When Anthropic launched Claude Cowork (an AI agent that can execute multi-step workplace tasks) and OpenClaw (formerly ClawdBot) emerged as another agentic AI capable of autonomous coding and task execution, the software market heard a single implication: seat-based SaaS economics might compress. The result was a broad "software derating" move that hit names people actually own: INTU, NOW, ADBE, CRM, DDOG, SNOW and more.
But the interesting question isn’t “will AI kill software?” The interesting question is: what gets disintermediated, what gets re-priced, and what becomes the new choke point?
This post is a framework for that. Not a hot take. Something you can reuse the next time markets panic.
The real shift: “Seats” → “Outcomes” (and the hidden math behind the selloff)
SaaS was built on a deceptively simple machine:
- Sell seats (users)
- Lock contracts
- Expand accounts over time
Agentic AI attacks the pricing unit. If an agent can do the work of 3 people, the CFO doesn’t negotiate seat discounts—she negotiates seat elimination. That’s why this selloff is not purely sentiment. It’s a preview of a renegotiation cycle.
Translation: The threat isn’t that software disappears. The threat is that the profit pool moves from “UI subscription” to “workflow execution.”
What investors miss: “AI feature adoption” ≠ “AI business model transition”
Many incumbents have demo-able AI features. That’s not the bar. The bar is: can they transition revenue and protect margins when AI introduces variable costs (inference + orchestration + data movement)?
In this regime, two things matter more than buzzwords:
- Pricing architecture: can they charge for value delivered (resolved tickets, qualified leads, invoices processed) rather than logins?
- Control plane ownership: do they own the system where work is initiated, authorized, and audited?
A practical map of who wins and who loses (with tickers people care about)
Instead of “software = doomed,” split the universe into four buckets.
1) Likely losers: “workflow wrappers” with weak control
These businesses monetize the act of a human clicking through steps. If agents do the steps, the UI becomes optional.
- High exposure: simple task management / note-taking / lightweight CRM layers where differentiation is thin.
- Tell: the product can be reconstructed by an agent + API access to systems of record.
What to watch: if management responds with “we added an AI assistant,” that’s a feature story. The market is pricing a business-model story.
2) Survivors: suites with distribution + contracts (but they must pivot)
Companies with massive distribution can be late and still survive—but they can’t stay seat-only forever. Think:
- Salesforce (CRM): has the data + workflow gravity in sales. The pivot question is whether “Agentforce”-style products can become outcome-priced without cannibalizing the base too quickly.
- Adobe (ADBE): creative tooling is sticky, but generative workflows shift value from “tool access” to “asset production pipelines.” Adobe’s edge is ecosystem + pro workflows; the risk is commoditized generation.
- ServiceNow (NOW): arguably better positioned than most because it sits in the enterprise workflow + approvals + ITSM control plane. If agents trigger actions, the system of record and audit trail matters.
What to watch: net retention quality (not just the number), seat counts vs usage, and gross margin impact from AI features.
3) Silent winners: data + observability + security (the “agentic plumbing”)
If agents actually do work, enterprise risk increases. More autonomous actions = more need for policy, identity, and monitoring.
- Snowflake (SNOW): if agents query and transform data, governed access + quality + lineage become more valuable. The question is whether SNOW captures that value or if compute economics compress it.
- Datadog (DDOG): “more software doing more things” means more telemetry, failures, cost optimization, and observability. If agent workloads proliferate, monitoring becomes non-optional.
- Security layer (examples): identity, access control, and threat detection become more central when non-humans take actions.
Non-obvious insight: agents don’t reduce software. They reduce humans-as-the-interface. That can increase machine-to-machine activity and therefore increase demand for monitoring/security.
4) Weird crossovers: finance workflow incumbents with “automation target” risk
The sharpest headline reaction often hits companies where agents threaten a very specific workflow. The poster child is:
- Intuit (INTU): if agents can automate tax filing and bookkeeping workflows end-to-end, the unit of value shifts from “software seat” to “return filed / books closed.” INTU has brand + distribution, but the market will pressure the pricing unit.
The decision framework: “Buy / Hold / Avoid” using 5 questions
If you want a repeatable way to decide what to do with a software stock during an AI panic, ask:
- Control plane: does the company own the system where actions are authorized and audited?
- Data gravity: does it sit on proprietary data or does it just display someone else’s data?
- Pricing pivot: can it charge for outcomes (tickets resolved, leads qualified, invoices processed)?
- Margin resilience: do AI features introduce variable costs that break the model?
- Switching costs: are customers locked in by workflows + compliance, or can they swap in an agent layer?
Interpretation:
- BUY candidates tend to be control-plane + data/observability/security owners (and/or incumbents with overwhelming distribution who can pivot).
- AVOID candidates tend to be thin wrappers with low switching costs and seat-only monetization.
- HOLD candidates tend to be incumbents that can pivot but will go through a messy transition (pricing, packaging, margin).
Use AlphaCrew to validate the framework on real tickers
If you want to apply this to the names above (or any other ticker), AlphaCrew gives you programmatic AI analysis pages you can refresh after earnings or major news:
- AI Stock Analysis Hub — search any ticker.
- CRM AI Analysis — suite + distribution case study.
- NOW AI Analysis — workflow control plane case study.
- ADBE AI Analysis — creative workflow disruption case study.
- INTU AI Analysis — finance workflow automation risk case study.
- SNOW AI Analysis — data gravity + governance case study.
- DDOG AI Analysis — observability in an agent-heavy world.
- PLTR AI Analysis — “agent operating layer” beneficiary thesis.
FAQ
Is this selloff rational? Partially. Markets are repricing the unit of value capture (seat → outcome) and demanding proof that incumbents can pivot without margin compression.
Does AI reduce software spending? It can reduce seat counts, but it can increase automation throughput and therefore increase spend on the control plane (workflow), data governance, security, and observability.