The short answer

The rebound in software shares after mid-May does not prove that the “SaaS apocalypse” fear has disappeared. It shows something more precise: the market is starting to separate software layers that AI agents may compress from operating-infrastructure layers that AI agents may call more frequently.

The original question was never whether all software disappears. The better question is which software becomes an agent-readable operating layer, and which software remains a seat-based screen that agents can bypass.

What changed after the first article

Many software and cloud names outperformed the broad Nasdaq benchmark over the follow-up window. That is a sign of relief, but not a final verdict. The market moved from indiscriminate fear toward selective differentiation.

Names connected to workflow systems, security, observability, data access, and enterprise permissions were treated differently from software models that depend mainly on user seats and manual screen interaction.

What AI agents threaten

• Repetitive click-based interfaces.
• Seat-based pricing with weak usage linkage.
• Thin workflow wrappers that do not own data, permissions, audit, or execution history.
• Products where the AI agent can complete the job without returning the user to the application surface.

What AI agents may strengthen

• Systems of record and permission layers.
• Security, identity, observability, and compliance infrastructure.
• Workflow platforms that agents must call to complete real business work.
• Usage-based or outcome-linked monetization models that can grow as agent activity rises.

Growth × Liquidity interpretation

Growth: Enterprise AI increases demand for data, workflow, security, and automation infrastructure. That supports selected software companies.

Liquidity and price: A fast rebound can also front-load the good news. Investors should not treat a one-week price recovery as proof that business-model risk has vanished.

Action rule

Keep the split. Core candidates are software companies that agents need to use. Waiting candidates are high-quality companies whose price already assumes smooth AI monetization. Watchlist candidates are products that still need to prove they are infrastructure, not just screens.