The workflow obsolescence risk

The market has been underwriting AI software like a feature cycle. Add agents, raise prices, hold the seat count, watch the multiple expand. That reading is too shallow. The sharper question for any software company in 2026 is not whether a competitor adds AI to its product. It is whether AI removes the workflow the product currently monetises.

Call this the workflow obsolescence risk: the risk that AI collapses or absorbs the workflow a product currently owns, leaving the product on the wrong side of the control point. It is not a competitive risk inside a category. It is a category risk. Diligence run at the feature level will miss it almost every time. Diligence run at the workflow level will catch it.

The market has begun to price this unevenly. Vertical software fell 43% YTD in early 2026, workflow software 39%, while data infrastructure and security categories each fell roughly 21% (Tomasz Tunguz, February 2026). The correlation between forward growth and forward revenue multiple held near 0.51. The translation is direct. Investors are not yet pricing AI replacement risk cleanly. They are pricing a growth signal that is itself a downstream symptom of workflow exposure. The slow growers are slow because the workflow underneath them is already migrating.

What the consensus is mispricing

The consensus underwriting case still treats AI as upside option value. Add agents, monetise them, expand the value metric. That model assumes the surrounding workflow stays intact and the vendor stays in the loop.

Two public signals already break that assumption. Satya Nadella, on the BG2 podcast in December 2024, argued that the business logic embedded in CRMs, ERPs, and project tools will migrate into the AI layer, with Copilot acting as the organising surface across multiple databases (Microsoft, December 2024, via Windows Central). Marc Benioff, defending Salesforce against the same thesis through 2025, conceded the structural point in a different form: software that makes humans do all the work is over, and Agentforce should be measured in Agentic Work Units, not seats (Salesforce, FY26 commentary, via CXToday and Fortune, 2025). Both CEOs are describing the same shift, from different sides of the trade. The workflow, not the seat, is becoming the unit of value capture.

Chegg makes the financial mechanic visible. After ChatGPT’s rollout and Google’s AI Overviews, traffic and subscriptions fell hard enough to force a 23% workforce reduction in June 2024 and $40 to $50 million of cost-out for 2025 (Chegg 8-K shareholder letter, June 2024, SEC filing). The product did not get worse. The workflow it monetised, paying for structured homework help, was absorbed at zero marginal cost by a general-purpose model.

The category did not lose share. The category lost the workflow.

The company-level mechanic

The diligence question is straightforward. For each material revenue stream, name the workflow being sold, then test what an agent leaves behind.

Three exposure patterns recur.

First, the workflow is the product. Customer support tickets, SDR prospecting, contract review, bookkeeping reconciliations, homework help. Klarna automated 67% of customer service chats inside one month of deployment, equivalent to 700 full-time agents, with cost per transaction falling from $0.32 to $0.19 between Q1 2023 and Q1 2025 (Klarna, OpenAI case study, 2024; CX Dive, 2025). The CEO later conceded the cuts went too far on complex cases and rehired (MLQ.ai, 2025). The relevant signal for an investor is not whether Klarna got the staffing right. It is that a single AI deployment compressed a workflow priced at human cost by roughly 40% in two years.

Second, the workflow is being absorbed by a vertical agent. Harvey raised $200 million at an $11 billion valuation in March 2026, with more than 25,000 custom agents running M&A, due diligence, contract drafting, and document review work inside firms including Ashurst, Baker Donelson, Cuatrecasas, and GSK Stockmann (Harvey, March 2026). Bessemer reports vertical AI companies founded post-2019 are reaching 80% of traditional SaaS contract values while growing 400% YoY (Bessemer, 2025, via L40). Harvey is not competing with a SaaS contract review tool at the feature level. It is selling the completed review.

Third, the workflow is being collapsed by a platform agent. Intuit deployed an Accounting AI that reconciles accounts roughly 3x faster against PDF statements, alongside payments, payroll, and bill-pay agents, with over 3 million customers using the toolset since August 2025 (Intuit Investors, 2025; QuickBooks). Gartner forecasts that 40% of enterprise applications will embed task-specific AI agents by end of 2026, up from less than 5% in 2025, and projects agentic AI could account for 30% of enterprise application software revenue by 2035, against 2% in 2025 (Gartner press release, August 2025; January 2026).

Where workflows stay sticky

Workflow obsolescence risk is not uniform. Four conditions still defend a workflow: proprietary transactional data the agent cannot reconstruct, regulated identity or audit trail, distribution into a buyer who will not change vendors easily, and a control point in the system of record where the agent has to ask permission. Atlassian’s January 2026 earnings showed the other side of the same coin: AI generates more code, which generates more code to manage, review, and ship, so the workflow expands rather than collapses (Atlassian Q2 FY26 earnings; Tunguz, February 2026).

The Gartner counter-signal is also worth pricing: over 40% of agentic AI projects are forecast to be cancelled by end of 2027, mostly on cost, governance, and unclear value (Gartner, June 2025). That delays exposure for some categories. It does not remove it.

Upside and downside

Upside case. A software company that owns proprietary workflow data, the system of record, or a regulated control point can absorb agents into its own surface, lift price per workflow, and convert AI from threat to operating leverage. The Salesforce framing of Agentic Work Units, if it holds in renewals, is one version of that.

Downside case. A company whose product is the workflow itself, sold by seat, with no proprietary data exhaust and a buyer who can switch to a vertical agent on a usage contract, faces a step-change in revenue exposure on a 12 to 36 month horizon. Chegg already shows what that looks like in public markets. Klarna shows what it looks like inside a buyer’s P&L.

Watch this, not that

Three diligence questions to put in front of any software investment in the next twelve months.

1. For each material revenue line, what workflow is the customer actually paying for, and what does the workflow look like with an agent in the loop?
2. What proprietary data, identity, or regulatory hook prevents the workflow from migrating to a vertical or platform agent, and how durable is that hook on a three-year view?
3. If the value metric moved from seats to completed work units, would the contract still grow, hold, or compress?

Watch the workflow, not the feature roadmap. Watch the value metric, not the ARR print. Watch the buyer’s cost per transaction, not the vendor’s price per seat.

Sources

1. How Markets Price AI Risk, Tomasz Tunguz / Theory Ventures, 6 February 2026.
2. SaaS Applications “Will Collapse” In The AI Agent Era: Microsoft CEO Satya Nadella, Windows Central, December 2024.
3. Benioff Rejects SaaS-pocalypse Fears as AI Reshapes Enterprise Software, CXToday, 2025.
4. Why Salesforce CEO Marc Benioff doesn’t see a white collar jobs apocalypse despite big gains from AI agents, Fortune, 30 July 2025.
5. Chegg, Inc. Form 8-K, Shareholder Letter, SEC / Chegg, 17 June 2024.
6. Klarna’s AI assistant does the work of 700 full-time agents, OpenAI / Klarna, 2024.
7. Klarna credits AI for slashing customer service costs, CX Dive, 2025.
8. Klarna CEO admits AI job cuts went too far, MLQ.ai, 2025.
9. Harvey Raises Growth Round at $11 Billion Valuation Co-led by GIC and Sequoia, Harvey, 25 March 2026.
10. Legal AI startup Harvey valued at $11 billion in funding round, CNBC, 25 March 2026.
11. Vertical AI SaaS: What Founders Need to Know in 2025, L40 (citing Bessemer Venture Partners), 2025.
12. Intuit Introduces Ground-Breaking Virtual Team of AI Agents, Intuit Investor Relations, 2025.
13. Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Gartner, 26 August 2025.
14. Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, Gartner, 25 June 2025.