The Ship of Theseus is a thought experiment from ancient Greece. If you replace every plank in a ship, one at a time, is it still the same ship? At what point does it become a different ship?

Customer support is going through the same transformation. You automate password resets. Then order status queries. Then billing questions. Then refund processing. Then basic troubleshooting. Each category, replaced one at a time with AI.

At what point does it stop being "support" and become "a system that processes requests"?

The Gradual Replacement

Year 1: You automate the 5 simplest ticket categories. 20% of volume. Customers barely notice. The remaining 80% still goes to humans.

Year 2: You automate another 15 categories. 50% of volume. Some customers start noticing. "I used to talk to a person. Now I get automated responses." But the responses are accurate and fast. Most people are fine with it.

Year 3: You automate medium-complexity categories. 70% of volume. The human agents now handle only the complex, emotional, and unusual tickets. The average ticket that reaches a human is harder than before. Agent skill requirements go up. Agent pay should go up too (but often doesn't).

Year 4: You automate complaint handling with AI that generates empathetic responses. 85% of volume. A customer who's angry about a billing error gets an AI response that acknowledges their frustration, processes the refund, and follows up. It's good. It's fast. It sounds human.

Is that still "support"?

What Gets Lost

The human connection. A customer who just lost their job and needs to cancel their subscription. The AI processes the cancellation. The human agent would have said "I'm sorry to hear that. I've cancelled your subscription, and I hope things look up soon. If you want to come back later, just let us know." The extra sentence costs nothing and means everything.

The judgment calls. A customer's situation doesn't fit any category. Their request is technically against policy but clearly deserved. A human agent makes a judgment call: "This person deserves an exception." The AI follows the rules. Rules don't cover every situation.

The feedback loop. When a human agent sees the same bug reported 20 times, they message engineering in Slack: "This is getting bad, 20 tickets today." When AI handles those tickets automatically, the volume is invisible. The bug reports are resolved but the pattern isn't escalated. Nobody knows it's happening.

The relationship. Some customers have a rapport with their support agent. They ask about the agent's weekend. They send a thank-you note after a good interaction. They feel a connection to a person at the company. AI doesn't create this. When the last human agent is replaced, the last thread of human connection goes with them.

What Gets Gained

Speed. AI responds in seconds, 24/7. No queues, no hold times, no "our agents are busy, estimated wait: 12 minutes." The customer gets help immediately, every time.

Consistency. AI gives the same answer every time. No variation based on which agent picks up the ticket. No Monday morning grogginess affecting response quality. No bad day from the agent spilling over into the customer interaction.

Cost. AI at $0.20 to $0.30 per interaction vs humans at $5 to $15. The economics are dramatic. A company that replaces 80% of human support volume saves hundreds of thousands per year.

Scale. AI handles 10 tickets or 10,000 with equal quality. Growth doesn't require proportional hiring. The cost curve stays flat while revenue grows.

Where the Line Should Be

The Ship of Theseus has no definitive answer. But customer support does: the line should be drawn where the customer's situation requires empathy, judgment, or relationship.

Automate the information retrieval. "What are your hours?" doesn't need a human. "How much does shipping cost?" doesn't need a human. "Where's my order?" doesn't need a human.

Automate the rule-based actions. Refunds within policy, account updates, subscription changes. These follow clear rules. AI applies them reliably.

Keep humans for the messy stuff. Complaints, grief, confusion, anger, VIPs, edge cases, ethical dilemmas, and anything where the customer's emotional state matters as much as their stated request.

Supp draws this line through classification. Each of the 315 intents is either fully automatable, AI-assisted (AI gathers context, human resolves), or human-only (AI routes, human handles entirely). The classification determines which mode. The customer gets the fastest possible resolution for simple requests and a human for everything else.

The Real Question

The Ship of Theseus asks whether identity persists through replacement. In support, the question is: does the customer care?

If the AI resolves their issue in 10 seconds with the right answer, do they care that a human didn't write it? For password resets, no. For a billing dispute, probably not. For a message about their dead spouse's account, absolutely.

The answer varies by interaction. And that variance is why the right system classifies first and routes accordingly, instead of applying one approach to everything.

A fully AI-driven support system is technically possible. It would be fast, cheap, and consistent. It would also be a system that processes requests, not a system that supports customers. The distinction matters to the people writing in at 3 AM because their business is down, or because their partner died, or because they're scared and confused and need a human to tell them it'll be okay.

Those moments are why support exists. Everything else is just information retrieval.