www.linkedin.com/posts/michae...

open.substack.com/pub/northwoo...

We need to stop waiting for infinite context windows and start building better harnesses.
Give your digital interns a notebook that survives the reboot. 4/

#AI #AgenticWorkflows #Severance #Security #SDD

This is huge for Security.
Imagine a Red Team agent finding a bug. If it doesn't leave a perfect map (External Memory) for the next agent, you waste tokens and miss fixes.
The "Spec" isn't just instructions—it's the Agent's long-term memory. 3/

I read Anthropic's new paper on "Long-Running Agents" and the solution is clear:
It's not just about smarter models (better interns).
It's about better Context (external memory).
You need a "Manager" agent to leave a detailed "To-Do" list for the "Worker" agent. 2/

Now security departments are going to have to figure out how to control AI agents in the hands of every knowledge worker?

Tell me what you think...
(3/3)

I mean, do we want our knowledge workers using AI to create and deploy AI agents? Companies expend a lot of effort to keep these knowledge workers from clicking on phishing emails or sharing sensitive information when they set up their out of office notifications. (2/3)

Security recommendations:
www.linkedin.com/posts/michae...

Broad implications of attack:
www.linkedin.com/posts/michae...

A look at the reliability of the attack information:
www.linkedin.com/posts/michae...

Attack playbook analysis:
www.linkedin.com/posts/michae...

If you would like to jump to one of the posts from here, links below...

What happened:
www.linkedin.com/posts/michae...

👍