The New Enterprise Platform Battle: AI Becomes the Interface

This series was inspired by what I found in Anthropic's S-1, the filing it made on October 26 for its IPO.

The enterprise is AI curious. We're only beginning to explore what it can do. 90% of CMOs are experimenting with AI, and just under 10% have scaled it. Meanwhile, like a movie with multiple subplots, a lot is going on. Enterprise software providers are in reinvention mode. The CISO office is in discovery. The AI companies are looking for capital to transform themselves. And the likely meeting point is the enterprise.

A $1.5 Trillion Question

Anthropic's IPO is expected to value the company at $1.5 to $2 trillion. $1.5 trillion divided by the $68.9 trillion value of the whole US stock market (World Federation of Exchanges, end of 2025) is about 2%. Against the world's $157.8 trillion, it's about 1%.

The business is real. The S-1 shows $16.3 billion of revenue in the first half of 2026, up from $4.6 billion for all of 2025. More than 1,900 customers pay over $1 million a year. But it spent about $2.76 to run the business for every $1 of revenue in 2025, and about $1.04 in the first half of 2026. And by Fortune's count it has committed to about $441 billion of compute.

So how does Anthropic live up to that valuation? Where is the real fight for cash flow? I think it's in the money companies already spend on enterprise software. That's the platform battle.

Where Work Starts

Picture a CMO and their entire team working in Claude or ChatGPT all day, without once logging into Salesforce or Marketo. The CMO asks how campaigns performed, for a forecast and for ideas for next quarter. An agent pulls the CRM data, looks at what marketing ran, updates the plan and sets up the next steps across several systems.

This is already happening for many people today. It's super powerful because the AI tools can talk to multiple systems through MCP connectors. It starts as a chat. Then it becomes a skill. Then an agent is created that delivers the experience, and then an app that contains the agents. It's a pattern we see frequently, for ourselves and for our clients. And as of Q4 2026, this experience is coming to mobile devices, through the AI apps and through custom apps.

The applications still matter. But the place where the work happens has moved.

Salesforce sees it. In March its co-founder Parker Harris asked, "Why should you ever log into Salesforce again?" In September it launched AIforce to bring Salesforce "to wherever people and agents work," Claude included. Marc Benioff called it "an interface revolution."

Everyone Wants to Be the Platform

Salesforce says the agentic enterprise needs an operating system, not just a model, and it wants to be that operating system. I read that as a platform strategy, not a product feature. The platform doesn't need to build every model. It wants to decide which model does which job, where the work runs and how you pay. Sending simple work to cheaper models and hard work to frontier ones is also how it protects its margin.

Anthropic and OpenAI are coming from the other side. Their models sit inside SaaS apps, which gives them distribution. Their own assistants give them a shot at being where people spend the day. Dario Amodei put it well when Anthropic expanded its Snowflake partnership last December: "Enterprises have spent years building secure, trusted data environments, and now they want AI that can work within those environments without compromise." An AI company doesn't need to replace your CRM or ERP. It needs to connect to them.

The S-1 shows how tangled this is. Anthropic lists Salesforce and Snowflake as partners that help bring Claude to customers, along with AWS, Google Cloud, Microsoft Azure and firms like Accenture and Deloitte. It also warns that leaning on a few cloud providers that are also its investors and customers "creates complex dynamics." Partners and rivals at the same time.

Generalized vs. Specialized

That's the real split. On one side are the general models from Anthropic, OpenAI and Google. They're the smartest across the widest range of work, and they get better every few months. On the other side is specialized AI: intelligence a company owns, tuned to its own data and workflows.

Gartner predicts that by 2027, organizations will use small, task-specific AI models three times more than general-purpose large language models. Its reasoning: general models lose accuracy when a task needs business context, while small ones respond faster and cost less to run.

The case for specialized is that it can beat a general model on the jobs that matter to you, with better performance, lower latency and lower cost. That's why software vendors want their own AI, where they can stand out and keep the margin, and why many enterprises want to own the intelligence instead of renting it.

The case for general is that the frontier keeps moving. A specialized model has to keep up with general models that keep getting smarter on their own.

Here's what the big software vendors are using, as of October 2026:

• Microsoft is doing both. It owns about 27% of OpenAI and offers Anthropic's Claude in Copilot. At Build in June 2026 it released its own MAI models. In July, Bloomberg reported that Microsoft had started using them in place of OpenAI and Anthropic models in Excel and Outlook, though by most reports OpenAI and Anthropic still handle most Copilot traffic.
• SAP chose general. At Sapphire in May 2026 it made Claude a primary reasoning engine for its Joule assistant and agents, connected to its systems through MCP. SAP owns the workflow and rents the intelligence.
• Salesforce sits in between, with its own Agentforce agents on one side and Claude on the other.
• Workday sells its own agents and is bringing its assistant, Sana, to Claude, with general availability expected late in 2026.
• HubSpot has a free connector for ChatGPT, next to its own Breeze agents.

So where does that leave Anthropic? If customers want specialized AI, you'd expect Anthropic to sell ways to build it. It does, but mostly around the model rather than inside it. Fine-tuning Claude itself has only been offered in a limited form, for an older Haiku model on Amazon Bedrock. What Anthropic sells instead are Agent Skills, folders of instructions and scripts that teach Claude a company's way of doing a task, which admins can roll out to everyone on Team and Enterprise plans. Add plug-ins for finance, engineering and design, and connectors to the systems a company already runs. That's specializing the work, not the model. The S-1 suggests it's working: more than half of customers spending at least $100,000 a year now use both Claude's platform and one of its own apps, such as Claude Code or Cowork, up from about 30% at the end of last year.

In practice I expect most companies to use both. Whoever decides which one runs on each task owns the platform.

The Pricing Problem

Here's the question I'd ask every SaaS product leader. If customers don't use your interface anymore, what exactly are they paying you for?

Seats worked when value tracked people. Agents break that. A smaller team can run far more work through agents. Gartner says up to $234 billion of enterprise application spending is exposed to agents between now and 2030, and calls it "an existential threat for vendors who are defending legacy dashboards and seat-based models." Buyers are moving too. In Futurum's first-half 2026 survey of enterprise software decision makers, fewer than 20% still favor per-user pricing.

At Dreamforce, Benioff said, "We've had to throw away our price book, basically." Agentforce now has a menu: $125 per user per month for the add-ons, $2 per conversation, or Flex Credits at $500 per 100,000, where a standard action is 20 credits, or 10 cents. That's a wider menu. It isn't proof that every outside agent calling Salesforce pays the same. Watch the contracts, the meters and the exclusions, not the announcements.

Salesforce isn't alone. Workday sells its agents on Flex Credits, launched in September 2025, and doesn't publish a dollar price per credit. It says AI now drives more than 25% of new contract value. Dow, one of its customers, paused rolling out Workday's agents, partly over how Flex Credits are governed and paid for. According to pricing guides from HubSpot partners, HubSpot now charges 50 credits, about 50 cents, for each conversation its customer agent resolves, down from about $1 for every conversation before April 2026.

The AI side has a meter too. The S-1 says Anthropic's revenue is "primarily consumption-based." Every task an agent runs through Claude earns Anthropic money. Every task run in Agentforce earns Salesforce money. Same work, two meters. That's the fight for cash flow.

And this business model will probably change. Today compute is rented, paid for by the token. In the future it will most likely be capital spending, owned instead of rented. Meanwhile Anthropic wants the direct relationship with the customer, and every channel that buys AI and delivers it inside its own app wants to build its own intelligence layer. They believe it can be better, both for the customer and for them.

Headless Doesn't Mean Worthless

It's tempting to say SaaS becomes a database when AI takes over the interface. I don't buy it. Enterprise apps enforce approvals, run business rules, reconcile transactions, keep audit trails and apply permissions. That's hard to replace.

But if the AI assistant becomes the daily workspace, the vendor loses the screen, and the screen is what justified more seats. A vendor can stay essential and become invisible at the same time.

What I'm Watching

• Contract terms. Longer terms, bundled AI commitments and renewal discounts that keep customers anchored while the interface changes. Signals to look into, not proof of lock-in.
• Bring-your-own-AI costs. When Claude or ChatGPT calls Salesforce, Marketo or NetSuite, what do you pay, and per what?
• Pricing beyond seats. Consumption, workflow and outcome pricing that doesn't punish useful automation.
• Model choice. "We support Claude" isn't the same as letting you run your business from Claude.
• The CISO office. Security teams will decide which AI tool gets to touch company data, and that may decide who owns the platform.
• Open access. APIs, connectors and permissions that let agents do real work without sending people back to the native screens.

I wrote about the same shift from the vendor side in SaaS Is Bringing the Work to You. Then It Sends the Bill. That one is about your meter. This series is about who owns the platform.

This prompt shows where your team's work starts today and who controls it.

What This Means for CMOs

For CMOs this isn't abstract. It will show up in our contracts, our budgets and our ability to pick the tools our teams use. Run the prompt above, and bring what it finds to your next renewal.

If you found this article interesting and want to talk about AI and your organization, reach out.

Sources: Anthropic, Form S-1, October 26, 2026 (revenue, customers, partners, risk factors; operating expenses of $12.6 billion on revenue of $4.6 billion in 2025). Shawn Tully, "Anthropic's S-1, by the numbers," Fortune, October 27, 2026. Yahoo Finance, Anthropic IPO live coverage, October 26, 2026 (valuation range). World Federation of Exchanges via SIFMA, global and US stock market value at the end of 2025. Anthropic, "Snowflake and Anthropic expanded partnership," December 3, 2025. Salesforce, AIforce announcement, September 15, 2026, and Dreamforce 2026 keynote. Salesforce, "Why the agentic enterprise needs an operating system, not just a model," April 2, 2026. Gartner, "$234 Billion in Enterprise Application Software Spend Is at Risk from Agentic AI," July 1, 2026. Futurum Research, 1H 2026 Enterprise Software Decision Makers survey, May 2026. Salesforce, Agentforce pricing page and Help article on Flex Credits. Gartner, "Gartner Predicts by 2027, Organizations Will Use Small, Task-Specific AI Models Three Times More Than General-Purpose Large Language Models," April 9, 2025. SAP News, "SAP and Anthropic to Bring Claude to SAP Business AI Platform," May 2026. Microsoft Build 2026; Bloomberg's July 2026 report on MAI models in Excel and Outlook, as summarized by other outlets. Workday Flex Credits announcement, September 16, 2025; Subscription Insider and Diginomica on Workday's AI contract value and Dow. HubSpot partner pricing guides on the April 14, 2026 credits change. Anthropic, Agent Skills and enterprise plug-ins announcements; TechCrunch, "Anthropic's IPO filing reveals its strategy for keeping enterprise customers hooked," October 26, 2026.