a builder's codex
codex · release log · 2026-07-21

AI agents collapse the seat-based SaaS expansion model, splitting software into infrastructure that AI feeds and applications that AI feeds on

2026-07-21 · +1 insight

One new card today. A signal that has been building across B2B software earnings reports came into sharp focus with Jason Lemkin's July 20 piece at SaaStr. The theme: what happens to seat-based SaaS when AI agents do multi-seat work without needing multi-seat licenses.

Theme 1, The seat-expansion model and how AI agents broke it

Jason Lemkin published a framework for reading the current B2B software market that names something the earnings reports have been circling without stating (AI agents eliminate the seat-expansion tailwind, sending sub-5% SaaS companies into terminal structural decline). Seat-based SaaS annuities compounded because customers expanded seat counts as their teams grew. One AI agent doing the work of ten support reps means the customer never buys those ten seats. Expansion stops being a tailwind and becomes a headwind.

Lemkin bifurcates the software market with a single diagnostic statement: "There is no longer one category called software. There is infrastructure that AI feeds, and there are applications that AI feeds on." Infrastructure companies on usage-based billing scale with AI workloads. Seat-based application vendors do not. Sub-5% annual growth is no longer slow growth. He calls it a terminal state: "Not slow growth. The end of growth, with a question mark over whether what's left is even safe."

The evidence is public market data. Dropbox guidance projects full-year revenue decline of 0.4% to 0.9%. PagerDuty net dollar retention has fallen to 97%. Zoom enterprise net dollar expansion sits at 99%. Software now trades at 22.7x forward earnings, below the S&P 500 for the first time ever. Roughly $450 billion in 2026 hyperscaler infrastructure spend targets AI. Each dollar there does not renew a seat license.

Open the full release log →