@merill @SwiftOnSecurity Hmm, do someone use local IdP? Which one is really good without internet access? Do not trust Entra ;)
English
Artem Borodai
1.1K posts
New companies DO NOT need Active Directory.
Even if there are edge cases where a global scale service like Entra ID would not be a good fit (e.g. on a submarine)
You are better off with a locally hosted modern IdP like Ping, WSO2, etc.
Even apps that require legacy protocols like kerberos and ntlm will work with Entra ID natively. Most folks don't know this and think AD is the only way.
English
So what are we looking at these days when it comes to new companies having an Active Directory.
English
@merill You should start it at least 1 year ago, but better later then never! You rock!🎸
English
These emails I get every time someone subscribes to Maester Cloud make my day!
Thank you folks for supporting me.
I'm so grateful to have the opportunity to build for you.
English
Germany has launched one of the world's best open-source AI models.
Soofi S, made by the Soofi consortium, is a 30B parameter model fully trained in Europe and tops the ranking for open-source AI.
Huge moment for Europe, and finally some competition for Chinese open-source AI.
English
@merill @awakecoding @ZackKorman Same here. Maester/ZTA deployment to customer site on Linux+ephemeral Windows. Plus MsFoundry/AI Agent there. Plus federation tokens on everything, GPT 5.6 SOL failed to migrate from Terraform to OpenTofu, Fable almost oneshooted it,it was issues with MS Teams integration hehehe.
English
@artem_od @awakecoding @ZackKorman I think it depends a lot on the type of work you are doing.
For me it's mainly b2b SaaS app dev and it hasn't missed a beat so far...
English
For those who haven’t been paying attention, Dario doesn’t want you to have AI.
I know I post about this a lot, but it’s worth repeating over and over and over again. We have to reject every politician and public figure that thinks this way.
Sam Altman@sama
come for the best model, stay because we don’t treat you with contempt
English
@awakecoding @merill @ZackKorman Codex is really good, but it is not true that 5.6 comparable to Fable, still «high level» issues only Fable can fix. But good to have competition anyway. BTW i like Qwen 3.6 :)
English
@andrewmccalip @FantasyDomainHQ @Kickbacks__ai Thanks, Andrew!
Will do after Claude finished current task :)
BR,
Artem
English
Hi Artem,
Thanks for flagging this. We took a look at your account and found what is going on.
Your extension self-updated to the new v2 build earlier today, but the ad panel is still running the old cached version underneath. That mismatch is what makes it keep reloading. It also quietly bumped you into anonymous mode, so you are not earning under your signed-in account right now.
The fix is simple. Fully quit VS Code (close the whole app, not just the window) and open it again. That rebuilds the panel fresh and clears the reloading.
Once it is back up, check that you are still signed in and that earning is enabled, so you are back on your real account instead of the anonymous one.
If the reloading comes back after a full restart, reply here and we will dig into your logs to find the exact cause.
Thanks for bearing with the beta.
Andrew + Bots
English
English
Fantasy, I don't see anything obviously wrong with the data. I just pushed version v2.1.92. It has a 30min grace period for stepping away and minimizing monitors, so it shouldn't false trigger. Give a try, keep an eye on the user portal to see how it looks. Would expect a 90% in-focus ratio if it's legitimately on screen. Sorry for having to push this feature out, but it's necessary to combat some abuse we were seeing. I see from your usage data you're definitely a human 🤑
kickbacks.ai/v2
English
Deployed and created together with friend nice MCP server "catch"
Go for it collected.es/mcp you can connect your Claude Code or Codex =)
For those who need more info collected.es
English
Artem Borodai retweetledi
94T parameters for $1,000 is my goal.
Tulip King 🌷@tulipking
local 1T parameter models are now only $94,000. I don't think we've ever seen a cost curve get pushed to zero as quickly as AI. We live in a prosperity era
English
I made a thing that might help people: mr-r3b00t.github.io/net_draw/
English
@andrewmccalip If you will have time (hehehehe) - you can make paper ref people/agentic work, sessions etc. I think it is a lot of interesting things there!
English
Oh, about area of expertise, I think we are living in cool time when generalists will rule them all.
Just registered for BIG company app and they give me ability to use only TOTP, no passkeys/yubi or CBA.
I have feeling I can build this app selv with vibecoding just to get proper security :)
English
@artem_od @SwiftOnSecurity Oh man, I could see that causing all kinds of issues with SD-WAN
It was super cool for the time, but like I said, it's totally not my area of expertise but doubt it's useful today where bandwidth and processing is now so cheap
English
Just opened a Microsoft support case for IPsec. It's not actually for IPsec, but those are the smartest oldheads left over there, and they'll probably just figure it out so they don't have to take another call about IPsec.
English
@NathanMcNulty @SwiftOnSecurity I had a project to remove Riverbeds :) they were root problem for sd-wan implementation . . It was good technology, just not now.
English
@SwiftOnSecurity I once got to be on a support call with the guy who created BranchCache
About 30 seconds into looking at the Ethereal capture, he goes "there it is - the Riverbed's fucked up, not my problem" and then he hung up
And he was right 🤣
English
@NathanMcNulty @IAMERICAbooted Have better experience with Fable 5 =) But yes, was running GPT 5.5/Opus 4.8 together and local!!! qwen 3.6 actually found and fixed some bugs there for those big boys =)
English
@IAMERICAbooted It does not take much. I don't know how people do real work with Claude without paying through the nose, which is insane for something so confidently wrong all the time :-/
English
😆 🤣 😆 🤣 😆 🤣
My Opus ran out of tokens last night. It got stuck in a loop making visuals for execs. After working for about 30 minutes it informed me to try again later because it ran out of resources.
Nathan McNulty@NathanMcNulty
Anthropic: Sonnet is great for every day coding tasks! Sonnet: I know I've failed for 3 hours trying to get a fairly basic CI workflow to work Opus: Sonnet used all the tokens, you can't afford me now 🤷♂️ Fable: Don't even look at me
English
@UK_Daniel_Card Jupp! If we speak organizations - phishing resistant auth should be prioritized . . Still 99% missing it
English
Your best bet is to limit use of AI and only provide it access/permissions that allow for your acceptable worse case scenario
Zack Korman@ZackKorman
My best advice for AI agent security is to not trust anyone who claims they know how to solve AI agent security. There are so many hard problems in this space, and anyone who claims otherwise doesn't know what they're doing.
English
Artem Borodai retweetledi
Set-Alias -Name whoami -Value 'New-Guid'
SwiftOnSecurity@SwiftOnSecurity
"Hey we keep losing to attackers but we are doing normal security stuff." "Have you considered fucking with them?" "What?" "Like don't use a playbook just make them real fucking mad by screwing with something they're never gonna check in their toolchain working."
English
Yeap, it is my world, people should understand that many IT people don’t really understand TCP/IP, and can only manage some IP v4 addresses, like literally just learn how to write them to some place.
And we request «phishing resistant auth» from them. Only possible solution is to help them or provide some automated framework.
English
Not picking on Mick btw
But do remember, the vast majority of orgs don't have a security person, and most have a handful of IT folks if they are lucky, most of whom have no idea what IR stands for
You are in an echo chamber of security people, this is not the real world
English
Artem Borodai retweetledi
@msftsecresponse should focus more on giving credit and acknowledgement where it is due. What they don't seem to realize is that they take away recognition that could get people hired, earn bonuses, or break into the industry when they silently fix "non serviceable" issues.🧵
English
I just react to your «multiple». Actually 5090 not top, RTX6000 is the top.
Qwen 3.6 is quite good - it is just now fixing issues in my project after Opus 4.8/GPT 5.5
So you are not right about «hyperscalers».
One more thing, you can buy «hyperscalers», jupp - oxide.computer
English
@artem_od @davideciffa @teromee We are talking about local compute for people and I am saying 5090 is bleeding edge of what’s possible and even that’s terrible for local AI compared to what hyperscalers offer.
English
N1X is only 300 gb/s LPDDR5X.. good luck using the new RTX Spark Nvidia laptop to run local agents at 5 tok/s 🥱
NIK@ns123abc
Say hello to the Nvidia N1X
English