Is AI Really Eating Software, Or Is The Market Confusing Threat With Panic?

The loudest debate in the market this year is whether AI agents will hollow out enterprise software. I want to slow down and separate the genuine threat from the panic, because right now the two are being priced as if they are the same thing.

The bear case is not stupid

The fear is concrete. If an AI agent can do the work of five people, a company buys one seat instead of five, and the per seat licensing model that funds most SaaS starts to leak. That is a real structural risk, not a meme. Salesforce and Workday are down hard this year for exactly this reason, and the market is right to demand answers.

But moats are not made of seat counts

Here is what the doomsday version misses. The value in enterprise software was never just the login. It is the data, the workflows, the integrations, the compliance, and the cost of ripping all of that out. A company does not replace a decade of finance and HR plumbing because a demo looked good. The firms with deep data and distribution, like $Microsoft(MSFT.US), are the ones most likely to sell the AI agents rather than be killed by them.

How I separate winners from losers

I am not buying the sector, I am sorting it. Platforms that own proprietary data and sit inside critical workflows can attach AI and charge more. Thin point solutions with no data moat and easily copied features are the ones genuinely at risk. The selloff has been indiscriminate, and that is usually where patient investors find mispriced quality.

What I am actually doing

I am adding slowly to the highest quality names on weakness and avoiding the cheap looking point solutions that screen as bargains. The rotation into chips is real and I understand it, but a great business at a fair price during a panic is a setup I have learned not to ignore. Software or chips is a false choice. The honest answer is that some software dies and some software ends up eating AI for lunch.