Great roundup of the year. The shift from "will AI work?" to "how do we architect this?" really captures where the industry is right now. I've been watching teams try to retrofit AI into existing SIEM workflows instead of rethinking the workflow for agents first, and the friction is real. The MCP security risks Willison flagged were spot-on, we ran into prompt injection attempts withn weeks of standing up our first agent. The harder question is balancing guardrails vs agent autonomy without killing usefulness.
Great roundup of the year. The shift from "will AI work?" to "how do we architect this?" really captures where the industry is right now. I've been watching teams try to retrofit AI into existing SIEM workflows instead of rethinking the workflow for agents first, and the friction is real. The MCP security risks Willison flagged were spot-on, we ran into prompt injection attempts withn weeks of standing up our first agent. The harder question is balancing guardrails vs agent autonomy without killing usefulness.
Balancing autonomy is definitely challenging, but possible with good tool and prompt scoping!