Enterprise Software Watch: Enterprise Software Redraws Itself Around Who Gets to Act
Strategic Intelligence
A $234 billion at-risk estimate, an $830 billion February stock rout, and a wave of execution-layer M&A all point to the same fight: the interface versus the agent that can act on it.
The Enterprise Software Lead
Enterprise software's central question this year is no longer whether AI agents belong inside the stack — it's whether the stack itself survives the transition intact. Gartner's most recent estimate puts $234 billion of enterprise SaaS spending at risk from "agentic arbitrage" between now and 2030, built on the observation that most software has historically been priced and sold to people, not to autonomous systems acting on people's behalf. That framing isn't theoretical: in early February, software and services stocks lost an estimated $830 billion in value over six trading days after Anthropic rolled out advanced agentic plugin capabilities — a direct market judgment that a meaningful slice of application-layer SaaS value has depended on workflow lock-in rather than genuine technical defensibility.
Vendors have responded not by waiting to be disrupted but by buying the pieces they need to control how agents actually execute work. A wave of 2026 acquisitions — Asana buying StackAI, Coupa buying Rossum, Salesforce buying Contentful, and Vertice buying Vendr — all targeted the execution layer specifically, not the conversational or summarization layer that dominated the first round of "AI features" bolted onto existing products.
AI & SaaS Transformation
The clearest evidence of category-level disruption is in pricing. AI agent software spending is projected to reach $206.5 billion in 2026, up 139% from $86.4 billion in 2025, but that growth is arriving alongside a structural rethink of how software gets priced at all. Deloitte's own 2026 outlook argues subscription and seat-based licensing are giving way to hybrid usage- and outcome-based models — a shift that fundamentally changes SaaS unit economics, since a vendor billing for "seats" occupied by employees has a very different revenue relationship with a customer than one billing for tasks an agent completes autonomously.
Gartner's adoption curve is the number every enterprise software buyer is now working against: task-specific AI agents are projected to sit inside roughly 40% of enterprise applications by the end of 2026, up from less than 5% a year earlier, with agentic capability potentially driving close to a third of all enterprise application revenue — north of $450 billion — by 2035. Deloitte's own analysts are notably more conservative on the pace of actual displacement, arguing that full agent substitution for existing enterprise applications is realistically five-plus years out because traditional SaaS providers still own deep, complex workflow footprints that are genuinely hard to replicate, not merely defended by inertia.
Enterprise AI Adoption
Adoption is real and accelerating, but it is also proving more expensive than most enterprises budgeted for. Agentic workflows consume five to thirty times more tokens per completed task than a simple chatbot query, and enterprises that piloted AI against chatbot-era usage assumptions are now seeing bills scale well beyond what their original ROI models projected — even as the underlying per-token price of intelligence keeps falling. That gap between falling unit costs and rising total spend is becoming the central budgeting problem for any CIO scaling agents past a pilot.
It's also reshaping how enterprises manage their existing software footprint. Sixty-eight percent of technology leaders say they plan vendor consolidation in 2026, most targeting a roughly 20% reduction in the number of providers they use, and the average enterprise now runs 106 SaaS applications, down from a peak of 130 in 2022. Budget pressure, underused licenses, and shadow-IT risk are the stated drivers — but "shadow AI," meaning employees and agents using AI tools outside sanctioned procurement, is increasingly cited as the more expensive hidden risk sitting underneath that consolidation push.
Software Competitive Landscape
The clearest winners right now are companies that control the layer where an agent actually takes action rather than the layer where it merely advises or summarizes. That's the strategic logic behind the Asana-StackAI, Coupa-Rossum, Salesforce-Contentful, and Vertice-Vendr deals — each acquirer bought a capability that lets an agent execute a task (build a workflow, extract structured data, manage content, negotiate a contract) rather than simply describe one. Sierra, the customer-support AI agent company founded by former Salesforce co-CEO Bret Taylor, and Snowflake — which has reportedly crossed $7 billion in lifetime AWS Marketplace sales — are both frequently cited as examples of vendors successfully repositioning around agent-native architecture rather than defending a legacy seat-based product.
The most exposed category is best described as "system of engagement" software: point-solution tools whose primary value has been serving as the human interface to an underlying dataset or workflow. Once an agent can query that same data and execute the same workflow without a human clicking through a UI, the pricing logic that justified per-seat licensing weakens considerably — which is precisely the dynamic behind February's stock rout and Gartner's $234 billion at-risk estimate.
SaaS Economics
M&A volume tells the clearest story about how the market is repricing itself. SaaS M&A hit 2,698 closed transactions in 2025, a 28% jump over 2024 and the highest annual count on record, and the first quarter of 2026 kept pace with an estimated 620-plus deals worth more than $95 billion in aggregate value. But the market has bifurcated sharply on valuation: high-growth companies with strong retention and genuine AI-native positioning are commanding six to eight times annual recurring revenue, while undifferentiated legacy businesses are compressing toward three to four times ARR. That spread — roughly double — is the clearest signal yet that buyers are now actively separating companies with real AI-native architecture from companies that have simply bolted AI features onto an existing product, and pricing the difference accordingly.
Early-stage funding reflects the same bifurcation. Recent weeks alone produced Series A and B rounds for AI-native cybersecurity, supply-chain, and financial-services agent platforms — Freehand's $75 million Series B for autonomous supply-chain and back-office agents, Encore AI's $30 million Series A for banking and insurance agents, and multiple cybersecurity rounds addressing "shadow AI" risk directly — evidence that investors are underwriting narrow, defensible agent workflows rather than broad horizontal AI platforms.
M&A & Strategic Moves
Beyond the four headline execution-layer acquisitions, the broader pattern in 2026 dealmaking is enterprise platforms, automation vendors, cloud infrastructure players, cybersecurity companies, and frontier AI labs all competing to buy agentic capability rather than build it from scratch — a much wider buyer base than the AI-native-startups-buying-startups dynamic that characterized 2024 and early 2025. One tracker puts the shift in stark terms: 35 agentic AI M&A deals closed in the most recent twelve-month period versus just 9 in the twelve months before that, a shift researchers describe as structural rather than cyclical.
The categories drawing the most acquisition interest — customer-service agents, coding agents, IT operations agents, procurement agents, compliance agents, agent security, agent search, inference infrastructure, and data governance for agents — map closely onto the operational plumbing an enterprise actually needs before it can trust an agent to act autonomously, rather than the flashier conversational layer that dominated the earliest wave of enterprise AI products.
Three Enterprise Software Signals
-
01
Buyers are now pricing the difference between AI-native and AI-retrofitted software explicitly. A roughly 2x valuation gap between high-growth, AI-native SaaS companies and legacy peers with bolted-on features shows the market has moved past treating "has AI features" as a meaningful differentiator.
-
02
M&A is concentrating around the execution layer, not the interface layer. The Asana, Coupa, Salesforce, and Vertice deals all targeted the capability to let an agent act — a clear signal that owning the moment an agent stops advising and starts doing is now viewed as the more defensible strategic position.
-
03
Vendor consolidation and agent adoption are reinforcing each other. As CIOs cut their application count by roughly 20% while simultaneously ramping agent deployment, the software that survives the consolidation wave will increasingly be the software an agent, not just a human, can operate directly.
The CODEW Take
Is AI strengthening the enterprise software market — or fundamentally changing what enterprise software means?
The evidence points toward the latter, though the transition is proving slower and more expensive than the most aggressive forecasts suggested a year ago. The $830 billion February stock rout and Gartner's $234 billion at-risk estimate both show that the market has already concluded seat-based, interface-dependent software carries real disruption risk. But Deloitte's more conservative five-year-plus timeline for actual displacement, and the fact that agentic workflows are proving far more token-expensive than budgeted, suggest full-scale software substitution is not imminent. What is happening now, concretely, is a redefinition of where value sits within the stack: from the interface that a human clicks through, to the execution layer an agent operates directly — and the acquisition spree by Asana, Coupa, Salesforce, and Vertice this year is best read as the clearest evidence yet of vendors racing to own that layer before someone else does.
Editorial Note
The CODEW Enterprise Software Watch examines the companies, platforms, business models, and technologies shaping enterprise software, SaaS, enterprise AI, data platforms, developer tools, and workflow automation.
Reviewed by Erwin Castro
on
Wednesday, August 12, 2026
Rating: