Larry Osterman
@larryosterman.github.io
Long term MSFT employee, Dad, Singer.
Expect cat pictures, tech nerding, etc.
Previous: SMB filesystem, Exchange Store, Windows Audio, Windows Runtime, COM, Confidential computing.
Current: Azure SDK for C++
Proud LGBTQIA+ ally.
Profile pic from Not
Expect cat pictures, tech nerding, etc.
Previous: SMB filesystem, Exchange Store, Windows Audio, Windows Runtime, COM, Confidential computing.
Current: Azure SDK for C++
Proud LGBTQIA+ ally.
Profile pic from Not
There's also the bit where the compiler maintainers say they are understaffed and underfunded, and the contributor responds saying (effectively) "I could maintain it, but I don't have the time or money to do it".
November 27, 2025 at 7:41 PM
There's also the bit where the compiler maintainers say they are understaffed and underfunded, and the contributor responds saying (effectively) "I could maintain it, but I don't have the time or money to do it".
OMG. I went into that PR with no context, but as a long time developer working in corporate OSS, who uses AI generated code daily, and all I can say is "Wow, what a shit show".
And I totally concur with everything Kathryn has written about this PR.
I especially liked "AI can understand it".
And I totally concur with everything Kathryn has written about this PR.
I especially liked "AI can understand it".
November 27, 2025 at 7:30 PM
OMG. I went into that PR with no context, but as a long time developer working in corporate OSS, who uses AI generated code daily, and all I can say is "Wow, what a shit show".
And I totally concur with everything Kathryn has written about this PR.
I especially liked "AI can understand it".
And I totally concur with everything Kathryn has written about this PR.
I especially liked "AI can understand it".
Argh, posted that and had to delete it for redundancy.
November 25, 2025 at 4:49 PM
Argh, posted that and had to delete it for redundancy.
Agents are 1 for 6 successfully following the prompt. I suspect humans have a somewhat higher success rate (fundamentally the task is "ensure that someone has modified a file in another repo, then run this command line").
That task is too complicated for most llms.
But they do get creative trying.
That task is too complicated for most llms.
But they do get creative trying.
November 25, 2025 at 7:36 AM
Agents are 1 for 6 successfully following the prompt. I suspect humans have a somewhat higher success rate (fundamentally the task is "ensure that someone has modified a file in another repo, then run this command line").
That task is too complicated for most llms.
But they do get creative trying.
That task is too complicated for most llms.
But they do get creative trying.
Welcome to the weird and wonderful world of non-deterministic programming.
Last week I got rid of a copilot prompt file because I realized that a similar length set of text instructions would be more helpful for partner teams - the AI was simply not reliable enough to be trusted.
Last week I got rid of a copilot prompt file because I realized that a similar length set of text instructions would be more helpful for partner teams - the AI was simply not reliable enough to be trusted.
November 25, 2025 at 7:08 AM
Welcome to the weird and wonderful world of non-deterministic programming.
Last week I got rid of a copilot prompt file because I realized that a similar length set of text instructions would be more helpful for partner teams - the AI was simply not reliable enough to be trusted.
Last week I got rid of a copilot prompt file because I realized that a similar length set of text instructions would be more helpful for partner teams - the AI was simply not reliable enough to be trusted.
OK, figured out how to get the old experience back, but YUCK.
November 21, 2025 at 6:42 PM
OK, figured out how to get the old experience back, but YUCK.
I lived in Pittsburgh from 1980-1984 - you couldn't leave a window open in the summer, because a black soot quickly covered every horizontal surface.
And this was *after* most of the steel mills had closed. I don't want to think about how bad it was when they were all running full out.
And this was *after* most of the steel mills had closed. I don't want to think about how bad it was when they were all running full out.
November 19, 2025 at 4:28 PM
I lived in Pittsburgh from 1980-1984 - you couldn't leave a window open in the summer, because a black soot quickly covered every horizontal surface.
And this was *after* most of the steel mills had closed. I don't want to think about how bad it was when they were all running full out.
And this was *after* most of the steel mills had closed. I don't want to think about how bad it was when they were all running full out.
Wow, congrats Leo!
November 16, 2025 at 5:40 PM
Wow, congrats Leo!
And it is infinitly cheaper to fix security bugs BEFORE you ship rather than after"
November 14, 2025 at 3:09 AM
And it is infinitly cheaper to fix security bugs BEFORE you ship rather than after"
I've never been so happy to have built a threat model before, it made all the effort worthwhile.
And it ratified the answer to the "Why are we threat modeling" question the dev lead for thst component asked me: "we threat model because it makes us think about the system like an attacker...
And it ratified the answer to the "Why are we threat modeling" question the dev lead for thst component asked me: "we threat model because it makes us think about the system like an attacker...
November 14, 2025 at 3:09 AM
I've never been so happy to have built a threat model before, it made all the effort worthwhile.
And it ratified the answer to the "Why are we threat modeling" question the dev lead for thst component asked me: "we threat model because it makes us think about the system like an attacker...
And it ratified the answer to the "Why are we threat modeling" question the dev lead for thst component asked me: "we threat model because it makes us think about the system like an attacker...
Usually I can basically figure out at least how he got to his tangent (glass wires? Magnets?) but on this one? I got nothin'
November 13, 2025 at 11:47 PM
Usually I can basically figure out at least how he got to his tangent (glass wires? Magnets?) but on this one? I got nothin'
And oh, 6.5% interest rate on a 30 year fixed. Not quite flashing back to the early 80s but still sticker shock.
November 12, 2025 at 4:00 AM
And oh, 6.5% interest rate on a 30 year fixed. Not quite flashing back to the early 80s but still sticker shock.