Commanding the Agentic AI
Sun Tzu wrote that energy may be likened to the bending of a crossbow; decision, to the releasing of a trigger. In 2026, the trigger is no longer pulled by a human finger. It is pulled by Agentic AI.
We have moved past the era of the "Chatbot" which merely answers a query. We are now in the age of the Autonomous Agent, a system that does not just identify a breach but maneuvers to intercept it. This is the ultimate force multiplier, but as any veteran knows, an army that acts without the General's oversight is an army that will eventually march off a cliff.
The Tactical Advantage: Movement at Machine Speed
The value of an AI agent is found in its Reaction Time. In the 2026 theater of operations, a human response is often a post-mortem. An agent, however, acts in the millisecond between the strike and the impact.
Autonomous Interception: When a node is compromised, the agent does not wait for a ticket. It severs the connection, reroutes the traffic, and isolates the threat before the adversary can even pivot.
The Labor Dividend: Agents handle the "Tier 1 Drudgery." They manage the routine patches and the mundane VLAN shifts. This allows the senior staff to focus on high-level strategy rather than getting bogged down in minor skirmishes.
The Five Pillars of the Perimeter
You cannot grant autonomy without establishing a perimeter. To keep your agents from turning into liabilities, you must enforce the Five Pillars of Governance.
Identity (NHI Management): Every agent must have a distinct "Non-Human Identity." It should never share a generic admin login. If an agent makes a mistake, the General must know exactly which scout to hold accountable.
The Registry: You cannot command what you cannot see. Every agentic workflow must be inventoried and registered. "Shadow AI" agents are unauthorized combatants and must be neutralized.
The Law of Least Privilege: An agent tasked with monitoring logs does not need the keys to the SQL treasury. Access must be scoped strictly to the mission parameters.
Constant Observability: You must watch for "Agent Drift." This is when the machine’s logic begins to deviate from the established tactical baseline.
The Audit Trail: Every action taken by a machine must be recorded in an immutable log. This is your "After Action Report."
Tactical Maneuvers: Tricks for the Front Lines
Beyond the formal pillars, I use a few veteran maneuvers to ensure the machine remains a servant and not a master.
The Burner Agent: For high-stakes reconnaissance on a legacy system, I deploy an ephemeral agent. Once the mission is complete, the agent and its permissions are deleted. It leaves no footprint for an adversary to follow.
Context Grounding: Hallucinations occur when an AI is forced to guess. I ground my agents in a Context Graph—a real-time map of our actual network terrain. It doesn't guess where a cable leads; it retrieves the truth from the heuristic engine.
The Physical Kill Switch: No matter how sophisticated the automation, the human must retain the ability to "unplug" the logic. If a cascading error begins, I want a single button that halts all autonomous movement without killing the hardware.
Doug’s Verdict: The Art of Command
Agentic AI is a magnificent spy and a tireless soldier, but it lacks Consequence Awareness. A machine does not feel the weight of a network outage or the sting of a failed audit. It only follows its objective function.
Command your agents with precision. Use them to win the war of speed. But never forget that the "Art" of IT is in the human intuition. The moment you stop supervising the machine is the moment the machine begins to dictate the terms of your surrender.