Self-disruption follows the value chain downward — software companies must eat their own agent layer before someone else does

Intercom’s trajectory — software company → agent company → AI model company — is a deliberate three-stage self-disruption. McCabe frames it explicitly: “as we did when we started to disrupt our software business with our agent business, we will now begin the processes of disrupting our agent business with our AI business.” The thesis is that value accrues to the model layer, so the winning move is to cannibalize your own higher layers before competitors do.

This connects to Sell the work, not the tool — model improvements compound for services, against software — Intercom is moving from selling the tool (software), to selling the work (agents resolving issues), to selling the engine that does the work (models via API at $250k/yr contracts). Each layer down captures more durable value. It also validates LLMs selectively destroy vertical software moats — 5 fall, 5 hold: Intercom isn’t waiting for LLMs to destroy their feature moats — they’re proactively shifting to where Frontier companies absorb every useful agentic pattern into their products can’t reach, because the model itself is the moat. The same proactive instinct applies inside the org, not just the product line: The gains come from redesigning work around AI, not bolting AI onto human workflows — redesign the work around the new material before the old shape is competed away, rather than bolting agents onto workflows built for humans. Jaya Gupta complicates the “downward” direction, though: value doesn’t only flow toward the model, it can relocate sideways to the data platform, the device, or wherever inference is cheapest — Where inference runs decides who captures margin, owns the context, and earns trust makes deployment location itself a place value can accrue.