AI Disclosure Tracker Part II

Recap of Part I

In Part I we examined how 61 US-listed software-majority public companies are communicating about AI three years after ChatGPT and what their revenue trajectories show. The shorthand of the data: disclosure is bimodal, the public cohort has not converged on a definition of AI revenue, hard-dollar disclosure is broadening but still concentrated in two companies at the application layer (Salesforce and Adobe), and absolute revenue dollars added per year are at all-time highs even as percentage growth has returned to pre-COVID baselines. This Part II turns from observation to implication.

‍

AI labor displacement is not yet visible in cohort employment data

"AI models for generating computer code have become so efficient that we have been restructuring our product development teams into smaller, more agile and productive groups. This new AI Code Generation technology is enabling us to build more software in less time with fewer people. Oracle is now building more SaaS applications for more industries at a lower cost." - Larry Ellison/ORCL

"AI has always been foundational to how we operate. AI flattens our hiring curve, and helps us innovate from idea to product faster. It streamlines go-to-market, improves customer outcomes, and drives efficiencies across both the front and back office. AI is a force multiplier throughout the business." - George Kurtz/CRWD

"The AI productivity gains we are seeing inside our own organization are demonstrating that we can grow revenue without growing headcount in lockstep, a dynamic we believe will be a meaningful driver of operating leverage over time." - Eliran Glazer/MNDY

‍

Sixteen percent of the cohort (10 of 61 companies) have disclosed staffing actions since January 2023 explicitly tied to AI investment or AI-driven efficiencies, totaling roughly 26,000 positions, about 2.2% of cohort employment excluding Amazon's retail workforce. Yet the cohort's headcount grew by about 37,000 net positions (+3.25%) from FY24 to FY25 ex-Amazon. The AI-driven cuts may be real, with the counter-argument that companies are just reacting to ZIRP-era indigestion (its not AI driven, they just over hired in years past), and are concentrated at specific companies, but they are currently offset by ongoing hiring elsewhere. The "SaaS apocalypse via AI labor displacement" thesis is not yet evident in aggregate employment data. The net position growth is well under the revenue growth but that is to be expected.

What has changed in CQ1 2026 is the commentary. The three quotes above span a systems-of-record incumbent (Oracle), a security infrastructure leader (CrowdStrike), and a horizontal SaaS leader (monday.com). The framing is the same in all three: AI is letting these companies grow revenue without growing headcount in lockstep. Cohort employment data has not caught up yet. Worth tracking through the next four quarters whether the chorus of commentary translates into a noticeable bend in the cohort's hiring curve.

For more historical analytics on employment growth amongst top software companies spanning ten years, see our Cover Drive Insights on productivity growth.

Five lessons for private software operators

1. Pick a pricing posture and own it

Five distinct AI pricing postures are visible in the cohort right now. Atlassian bundled Rovo into existing Cloud tiers and reported about 5 million monthly active Rovo users with no separate AI revenue line; the bet is volume of adoption and incremental tier upgrade rather than a discrete AI revenue stream. ServiceNow kept Now Assist as a separate Plus SKU with a $1 billion ACV target, monetizing AI as premium pricing on top of seats. Adobe split the disclosure cleanly into AI-first ARR (incremental new SKUs) and AI-Influenced ARR (existing revenue with AI features added), letting the market choose what to value. Salesforce charges Agentforce as a discrete SKU with an agentic work unit consumption metric (over 2.4 billion agentic work units delivered to date). monday.com launched a seats-plus-credits model in CQ1 2026 that combines fixed seat licenses with variable AI consumption pricing on top.

Each posture is defensible. None is right or wrong on its face. The trap, observed across multiple cohort companies, is undisclosed bundling: launching premium and quietly bundling, or launching bundled and claiming pricing power. Pick the posture, communicate it explicitly, and stay with it until you have enough feedback data. Customers and capital allocators are paying close attention to whether the posture and the financial reality on ROI match.

2. Moat construction: AI amplifies, it does not create

The categories accelerating fastest in the cohort are the ones that already had defensible moats. Hyperscalers have capital and scale moats on top of cloud capacity. EDA companies (Synopsys, Cadence) sell into the AI silicon boom but their moat is decades of deeply embedded chip design and verification workflow with the world's chip designers; AI is amplifying that, not creating it. Vertical SaaS leaders have proprietary data capture: CCC's 14 million claims, Veeva's life-sciences workflow, Autodesk's design data, Blackbaud's social impact dataset, Guidewire's insurance workflow data. AI is the layer that compounds these assets if executed well.

The horizontal SaaS cohort that lacks structural moats is shipping AI products at velocity but has not seen revenue lift above pre-COVID growth rates with rare exceptions. For private operators the lesson is clear. Audit the moat before building AI features. Proprietary data, regulatory workflow, industry expertise, customer integrations and switching costs are the real assets. Renting frontier model access without proprietary data is renting distribution from the model providers, and the operating economics get worse over time as more entrants compete on the same models. The companies in the cohort with the most AI runway ahead are vertical SaaS leaders with both deep data and embedded workflow, and infrastructure vendors who make the agentic layer secure, controllable, and measurable. The cohort's disclosure tier today materially understates where the underlying AI opportunity sits in vertical and infrastructure software.

3. AI as an internal operating leverage source

The 10x TAM expansion is about redirecting customer labor budgets that today fund human work into software P&L. The companies that capture this redirection earliest will own the next decade of software revenue growth. At the same time, companies should look inward at their own operations.

For private operators with capital efficiency mandates, this is actionable now. Engineering productivity through AI code generation, customer support automation through agentic deflection (Q2 Holdings disclosed in CQ2 2025 over 50% reduction in account takeover fraud and material customer-support deflection via AI), finance and operations workflow automation, and go-to-market efficiency through internal copilots, top of funnel agents and analytics agents for revenue operations.

While the belief universally is that there ought to be productivity enhancement, measuring it is nontrivial. Landing on new KPIs and communicating them becomes important.  

4. Internal disclosure discipline replaces public disclosure muscle

The public cohort took about twelve quarters to move from silence through qualitative commentary through attached cohort metrics to driven hard-dollar AI disclosure. Each step required defining the metric, aligning auditors, briefing the sell side and picking a cadence. Private companies do not have a sell side or the SEC-grade auditor process the public cohort goes through. But the discipline of defining AI revenue, AI-attached revenue, AI consumption ARR, and AI productivity inside the business is its own competitive advantage. It builds management reporting muscle. It forces strategic clarity about what AI is actually doing for the business. It lets the board make capital allocation decisions on real data rather than on narrative.

The freedom from quarterly public disclosure means private operators can pick the metric that fits their business. They are not constrained by the cohort's failure to converge on an industry standard (the cohort is using at least nine distinct framings of AI revenue today, with no convergence in sight). Pick what you can measure, define it precisely in board materials, hold to it consistently util feedback data emerges, and tell employees and customers the story the data is telling. That clarity is what compounds into competitive advantage.

5. Architect billing for seats and consumption from day one

The consumption layer is the fastest-emerging attach form in the cohort. Adobe generative credits have been live for multiple quarters and contribute to the consumption side of its AI-Influenced ARR. Box AI Units are sold as a discrete consumption SKU on top of Box subscriptions. ServiceNow charges Now Assist consumption through its Pro Plus and Enterprise Plus tiers. monday.com just launched seats-plus-credits in CQ1 2026, an explicit pivot from pure seat pricing. The math works differently than pure seat-based subscription pricing because AI features have variable per-customer compute load that does not correlate neatly with seat count.

For private operators, the implication is to architect billing systems for both fixed (seats) and variable (consumption) revenue from the start, not retrofit later. Finance teams need new forecasting models because consumption forecasts differently than seats and tends to be more volatile quarter to quarter. Sales and customer success teams need updated playbooks because the customer conversation moves from license count to value delivered per unit of consumption. The cohort companies that have done this transition successfully have generally found that customers prefer the optionality (variable consumption above a fixed base) and that consumption ARR can carry higher gross margin and lower customer acquisition cost than pure seat ARR over time.

What we are watching through 2026 and 2027

Six items on our watchlist for the next four quarters.

1. When does hard-dollar AI revenue disclosure broaden beyond Salesforce and Adobe?

About 85% of disclosed application AI ARR sits at two companies today on a base of roughly $3 billion. We will be watching several companies including HubSpot, Klaviyo, Datadog and Atlassian, all of which have AI agent SKUs in market without any hard-dollar AI revenue line yet. The market will look for more hard disclosure, and assess whether the AI revenue is really Net New.

2. Does the labor commentary translate into aggregate headcount decline?

There is increasing commentary on this, which is leading actual data. Cohort headcount still grew +3.25% from FY24 to FY25 ex-Amazon. We will be watching FY26 headcount progression at the 10 companies with explicit AI-tied actions, and tracking whether additional CEOs and CFOs echo Ellison-style claims on Q3 and Q4 2026 earnings calls. Tipping point: 20 or more companies making explicit AI-driven structural headcount commitments.

3. Does horizontal SaaS find AI-driven acceleration broadly?

The data through CQ1 2026 says the horizontal SaaS cohort is still stuck at pre-COVID growth rates with rare exceptions. The market will be watching names including HubSpot, Atlassian, monday.com, Klaviyo, Dropbox, Box, Pegasystems, and Appian over the next two to three quarters for any meaningful acceleration tied to AI products. Jury is out on the horizontal-SaaS-can-grow-with-AI thesis vs commoditization risk thesis for horizontal SaaS.

4. Does consumption pricing become the dominant new-customer billing model?

Several of the companies in cohort are on consumption layers today. We will be watching for any vertical SaaS adopters (CCC, Veeva, Guidewire, Alkami), the disclosed consumption ARR breakouts from Adobe and Box, and the performance of monday.com's seats-plus-credits launch through Q2 2026. If consumption ARR crosses a meaningful threshold (say 15% of cohort total) in any sub-segment, the billing model conversation shifts industry-wide.

5. Does the AI agent governance layer consolidate or fragment?

Okta Auth0 for AI Agents, ServiceNow AI Control Tower, CrowdStrike Charlotte AI, Datadog Bits AI and Rubrik Agent Cloud are all going after the same identity, observability and security layer for non-human agents. Palo Alto Networks recently announced its intent to acquire Chronosphere in an adjacent observability play. This is the most likely cohort consolidation area for M&A in the next four quarters. Would be interesting to see who claims early leadership here.

6. Does MCP become the standard agentic protocol or does fragmentation set in?

Figma, MongoDB, and Box have shipped Model Context Protocol servers integrating with Claude, Cursor, VS Code and other agentic clients. We will be watching whether enterprise software companies converge on MCP versus building proprietary integration protocols. If MCP becomes a true standard, the composable agentic stack thesis wins and integration costs fall industry-wide. If proprietary protocols proliferate, vertical integration wins and the closed-platform AI players have a moat.

Methodology

All counts in this paper are drawn from documents filed with the SEC: earnings 8-Ks (Item 2.02 with Exhibit 99.1 press release), 10-Qs, 10-Ks, 6-Ks for foreign private issuers, and 20-F annual reports. Sell-side reports are not consulted for this analysis. For the full cohort list and categorization, see Part I.