In December 2025, Anthropic’s coding agent started actually working. Not a demo. Not a research preview. The thing that had been promised for two years finally landed.
You don’t have to take our word for it. Andrej Karpathy said the same thing in an interview: “a lot of people experienced AI last year as ChatGPT-adjacent. But you really had to look again, and you had to look as of December, because things have changed fundamentally — especially on this agentic coherent workflow that really started to actually work.”
And in a recent podcast, Marc Benioff said Salesforce is spending $300 million on Anthropic this year — to code. His words on what shifted: “when 4.6 hit, boom, everyone could code in their companies. And before that, they really couldn’t.” And: “I can implement my software and sell it at the same time. I’ve never been able to do that before.” He called Anthropic “a rocket ship that will not stop.” (Source: The Next Web)
One of the most-cited researchers in the modern AI era and the CEO of the world’s largest CRM independently mark the same month as the turning point, that’s not hype. That’s a signal.
Here’s the curve behind it. Anthropic went from $1 billion to $30 billion in annualized run-rate in about 15 months. Salesforce took 20 years to reach $30B. IBM took 79. That shape doesn’t exist anywhere else in enterprise software history.
And it isn’t only Anthropic. Microsoft reportedly canceled internal Claude Code licenses after consumption blew past forecast. Uber reportedly burned its entire 2026 AI budget in four months. GitHub Copilot switches to usage-based billing on June 1. The market noticed: Jefferies named the SaaSpocalypse on February 4, after roughly $1 trillion in software market cap got erased from the October 2025 peak. Their thesis was simple — per-seat licensing breaks when one agent does the work of five employees.
We delivered this argument in May Flexera Live in Houston: capability is moving faster than understanding, and every discipline we built to translate capability into understanding — ITAM, SaaS management, FinOps — assumed a unit of work that just moved. The application is no longer the unit. The seat is no longer the unit. The inference is.
Shadow SaaS and SaaS pricing are making it harder to see and control usage
Shadow SaaS 2.0
The first we are calling Shadow SaaS 2.0. Classic shadow IT was about apps your CISO didn’t approve. Shadow SaaS 2.0 is different. The agent doesn’t log in. It runs on a personal API key, a developer’s credit card, a CI/CD pipeline, a cloud function. Your SaaS management tools were built around identity and login events. The new unit doesn’t generate either one. Identity-based discovery covers only a fraction of the AI running in your environment. The rest is invisible.
Regulators are watching Shadow SaaS 2.0 too. They are already enforcing AI governance requirements, including:
- EU AI Act enforcement
- State-level AI regulations
- NIST AI risk frameworks
- SEC prioritization of AI governance
- And now cyber insurers require AI governance alignment for coverage
SaaS Pricing Has Shifted from Seats to Consumption
SaaS pricing is moving from predictable, seat-based models to variable, AI-driven consumption.
What is changing:
- Pricing is multi-metric and units now include tokens, API calls, and compute minutes instead of users.
- Usage becomes non-linear, where a small number of workflows can drive disproportionate cost
- Consumption data is often opaque and aggregated at the account level
The impact of this change is spend visibility and volatility, difficulty forecasting budgets, and challenges with allocating costs to departments or projects. Traditional license optimization simply does not work in this model. Most SaaS management tools were built to track one pricing model. They were not designed to reconcile four.
With AI, SaaS ownership is even more fragmented
In Gartner’s Critical Capabilities for SaaS Management Platforms (published in July 2024), they stated most organizations don’t have a clear owner for SaaS, with several departments involved in various aspects of the lifecycle. Fast forward to 2026, and we see evidence of this in survey data. According to the Flexera 2026 State of ITAM report, 64% of ITAM pros say they manage SaaS. However, in the FinOps Foundation’s 2026 State of FinOps report, 90% of FinOps pros say they manage or plan to manage SaaS.
If everyone is doing it, is anyone keeping track of the entire program?
AI accelerates that fragmentation. This is because of different buying teams and how you pay. SaaS is now a continuous relationship with dynamic usage and dynamic cost. Effective management requires coordination across teams that historically worked in parallel, not together.
SaaS lifecycle management with AI agents
AI introduces a new lifecycle challenge: agents. Traditional Joiner-Mover-Leaver (JML) processes work for humans. They do not work for agents.
When a human leaves, SSO is revoked, licenses reclaimed, mailboxes archived. Clean handoff. Agents don’t get reclaimed. They keep running. They keep billing. They keep accessing data. So the question is simple.When the human leaves, who takes responsibility for the agent?
Organizations must define:
- Ownership for every agent – at registration
- Lifecycle management for agents
- Governance triggers when ownership changes
The new application rationalization – contract to inference rationalization
The second is we named on stage as contract-to-inference rationalization. The discipline of matching each inference to the right model and the right meter, priced against the contract you actually have. Not against a rate card. Not against a vendor pitch deck. Against the contract sitting in your repository today.
Benioff described the same practice in different words on the same podcast: “the vast majority of those tokens don’t need to go to Anthropic. There needs to be some intermediary layer that… can route it to the most affordable for the job.” And: “there’s going to be a hot new company that’s going to come along and say… I’m going to sit between you and Anthropic and OpenAI.”
He’s describing the discipline. He just isn’t naming it.
That naming matters, because this is what SaaS management was always supposed to do — verify you aren’t paying a dollar more, or a dollar less, than the workload actually needs. Across every vendor in your environment. Across every license tier and model available today. And re-verified as the market moves underneath you. That job didn’t disappear when the unit changed. It got harder.
We’ve solved a problem of this shape before. In 2014, cloud broke the contract. Finance saw the bill. Engineering owned the workload. Neither side saw the other’s half. We didn’t fix that with a tool. We built FinOps — a cultural reset, a shared vocabulary, a new operating model. The tools followed the thinking. They had to.
Same shape now. Six functions at the table instead of two: finance, engineering, security, legal, the business, vendor management. Four meters running at once. No shared vocabulary yet. The vocabulary comes first. The tools follow.
How can I manage SaaS in an AI-first world?
Here are some recommended first steps to start managing SaaS and AI.
- Unify your SaaS data: Most SaaS/AI apps are accessed via a browser. Leverage broad discovery capabilities such as a browser extension or CASB to get an idea of what is used in your organization. For example, if you’ve decided to standardize on Microsoft Copilot and others are using Claude, this is a way to find out how widespread use is of a non-standard app and dig in further into vendor portals or use your SaaS management platform’s API connectors to dig into the details. In addition to identifying use, it’s important to understand your SaaS contract and align that to use to find potential issues with overages.
- Align your team ownership structure: If everyone is managing SaaS, is anyone looking after the entire program? Identify all the SaaS used in your organization, where the billing is coming in (direct, via cloud bill, etc.), identifythe app owner and the data and processes used to keep costs in check. Get clarity on who is responsible for identifying shadow SaaS/AI and making rationalization decisions across the organization.
- Develop a continuous discipline: The constant of managing SaaS and AI is the way you do it today is not the way you will do it tomorrow. Build change into your processes and policies. Build in automation where possible and consider allocating more time to application rationalization/build vs buy decisions.
