Variations in Tactics: When to Trust the Machine (and When to Grab a Console Cable)
In the fourth century BC, Sun Tzu wrote about "Variations in Tactics." He argued that there are roads which must not be followed and towns which must not be besieged. In 2026, I am applying that same logic to my AI agents. Just because the bot suggests a configuration change at 2:00 PM on a Tuesday does not mean I am going to let it touch the core switch without a human looking over its shoulder.
The "Art" of modern IT is knowing exactly where the machine’s intelligence ends and its "hallucinations" begin. We are living in the era of the Synthetic Admin, but if you do not maintain a "Human-in-the-Loop" philosophy, you are going to find yourself besieged by a broadcast storm of your own making.
The Reality of the Digital Hallucination
We have all seen LLMs confidently lie about historical dates or movie cast lists. In a chat window, it is funny. In a global routing table, it is a catastrophe.
AI models in 2026 are great at identifying patterns, but they lack "Physical Intuition." An AI might see a spike in latency and suggest rerouting traffic through a backup link that it thinks is available. What it doesn't know—because it can't feel the humidity in the room—is that the backup link is currently hanging by a thread on a patch panel that was accidentally kicked by the night janitor.
If you blindly trust the "Automated Optimization" button, you are following a road that leads to a cliff.
The Three-Step Verification Maneuver
When the machine suggests a "Variation in Tactics," I follow a strict protocol before I hit Commit.
The Logic Check: Does the suggested change actually address the root cause, or is the AI just treating a symptom? If the AI wants to increase a timeout value to fix a "slow link," it might be ignoring the fact that the link is slow because of a physical CRC error.
The Blast Radius Assessment: If this change fails, how many people stop working? If the answer is "the whole building," the machine stays in the passenger seat.
The Console Cable Reality: I always ask myself: If I lose remote access because of this change, am I prepared to drive to the data center with a laptop and a serial cable? If the answer is no, I do not let the AI push the button.
Comparison: Machine Logic vs. Human Intuition
The machine is your scout; you are the General. Here is how to divide the labor.
| Tactical Scenario | Trust the Machine? | The Human's Role (The "Art") |
|---|---|---|
| Log Aggregation | YES: Full Trust | Scans millions of lines instantly. The human simply identifies which flagged "anomalies" are actually business-critical. |
| Root-Cause Heuristics | YES: High Reliability | Tools like pathSolutions find physical faults (CRC errors, bad SFPs). The human physically replaces the hardware. |
| Core Routing Changes | NO: High Risk | AI hallucinations in BGP or OSPF can isolate entire regions. Review the "suggested" config before deployment. |
| VLAN Provisioning | PARTIAL: Human-in-Loop | Good for routine tasks, but verify that "Auto-Provisioning" didn't bypass security silos or core ACLs. |
The Digital Friendly Fire
I have seen an AI agent "optimize" a network by shutting down redundant ports to save power, only to realize (too late) that it just broke a high-availability cluster. That is "Digital Friendly Fire."
The goal for 2026 isn't to be a Luddite who refuses to use the new tools. It is to be the strategist who knows that the "Spirit of the Staff" is what wins the war. Use the AI to gather the "Foreknowledge," use it to run the diagnostics, and use it to suggest the fix.
But when it comes time to alter the fabric of the kingdom? Keep your hand on the hilt of the sword. Or, in our case, keep your console cable in your bag.