I don't know which would be sadder. If the Microsoft employees hate their jobs more than anything for being forced to publicly defend this, or if they're being sincere.
I was under the impression that dogfooding was a good practice for quality control that improves your products. Not anymore, apparently. Now we make fun of them for doing that.
Of course the team developing these AI models should be dogfooding them. Telling the rest of the company they have to use them is pushing tools without understanding contexts.
If AI is that much more amazing, just measure employees on the usual metrics and those using AI should be so obviously far ahead that you can get rid of the others. Measuring "usage of AI" is a garbage metric that will not achieve anything good.
It could be they need to force usage to generate more quality training data over the MS codebase. You'll get a huge amount of prompt/llm_answer/human_correction instances if all MS programmers are forced to produce these e en if initially it does not increase or even decreases their productivity.
We saw the same in the early offshoring days. Firms forcing this on their project leads, knowingly expecting, accepting and swallowing the productivity losses to learn how to set this up for the promise of future cost savings.
> Not anymore, apparently. Now we make fun of them for doing that.
What ? do things when it makes sense, not as a top-down mandate.
Somehow, ceos/management have gotten AI religion, are now thrusting it top-down. This is like management dictating the tech-stack to use when it is not in their competency area.
Yeah the shoving is stupid. They are trying to accelerate it because there is so much FUD around LLM code generation slowing down adoption. All the biggest model shops (MS/DS/Anthropic) are at an extreme advantage right now. Internal use would be scheduled ahead of paying customers if I was in charge of it.
Imagine your company announced a new initiative where everybody's job would become "manager".
Would you do to work for a company where every single person including yourself was a manager?
This idea is silly because it asserts that we can replace work with management. But management is just more work, so we can just replace it with more management, and I think if you follow the logic to its outcome no work needs to be done by anyone anymore ever
They're even pushing their customers. Microsoft always pushes "adoption". They don't understand some users just have no need for or don't like certain features.
I work a lot with Microsoft consultants and that constant adoption evangelism is exhausting. It's not like their tools are even very good anyway. But they're as pushy as Jehovah's Witnesses.
They also see themselves too much as a 'partner', with the associated air of wanting to tell us what to do. They're a vendor, nothing more.
My guess...they want to their employees actions as some sort of training data?
Makes sense if you think about it. Really force everyone to use AI tools, check what prompts they've entered, and analyze areas where you can cut headcount.
A much more likely scenario is that it's to hit some sort of internal metric, to have a certain percentage use of AI tools is a goal for management somewhere.
Yes this is exactly how Microsoft works. Every little metric is a KPI for someone. The little king of the hill for using @mention in teams. The king for using using the Edge shopping bar. The king for using the weather widget on the task bar.
And each of them seems to have free reign to Hassle the users as much as they can.
Or the flip side - copilot is an unorganized mess and feels way behind OpenAI. Maybe making people use it will put pressure to make it better. Windows NT -- David Cutler has a rule about eating your own dogfood. I don't think they eat enough of it at Microsoft anymore with the lack of build quality in the last few years when they laid off their big QA team.
Some of the stuff I've seen M365 Copilot do is just flat out broken and awful and it's presented in their enterprise (stable) channels.
My personal (probably unlikely) theory is that they plan to lay off a lot of staff anyway, so they wanna get them hooked and reliant on Copilot first so that they're lifelong customers wherever they go next
If that were true, they'd just open up copilot agent for general access for any open source project, otherwise their training data will skew toward Microsoft-centric interactions on Microsoft-centric languages and processes
I was going to offer the terraform or ansible organizations as examples, since I'd guess a lot of those bugfixes are rote work, but then realized there's no way Microsoft would assist IBM to ship products
Can't be qemu, given their recent no-slop policy, so maybe the idea is harder than I thought
I don't really think about azure because I choose not to support Microsoft. Also, as of now I have had very little personal use for running servers, besides things that can easily be hosted almost anywhere.
I'm sure it's convenient for some people with certain projects.
I think Linux is a superior operating system to Windows in almost every way. This is supported by the fact that over 60% of azure servers run Linux [1].
Windows operating system is a walled off garden of intellectual property that wastes billions of dollars every year. Schools, government agencies, hospitals, etc all are spending an incredible amount on proprietary software and this is mainly benefitting a very small amount of people. Yet, thier shitty spyware OS is actually worse then the open source alternative. I think this is a moral failure for humans.
A startup I worked at a few years ago used Azure. In theory it’s _fine_ but I wasted many hours tracking down phantom issues because the Azure admin UI was super buggy.
I am all for it the copilot agent PRs on GitHub and the interactions of their employees with it are hilarious.
https://github.com/dotnet/runtime/pull/115762
https://github.com/dotnet/runtime/pull/115743
https://thedailywtf.com/ will have practically unlimited content. It may have to change the unit of time in its name just to keep up
I don't know which would be sadder. If the Microsoft employees hate their jobs more than anything for being forced to publicly defend this, or if they're being sincere.
I was under the impression that dogfooding was a good practice for quality control that improves your products. Not anymore, apparently. Now we make fun of them for doing that.
Of course the team developing these AI models should be dogfooding them. Telling the rest of the company they have to use them is pushing tools without understanding contexts.
If AI is that much more amazing, just measure employees on the usual metrics and those using AI should be so obviously far ahead that you can get rid of the others. Measuring "usage of AI" is a garbage metric that will not achieve anything good.
It could be they need to force usage to generate more quality training data over the MS codebase. You'll get a huge amount of prompt/llm_answer/human_correction instances if all MS programmers are forced to produce these e en if initially it does not increase or even decreases their productivity.
We saw the same in the early offshoring days. Firms forcing this on their project leads, knowingly expecting, accepting and swallowing the productivity losses to learn how to set this up for the promise of future cost savings.
> Not anymore, apparently. Now we make fun of them for doing that.
What ? do things when it makes sense, not as a top-down mandate.
Somehow, ceos/management have gotten AI religion, are now thrusting it top-down. This is like management dictating the tech-stack to use when it is not in their competency area.
Not only is it good for the product, but in this case we are talking about Gemini Pro, one of the top models in the world.
It's good for the engineers involved too. In 5 years, developers not using LLMs for code generation are going to be unemployable.
I really hate all of these executives trying to shove AI our throats.
Instead of rewarding our efficiency using AI with maybe a 3-4 day work week, they’ll just lay us off instead
Yeah the shoving is stupid. They are trying to accelerate it because there is so much FUD around LLM code generation slowing down adoption. All the biggest model shops (MS/DS/Anthropic) are at an extreme advantage right now. Internal use would be scheduled ahead of paying customers if I was in charge of it.
They're doing to to eventually replace you or your colleague with said AI
Why?
Imagine your company announced a new initiative where everybody's job would become "manager".
Would you do to work for a company where every single person including yourself was a manager?
This idea is silly because it asserts that we can replace work with management. But management is just more work, so we can just replace it with more management, and I think if you follow the logic to its outcome no work needs to be done by anyone anymore ever
... Because of the shoving
They're even pushing their customers. Microsoft always pushes "adoption". They don't understand some users just have no need for or don't like certain features.
I work a lot with Microsoft consultants and that constant adoption evangelism is exhausting. It's not like their tools are even very good anyway. But they're as pushy as Jehovah's Witnesses.
They also see themselves too much as a 'partner', with the associated air of wanting to tell us what to do. They're a vendor, nothing more.
My guess...they want to their employees actions as some sort of training data?
Makes sense if you think about it. Really force everyone to use AI tools, check what prompts they've entered, and analyze areas where you can cut headcount.
A much more likely scenario is that it's to hit some sort of internal metric, to have a certain percentage use of AI tools is a goal for management somewhere.
Yes this is exactly how Microsoft works. Every little metric is a KPI for someone. The little king of the hill for using @mention in teams. The king for using using the Edge shopping bar. The king for using the weather widget on the task bar.
And each of them seems to have free reign to Hassle the users as much as they can.
Or the flip side - copilot is an unorganized mess and feels way behind OpenAI. Maybe making people use it will put pressure to make it better. Windows NT -- David Cutler has a rule about eating your own dogfood. I don't think they eat enough of it at Microsoft anymore with the lack of build quality in the last few years when they laid off their big QA team.
Some of the stuff I've seen M365 Copilot do is just flat out broken and awful and it's presented in their enterprise (stable) channels.
My personal (probably unlikely) theory is that they plan to lay off a lot of staff anyway, so they wanna get them hooked and reliant on Copilot first so that they're lifelong customers wherever they go next
Eating your own dogfood as they say is a great way to improve a product.
If that were true, they'd just open up copilot agent for general access for any open source project, otherwise their training data will skew toward Microsoft-centric interactions on Microsoft-centric languages and processes
I was going to offer the terraform or ansible organizations as examples, since I'd guess a lot of those bugfixes are rote work, but then realized there's no way Microsoft would assist IBM to ship products
Can't be qemu, given their recent no-slop policy, so maybe the idea is harder than I thought
Liuson told managers that AI "should be part of your holistic reflections on an individual's performance and impact."
Running MSFT like an Asian sweat shop will sink it long term, but perhaps that is the goal.
>Running MSFT like an Asian sweat shop will sink it long term
It's not all bad news then.
Damn AI must be pretty good if they gotta force their employees to use it.
I'm convinced people only use Microsoft because of licensing contracts and familiarity. It's an awful company with subpar products.
What abt azure? is it any good?
I don't really think about azure because I choose not to support Microsoft. Also, as of now I have had very little personal use for running servers, besides things that can easily be hosted almost anywhere.
I'm sure it's convenient for some people with certain projects.
I think Linux is a superior operating system to Windows in almost every way. This is supported by the fact that over 60% of azure servers run Linux [1].
Windows operating system is a walled off garden of intellectual property that wastes billions of dollars every year. Schools, government agencies, hospitals, etc all are spending an incredible amount on proprietary software and this is mainly benefitting a very small amount of people. Yet, thier shitty spyware OS is actually worse then the open source alternative. I think this is a moral failure for humans.
[1] https://azure.microsoft.com/en-us/products/virtual-machines/...
A startup I worked at a few years ago used Azure. In theory it’s _fine_ but I wasted many hours tracking down phantom issues because the Azure admin UI was super buggy.