employees spend. The more of a black box it is, the more governance and corporate bullshit you have to deal with.

The copilot part of it runs "unlimited" on the license. Except it's not unlimited, and this is even more of a blackbox because you can't see any sort of spending and the limit is listed as "extensive use".

Just btw, most assume when you say Cowork you're referring to Anthropic Cowork as it's the original rather than Microsoft's white labelled Cowork. And when you say Sol or whatever model you're referring to OpenAI and use on its servers. Microsoft's versions are not at parity with the originals. I feel your pain on being limited by silly enterprise restrictions.

On your point about "monitoring", personally I feel this is toxic corporate IT culture, enabled and perhaps pushed by the likes of Microsoft with all their tools, which they of course make money off. People have cellphones with cameras making most points in this area moot.

An alternative used in other big corporates is to set budgets, with tiered authorisation approvals for higher limits. The users and their managers can justify why and what they're doing that they need the additional tokens. This also encourages more efficient use of tokens on other work. More efficient use is sometimes counterintuitive. Laissez-faire generally works best.

Cowork requires user approvals for high risk actions such as emailing.