The point about "midnight drift" stood out to me because most discussions around AI focus on productivity gains, while the security implications of unattended autonomy are often overlooked. I also appreciated how you framed AI agents as identities that require continuous oversight rather than one-time authentication.
Thank you. I think that's exactly the shift many organizations are underestimating. Authentication answers who the entity is, but not what it's doing once it starts operating autonomously.
the way you break down complex concepts like non human identities and prompt injection make it so easy to understand without losing depth. Overall, very insightful and well structured, def looking forward to more of your work on this topic!!!
Thank you, that means a lot! These topics can become incredibly technical very quickly, so I try to focus on the underlying ideas rather than the jargon
That's an amazing read..The idea of Non-Human Identities running 24/7 and breaking the “quiet night window” feels like a real shift most people are still underestimating.
This gave me a lot to think about. I really enjoyed your perspective and the way you broke down such a complex topic.Really interesting perspective and a great read. Looking forward to reading more of your work! 👏✨
Great write. This is indeed a good piece! You tried to make sure that normal user like me fully understood and stay aware of the AI development, thanks for that snippet of prompt though 😂
I wasn't sure what to expect from this but I love that your take is balanced and erring on the side of security over the perceived convenience of AI. I'd been doing a ton of research on implementations and the come away was for that in general we're not accounting for issues we already solved that get circumvented by AI and that we're not accounting for the new issues that AI presents. I'm glad more people are talking about this.
Thank you! That's what keeps standing out to me as well. A lot of AI discussions focus on what the technology can do, but not enough attention goes to the security assumptions it quietly bypasses or the entirely new risks it introduces. It's encouraging to see more people looking at AI through that lens.
I liked your definition of Agentic AI: an entity that cannot validate its own sanity i.e it lacks meta-cognition and hence cannot reason about what it's actually doing beside what's been fed into it
I'm glad it's being talked about it this way since nowadays almost everyone thinks of it as silver bullet solution that can magically solve all of the problems with pure net positive outcome
I appreciate that. The lack of meta-cognition is exactly what concerns me. These systems can execute complex workflows, but they can't step back and ask "does this still make sense?" That's why I think the conversation needs to shift from capability alone to capability plus oversight.
This is a topic which is not getting enough attention in my view. We are / have moved into a continuous operating model and if you don’t have the architecture, processes and governance in place you are sleep walking into problems whose scale will not be evident.
As an aside we’ve had similar discussions on unknown blast radius issues with things like Infrastructure as Code where one error can be compounded exponentially.
But I’d like to throw in a side curve ball - there’s a reason why humans need sleep, that downtime to process the days events, to have the subconscious working on a problem, to recover from the days events. AI doesn’t have that so we risk losing something in the human workforce psyche - that 1am eureka moment, the spark of an idea, that time you wake up knowing what’s wrong and what needs doing.
If everything is the same where does the innovation come from?
That's a fascinating perspective. I completely agree on the continuous operating model point that many organizations are inheriting 24/7 execution without fully appreciating the implications of 24/7 risk. And your point about sleep is interesting because it highlights a difference we don't talk about enough: humans pause, reflect, and occasionally rethink assumptions.
It doesn’t get paid so it’s not under employ
Fair point
The point about "midnight drift" stood out to me because most discussions around AI focus on productivity gains, while the security implications of unattended autonomy are often overlooked. I also appreciated how you framed AI agents as identities that require continuous oversight rather than one-time authentication.
Thank you. I think that's exactly the shift many organizations are underestimating. Authentication answers who the entity is, but not what it's doing once it starts operating autonomously.
the way you break down complex concepts like non human identities and prompt injection make it so easy to understand without losing depth. Overall, very insightful and well structured, def looking forward to more of your work on this topic!!!
Thank you, that means a lot! These topics can become incredibly technical very quickly, so I try to focus on the underlying ideas rather than the jargon
That's an amazing read..The idea of Non-Human Identities running 24/7 and breaking the “quiet night window” feels like a real shift most people are still underestimating.
I'm glad you found it informational!
i dont sleep
We're slowly turning into AI, or is AI turning into us?
This gave me a lot to think about. I really enjoyed your perspective and the way you broke down such a complex topic.Really interesting perspective and a great read. Looking forward to reading more of your work! 👏✨
Thank you, means a lot to me!
Great write. This is indeed a good piece! You tried to make sure that normal user like me fully understood and stay aware of the AI development, thanks for that snippet of prompt though 😂
Haha I'm glad you found it useful!
I wasn't sure what to expect from this but I love that your take is balanced and erring on the side of security over the perceived convenience of AI. I'd been doing a ton of research on implementations and the come away was for that in general we're not accounting for issues we already solved that get circumvented by AI and that we're not accounting for the new issues that AI presents. I'm glad more people are talking about this.
Thank you! That's what keeps standing out to me as well. A lot of AI discussions focus on what the technology can do, but not enough attention goes to the security assumptions it quietly bypasses or the entirely new risks it introduces. It's encouraging to see more people looking at AI through that lens.
I liked your definition of Agentic AI: an entity that cannot validate its own sanity i.e it lacks meta-cognition and hence cannot reason about what it's actually doing beside what's been fed into it
I'm glad it's being talked about it this way since nowadays almost everyone thinks of it as silver bullet solution that can magically solve all of the problems with pure net positive outcome
I appreciate that. The lack of meta-cognition is exactly what concerns me. These systems can execute complex workflows, but they can't step back and ask "does this still make sense?" That's why I think the conversation needs to shift from capability alone to capability plus oversight.
This is a topic which is not getting enough attention in my view. We are / have moved into a continuous operating model and if you don’t have the architecture, processes and governance in place you are sleep walking into problems whose scale will not be evident.
As an aside we’ve had similar discussions on unknown blast radius issues with things like Infrastructure as Code where one error can be compounded exponentially.
But I’d like to throw in a side curve ball - there’s a reason why humans need sleep, that downtime to process the days events, to have the subconscious working on a problem, to recover from the days events. AI doesn’t have that so we risk losing something in the human workforce psyche - that 1am eureka moment, the spark of an idea, that time you wake up knowing what’s wrong and what needs doing.
If everything is the same where does the innovation come from?
That's a fascinating perspective. I completely agree on the continuous operating model point that many organizations are inheriting 24/7 execution without fully appreciating the implications of 24/7 risk. And your point about sleep is interesting because it highlights a difference we don't talk about enough: humans pause, reflect, and occasionally rethink assumptions.