The two failure modes around Service as a Software are equally expensive
Most market thesis debates have one correct failure mode and one category error. Service as a Software is unusual because both failure modes are genuinely costly.
Treat the shift as inevitable and you overpay for an early position. The counter-evidence is real: enterprise survey data consistently flags human oversight as dominant in regulated services, hallucinations as the top-cited production risk, and trust as a constraint that takes longer to build than valuations imply. The loudest founders on the timeline have been quietly underweighting this.
Treat it as overblown and you miss the only structural reset business software has had in twenty years. The 1:6 ratio is real. The early traction in narrow workflows is real. YC and Sequoia are not writing checks into the same thesis simultaneously by accident. The services pool dwarfs the software pool by six to one, and that entire pool is being repriced — in slow motion, with very different timelines per vertical.
The businesses that survive will own distribution, domain expertise, workflow integration, risk controls, and trust — not just the model. The businesses that miss it entirely will be competing, by 2028 or 2030, against agents pricing outcomes at software margins in their category.
The symmetric failure modes make this thesis unusually punishing to misread. Overconfidence in the bull case and dismissal of the shift are both expensive. Pick a side. Pick it with eyes open.
Source claim: The two failure modes around Service as a Software — treating it as inevitable versus treating it as overblown — are symmetrically expensive, because the counter-evidence is real and the structural shift is also real.