Part 6 of Defense at Machine Speed: The Operating Model. The assembly of everything before it.
The defense organization that matches an agentic threat runs as a loop, not a calendar. Five stations, never stopping:
Hunt. Continuous scanning of the whole estate with the best available models, on the attacker's cadence. New disclosures become scan targets the day they drop.
Verify. Every finding earns its queue place by surviving an attempted exploit. The queue sorts by demonstrated exploitability, not severity scores. Nobody fixes a rumor.
Patch. Remediation as assisted throughput: AI drafts the fix with the exploit test in context, engineers spend their judgment reviewing diffs instead of reproducing scanner output cold.
Prove. Every fix is re-attacked, by a function independent of whoever wrote it. The exploit must fail before the ticket closes. Fix survival under re-attack is a headline number.
Deceive and tax. Decoy endpoints and credentials charge the attacker tokens, time, and exposure; short-lived, narrow identity removes the cheap chaining paths.
Two structures hold it up. A range, where the real offensive harnesses run against replicas of your own systems under your control, which is what DAST has to mean now. And outcome measurement: attack success rate, detection and containment time, fix survival under re-attack, cost per resolved incident. Programs that report activity get funded to produce activity. Programs that report outcomes get funded to produce outcomes.
The team gets smaller at the top of the funnel and stronger at the points of judgment. The loop's labor is machine labor. Humans decide which risks to accept openly, which fixes are good enough, and how the next disclosure reorders the queue.
Nothing here requires a breakthrough. The models are available, the harnesses are public, the receipts are in the open. The organizations that come through this transition will be the ones that stopped taking baby steps. Full piece on Substack.