Interesting stuff i found - July 2025
Azure Icons
Deep down in a Onenote note I found an old link to a third-party maintained collection of Azure icons (for presentations and documentation)
That has not been updated in a while, but it turns out there is now an official Microsoft collection of Azure icons.
Azure icons to use in architecture diagrams and documentation
What if we could eliminate accounts
I love when people challenge common wisdom. Not sure the suggestions are practical right now, but as someone who is working with CIAM day-to-day, I know that what we do today is not great. Maybe it could be a way forward with some industry support, but for now, I try to passwordless all the tings and stick with the accounts, warts and all.
What if we could eliminate accounts
American politics
The current politics of the US seem extreme - and it is nice to hear significant voices in tech call it out
Instead of gutting health care to ~balance the budget~ deliver tax cuts to the rich, maybe they should make the ultra rich contribute proportionally instead, so they can re-attach to reality?
Entra Sign-in URL Builder
Useful for testing entra apps, and for browsing Graph scopes
Google NotebookLM & OneNote Copilot Notebooks
In the note-taking part of my info pipeline, NotebookLM keeps popping up. It has the ability to take a number of documents (markdown, PDF - up to 50 for free) then allow you to chat about it and generate an audio overview. I plan to try it out for my upcoming Azure certification.
Currently, I use OneNote for note taking, so I am also looking forward to trying out what looks like a copycat feature, Copilot Notebooks
The problem with “Vibe” coding
Dylan Beattie makes the important distinction between “program” (what I sometimes call “me-ware”), something that is whipped up quickly, and works for one specific, hardcoded scenario. And “product”, the program battle-hardened and extended to the point where it can be actually useful for other people, tested, maintainable and have other important quailities.
His experience match my own - LLMs are absolutely fantastic at cranking out me-ware, but struggles as soon as we are talking products.
Logout isn’t a signal - it’s a decision
If I understand the message correctly, the author wants to reverse the traditional logout flow of say OIDC. Here the RP asks the IdP to initiate federated logout and in turn expect it to request each client it knows to be participating in the session to perform local logout. In the proposed flow, as I understand it, instead each RP should expicitly ask for a session to be started, gets a session token, and can then ask for the session status by requesting a revocation list published by the IdP. This sounds a bit weird to me, as it does nothing to solve the “how long should the IdP SSO session live in face of individual client logout” (in Enterprise typically you want everything signed out, consumer login based on Facebook not so much), and I have a hard time imagining risk signals being better at the IdP than at the actual client, unless we also invent a “communicate risk back to the IdP” protocol…
Time will tell if any of this moves towards standardization, I at least agree there is a lot of room to improve the current situation.
ServicesDebugger
Nice little tool to help figure out why something isn’t resolving in ASP.NET
The valley playbook for business software
This quote
This is the valley playbook for business software:
Create a great product by selling it to small teams in the early days, then pivot to the enterprise sales model and “call us” pricing after a while.
The reputation from the early days can then slowly be drawn down, and by the time it’s gone, you’ve locked in a bunch of top-dollar enterprise customers who are never going to switch.
hits the nail on the head. At work I have been at the “small team” end of this a couple of times, and been burned enough to list it as an explicit risk point to consider when choosing new third party dependencies.
On Basecamps “no nonsense” approach
Data Strategy for SREs and Observability Teams
Great stuff, but the resources required to reach the high maturity levels are difficult for SMVs. Right now the focus at work is on deriving value from our current, vendor-specific setup, while transitioning to OpenTelemetry.
Data Strategy for SREs and Observability Teams
Public speaking tips
Hadi Hariri made ChatGPT gather up an old series of public speaking tips to an ePub book. It is currently available for free. I have little experience with public speaking, but semi-frequently do internal presentations, and I guess the same tips apply for smaller audiences too.
Polyfill for .NET
Great idea borrowed from JS (and others), this library polyfills features from newer .NET.
Why Bell Labs worked so well
A critique of the current approach to science R&D funding and management, with a look back at what made (the setup around) Bell Labs great
Your Career Needs a Vision, Not More Goals
A friendly reminder that career management (and life!) involves more than simple goal setting. Also don’t expect your employer to manage your career!
The message aligns well with the systems over goals approach Scott Adams (whose philosophical writings including old Dilbert I like, while I strongly dislike his more recent public politics) advocates.
Your Career Needs a Vision, Not More Goals
Pyright
Python is emerging as an unavoidable language for data science and embedded work, and with uv its dependency management is looking better than ever. I also suspect that LLM agents
have an easier time working with Python than .NET. At least I would like to try it out.
However, I strongly believe in the value of implicit, strong typing. Modern Python appears to support type annotations, and then there is Pyright from Microsoft.
Bitchat
While there has been a lot of writing that the code is not security reviewed, vibe coded etc. etc., even if the implementation is pre-prototype worthy, the concept of a peer-to-peer network not involving the internet, being carried from phone to phone, is fascinating. But I guess it would quickly be abused for crime, leaving it in legal quagmire like hosting a Tor exit node.
Xbow - AI automated pentesting
Maybe their claims are exaggerated, but it makes you wonder: If this capability is close to being commercially available, the current hybrid warfare tactics employeed by bad nation states may be about to get a lot worse…
Xbow has reached the top spot on the US pentester leaderboard