TwelveLabs Just Raised $100M to Make Video Searchable. Here's Why That's an Enterprise Infrastructure Story.

In the morning of July 1, 2026, Gartner published a press release with a number that landed differently than most analyst forecasts: $234 billion in enterprise application software spend will be at risk from agentic AI by 2030. That same afternoon, Microsoft's July 1 M365 price hike took effect — raising Microsoft 365 E3 from $36 to $39 per user per month, and Office 365 E3 by 13%. Same day. Opposite directional bets.

The symmetry is not a coincidence. It is the clearest possible illustration of where enterprise software stands in mid-2026: two simultaneous forces pulling in opposite directions, with the winner determined by which argument reaches the enterprise buyer's budget committee first. Microsoft's bet is that bundling Copilot into existing licenses, combined with the switching friction of a deeply embedded platform, will sustain per-seat pricing. Gartner's forecast suggests that bet is being tested in real time.

The mechanism Gartner is describing has a new name: agentic arbitrage. The idea is that AI agents — systems capable of completing multi-step business tasks autonomously by calling APIs, reading data, and taking actions — can now deliver business outcomes that previously required human interaction with enterprise software UIs. When an AI agent schedules a meeting, logs a sales interaction, or routes a support ticket without a human ever opening Salesforce, Workday, or ServiceNow, the usage pattern the enterprise license was built on stops materializing.

The seat-based license model assumes that value is delivered when humans use software. Agentic AI is severing that assumption at scale.

What "Agentic Arbitrage" Actually Means

George Brocklehurst, Managing VP at Gartner, put the mechanism plainly in the July 1 release: "Agentic systems deliver outcomes directly, bypassing traditional UX-heavy applications and making the software invisible. This breaks the link between user growth and revenue growth."

The economic logic runs like this. A company pays $150 per user per month for a CRM like Salesforce. That price is justified by the number of sales reps who log calls, manage opportunities, and review pipeline in the platform. Now an AI agent does the logging. It reads the email thread, extracts the interaction summary, and writes it to the CRM via API — without the rep ever opening a browser tab. The outcome — a logged interaction, an updated pipeline stage — is identical. The user who consumed the license's value is gone.

"A CRM is not your sales process," Brocklehurst noted in Gartner's accompanying CIO coverage. "It's a component in your sales process, but there's a lot that sits outside of it." That is the tell. The vendors who built CRMs on the premise that their UI was the process are now discovering that AI agents can execute the process while treating the CRM as read/write infrastructure — valuable, but not separately licensable at per-seat rates.

The $234 billion number represents Gartner's estimate of the total enterprise application software spend that will face this structural pressure by 2030 — approximately 20% of projected total enterprise software revenue. It is not a forecast that all of it disappears; it is a forecast that the commercial model undergirding it becomes increasingly difficult to defend against buyers who can demonstrate, empirically, that fewer humans are touching the software.

The Six Categories Most Exposed

Not all enterprise software is equally exposed to agentic arbitrage. The vulnerability scales with two factors: how much of the software's value depends on human UI interaction versus data storage and API access, and how well-documented the APIs are for agent interaction.

The highest-risk categories share a common characteristic: their primary value is workflow coordination, not proprietary data or deep computational insight. When AI agents can coordinate workflows directly — reading, writing, and updating records across systems without a human navigating between tabs — the per-seat justification collapses.

The lower-risk categories have structural advantages that agents cannot replicate: Microsoft's M365 advantage is that it stores the documents, emails, and collaboration history that agents need access to. Becoming the data layer — rather than just the UI layer — is the defense. The price hike and Copilot bundling are Microsoft's explicit bet that this data layer position sustains per-seat economics even as agent adoption grows.

Who's Already Bypassing the Stack

The $234 billion risk is not a 2030 problem waiting to arrive. Levi Strauss & Co. is the clearest documented case of agentic arbitrage in production at enterprise scale.

According to Gartner's analysis, Levi Strauss has built specialized AI agents for HR, finance, IT, and retail operations — each agent handling a defined category of workflow within its domain. The company is now building what it calls a "Super Agent," an orchestration layer that connects those specialized agents into a unified interface, enabling employees to complete tasks that previously required switching between multiple enterprise applications. The Super Agent does not eliminate Levi Strauss's enterprise software stack; it routes work through that stack via API while reducing the human UI sessions that justify per-seat licenses.

Ramp, the corporate card and spend management company, launched Applied AI Solutions in June 2026 — agentic workflows for accounts payable, expense management, and vendor negotiation across multiple enterprise systems. Ramp's value proposition to CFOs is that its AI agents can execute financial workflows that previously required human interaction with multiple software platforms. The implicit competitive claim: you do not need per-seat licenses in as many systems if agents are handling the workflow.

The pattern at both Levi Strauss and Ramp is consistent: AI agents become the integration and orchestration layer, treating enterprise software as infrastructure rather than as a primary work surface. The human-facing interface shifts from the enterprise application to the AI agent, while the enterprise applications become data stores and action APIs that agents call. The license still matters — but it matters as infrastructure access, not as productivity software for human users.

The Irony of the Microsoft Price Hike

The Microsoft 365 price increase — M365 E3 to $39/user/month, Office 365 E3 up 13%, effective the same day as the Gartner report — is not a sign that Microsoft is ignoring the agentic threat. It is a sign that Microsoft is making a specific bet about where the defense lies.

Microsoft's logic is coherent on its own terms. Copilot is bundled into higher M365 tiers as justification for the price increase, converting Copilot from an add-on expense to a core platform feature. The bet: enterprise customers who are deepest in the Microsoft ecosystem — with decades of documents in SharePoint, email history in Exchange, meetings in Teams, files in OneDrive — will not walk away from the platform that holds all of that context, regardless of how agentic the software landscape becomes. The data moat is the defense.

As Signal's analysis of Microsoft 365's Copilot bundling strategy documented, the price hike arrives with defensible logic for the specific segment of Microsoft's customer base that is deeply embedded in M365 workflows. It is considerably less defensible for the enterprise software vendors who do not have that data depth — whose primary value delivery mechanism is a UI that AI agents can now navigate around.

The tension between the Microsoft price hike and the Gartner risk report is a compressed illustration of the 2026 enterprise software bifurcation: companies with genuine data moats and deep workflow embedding are raising prices with justification; companies without them are about to discover that the buyer's willingness to pay for per-seat licenses is being systematically eroded by the same AI capabilities they've been promoting as productivity features.

The Pricing Models That Survive

The Agentforce $800M ARR benchmark is the most compelling empirical evidence that enterprise buyers will pay for AI-powered software — if the commercial model is structured around outcomes rather than seats. Salesforce Agentforce charges approximately $0.002 per action, not per user. Intercom Fin charges per conversation resolved, not per support agent seat. These are not niche experiments; they are the commercial architecture that enterprise software companies need to build toward if they want to survive agentic arbitrage.

The shift requires rethinking not just the price point but the entire value narrative: from "your team uses our software" to "our software achieves outcomes for your team." Four commercial models are defensible against agentic arbitrage:

1. Outcome-based pricing ties revenue directly to the business result the software produces. Intercom's per-resolution model. Salesforce Agentforce's per-action model. Gong's deal-influenced-revenue attribution. The value argument is clear and measurable, and it does not depend on how many humans are touching the software to produce the outcome.

2. Data ownership moats create value that agents need but cannot replicate. Salesforce's CRM value is partly the historical customer data it holds; Workday's value is partly the HR and compensation data it stores. Software that becomes the authoritative system of record — the source of truth that agents query and update — retains enterprise value even as the UI interaction model changes.

3. Workflow embedding positions the AI agent experience as native to the product rather than as a bypass of the product. The companies that survive agentic arbitrage are those that make their software the ambient layer where agents operate — rather than the layer agents route around. Salesforce's Agentforce strategy is this pattern: put the agents inside Salesforce, rather than let them bypass it.

4. Human-in-the-loop compliance gates create licensing positions that agents cannot eliminate because human approval is legally or operationally required. Financial approval workflows, healthcare decisions, legal reviews — these are categories where named-user or seat-based licensing retains enterprise justification because the human in the loop is a compliance requirement, not just a usage pattern.

The Agentic Exposure Audit

For product leaders at enterprise software companies, the Gartner $234B number is most useful not as a forecast to track but as a framework for assessing your own company's exposure. The relevant diagnostic questions:

What percentage of your product's core value is delivered through UI interaction versus data storage and API access? If the answer is primarily UI interaction, your exposure is high. If your product holds data that agents need and exposes it through well-documented APIs, you are becoming infrastructure rather than being displaced by it.

What is your customers' current API usage pattern relative to UI usage? Teams already using your APIs heavily are the early leading indicator of what happens when those API interactions are handled by agents rather than humans. If API usage is growing faster than seat count, that is not a growth signal — it is the agentic arbitrage signal arriving early.

Can you define a measurable outcome that your product produces? Companies that can cleanly define "our product reduces invoice processing time by X hours" or "our product increases sales cycle velocity by Y percent" can restructure pricing around that outcome. Companies that can only define value as "our teams use this software" cannot.

Do your enterprise contracts have provisions for what happens when AI agents are the primary users? Most enterprise software contracts were written assuming human users. As enterprises deploy AI agents that interact with software at scale, what constitutes a "user" will become a significant negotiation point at renewal. Getting ahead of this conversation is significantly better than being reactive when the contract language becomes the battleground.

The AI inference price war that has driven inference costs down 90% in 18 months is directly accelerating the agentic arbitrage dynamic: as the cost of running AI agents falls, the economic case for deploying agents to handle workflow automation rather than having humans operate enterprise software UIs becomes compelling at an earlier scale point. The cheapest frontier inference is now under $0.15 per million tokens. At that price, automating an enterprise software interaction that costs $0.003 in compute is economically rational at any meaningful volume.

How the Transition Plays Out

Gartner's $234B risk is spread across a four-year window (2026-2030), which is both reassuring and sobering depending on your position. For enterprise software companies in the high-exposure categories, four years is enough time to transition commercial models if the work starts now. It is not enough time to ignore the signal and pivot successfully when buyer pressure becomes explicit in contract negotiations.

The transition window is already closing for some categories. CRM workflow automation via AI agents is a deployed reality at companies like Levi Strauss, not a future risk. ITSM automation is in active deployment across enterprise IT operations. The Gartner 40% enterprise AI agent mandate published in late 2025 was about deployment of AI agents within enterprise applications; the July 2026 report is about the commercial consequences of those deployments for the applications' own pricing models. These are sequential chapters of the same story.

The companies that will look back at July 2026 as the inflection point they caught are those that use the next two years to define outcome metrics for their products, restructure enterprise contracts around outcomes rather than seats, build agent-friendly API surfaces that make their product more valuable as AI infrastructure, and create proprietary data positions that agents need access to rather than compete with.

The companies that will look back at July 2026 as the inflection point they missed are those that treat the per-seat pricing model as a constant of enterprise SaaS rather than a variable that the underlying technology is actively changing. Microsoft's price hike is defensible because Microsoft has the data moat. For the enterprise software companies without Microsoft's data depth, waiting to find out if that bet pays off is not a strategy.

Takeaway: Gartner's $234 billion forecast is the analyst community's formal recognition of what enterprise buyers are already discovering: AI agents can deliver workflow outcomes without the human UI interactions that justify per-seat licensing. The $234B figure is less a prediction than a quantification of a structural shift already underway at companies like Levi Strauss and Ramp. The response that works — outcome-based pricing, data ownership moats, workflow embedding, compliance gate monetization — requires treating the agentic transition not as a product feature to add but as a commercial model to rebuild. The SaaS companies that start that rebuild now have a four-year window to complete it. The ones that wait may find the window closing while they are still debating the menu price.