Ramp has raised $750 million at a $44 billion valuation, while Stripe reportedly paid more than $7 billion to acquire OpenRouter. In a column cited by BlockTempo, Simon Taylor argues that both moves are not really about ordinary payments growth. They are bets on new customer jobs created by AI.
Taylor’s central point is that investors are rewarding companies that can capture AI-native demand. He points to token spend management, agent identity, and model routing as examples of customer needs that barely existed before AI but are now turning into attractive fintech revenue pools.
Why Taylor says AI-native demand matters more than legacy payments
According to the column, Ramp recently completed its $750 million financing round at a $44 billion valuation. Taylor writes that Ramp may not even need the money, but the round is buying into more than strong growth. It is buying into a larger story about what it means to become AI native.
That story, he argues, is credible because AI is reshaping what customers need done. One example is routing. Taylor says Router finds the lowest-cost and highest-quality model for each request, and that Ramp says it can save customers 40%.
He describes Stripe as building a similar stack. Payments sit at the base, Metronome handles usage metering and billing, and OpenRouter sits on top routing traffic across more than 400 models. Stripe, he writes, reportedly paid more than $7 billion for that top layer.
Taylor also says growth investors are actively looking for companies that can benefit from the AI boom and become more resilient because of it. He writes that he spoke this week with three founders who were close to raising large rounds under an AI-native label, and each asked versions of the same question: how should they position themselves as AI native, and what should their product actually be?
His answer is that this is a product question before it is a branding question.
A task ladder: old jobs, adjacent jobs, and new jobs born from AI
Taylor leans on Clayton Christensen’s idea that customers do not buy products; they hire them to get a job done.
In finance, old jobs include reimbursing expenses, reconciling invoices, and managing fraud. Those jobs existed long before AI. Modern models may let companies do them faster, at lower cost, and with less labor, but the underlying demand is not new.
AI, in his view, has also created a second category of jobs: monitor my token spend, route each inference request to the best model, and let my agent buy something without handing over my full identity and bank account. Before AI created these problems, he says, nobody had those jobs to solve.
To explain the shift, Taylor uses a 2x2 matrix. One axis distinguishes old customer jobs from jobs created by AI. The other separates companies and products that existed before AI from those that could only exist because modern AI arrived.
In that framework, Nubank using NuFormer to improve credit decisions falls into deep adoption because underwriting is still an old banking job. Harvey, Hebbia, and Rogo belong to a different box: they were not possible before modern models, but they still address long-standing legal and financial research work. Ramp and Stripe sit in the tailwind zone because they already had businesses in place and AI has created new jobs beside them. OpenRouter and fal make sense only because inference has become an industry.
Taylor adds that real companies rarely fit neatly into one square. Ramp’s spend product sits on the old-job side, while its token dashboard and Router sit on the new-job side. Stripe’s payments business also belongs to the old side, while Metronome and OpenRouter pull it toward the AI-created task category.
He then proposes three labels for how companies show up in this market:
- Adopters: companies that use AI to serve existing demand. They may add a copilot or an agent, and they may get faster, cheaper, and leaner, but the structure of the sales pipeline does not change.
- Beneficiaries: companies that both improve old jobs and capture new demand created beside them by AI. Taylor says Ramp and Stripe are the cleanest examples.
- Natives: companies that exist because the new demand exists. OpenRouter routes requests across more than 400 models, while fal runs generative media inference at scale. In a world without inference to sell, he says, neither business would exist.
Token spend, agent interfaces, and discovery are becoming new fintech jobs
Taylor argues that the opportunity set expands as companies move up the ladder from familiar work to unresolved AI-specific problems.
He says AI-driven operating models are already changing how companies are built, governed, and run. Teams are creating plugins around employees, connecting internal data, sharing skills, and changing who can do what. Product managers and designers are writing thread-like code. Engineers are moving deeper into product work. Teams are also building their own internal tools. Even so, he describes this as the price of admission, not the thing that changes what a company sells.
Some AI-native products still serve very old jobs. He says that does not make them any less native. Compliance screening, document and email reconciliation, legal research, and financial reporting were all painful paperwork tasks before AI. He names Beacon, Sardine, Gradient Labs, Harvey, Hebbia, and Rogo as examples of companies using modern models to summarize hundreds of documents and data sources, then return answers or manage workflows. Taylor notes in the column that he is an advisor to Sardine.
For companies working in spend management, account management, or payments, he says old jobs remain in place but new ones are appearing right beside them.
One is: help me manage AI usage. Taylor describes Ramp’s token dashboard as a clear example of an AI-created job living inside an existing product. He acknowledges that cost dashboards are not new in themselves, since Vantage and Datadog have priced cloud and compute spend for years. The difference is the billing object. Before AI token bills existed, nobody needed to categorize token costs, forecast them, or separate COGS from OpEx for token usage. On Ramp, he writes, AI token spend grew 20.7x from June 2025 to June 2026. Metronome, now part of Stripe, sits on the other end of the same job by handling usage metering and billing.
Another is: help my agent connect to your product and complete work. Taylor says Mercury, Visa, and Ramp are rolling out command-line interfaces, or CLIs. Stripe launched its own CLI seven years ago, he writes, but usage jumped after Claude Code arrived. A CLI is a user interface built around the command line rather than an app or a web page, and Taylor argues that agents navigate those interfaces better while using fewer tokens. He gives a Ramp CLI example: in --agent mode, a transaction is returned as JSON and uses about 105 tokens. In --human mode, the same transaction needs 280 tokens for formatting and display.
A third is: help my store get discovered by AI agents. Taylor writes that about one-third of Gen Z now use AI instead of Google to research what to buy. If a store and its SKUs are not visible there, merchants may miss demand. He says Shopify, WooCommerce, and payment service providers working with e-commerce merchants are now optimizing for that shift.
His practical conclusion is that AI is creating an entire category of adjacent demand around existing businesses, and the opening is to build interfaces for those jobs rather than rewrite the whole company overnight.
The control plane question: who gets to orchestrate agents
Taylor then pushes the idea further with a thought experiment. If agents eventually account for 80% to 90% of internet traffic, commerce, and economic activity, how should a company reposition itself, and what becomes its core unit of value?
In that scenario, AI is no longer an add-on feature layered on top of existing business lines. The harder questions are new ones that did not exist before the AI wave.
The first is trust around data handling and decision-making. Taylor says businesses are entering a world where third-party agents built elsewhere may interact with them directly. That creates a need for some mechanism to ensure safety, reliability, privacy management, and accountability tied to a legal entity. A simple pattern is to keep the agent inside a SaaS provider already covered by enterprise agreements. But agents are increasingly becoming the product itself, and they may come from a lab such as Anthropic, a startup, or an internal team. Taylor says he is seeing companies build or buy shells or control planes, such as Primitive, to wrap these agents. He also points to AIUC, which is aiming to underwrite and certify agents.
The second is trust around transactions. If an agent shows up in a store and tries to buy something, Taylor asks how the merchant should assess its reputation, identify its creator, and verify whether the user actually authorized the purchase. He lists several early standards, including Google’s A2A, Visa’s Trusted Agent Protocol, and identity standards being built by FIDO. He also names Natural Payments, Skyfire, and A-comm as companies planting flags in this area by managing the part of the agent commerce workflow where money actually moves.
The third is lower-cost inference and compute. Taylor cites Brex as saying that companies add their first open compute supplier within five months of first incurring OpenAI or Anthropic API charges. Five months, he writes. One moment a company is paying model labs directly for API usage; soon after, it is comparing Together AI, Fireworks, and Baseten, deciding where workloads should run, and asking finance how to hedge cost exposure. He says AI has created a software supply chain and then started building a capital stack under it. Nvidia and Wall Street, he writes, are trying to mobilize $500 billion around that layer.
These jobs are larger than any single feature because they cut across product, data, identity, and money. Once that happens, Taylor says, the question becomes which party the customer trusts to sit above all of them.
He sums up the shift in simple terms: customers do not want your agent. They want their agent to enter your product.
In his view, most business customers will end up running some kind of control plane or shell, a single place to orchestrate the agents they own across the tools they use. A CFO’s finance agent, an engineer’s coding agent, and an operations team’s procurement agent would all be budgeted, authorized, and observed from a layer above the products themselves. Consumers, he says, will get their own version too, whether that ends up being Grokbot, Instinct, a new consumer app from OpenAI or Google, or a future Siri that Apple finally improves.
Three strategic paths from here
Taylor closes with three strategic options.
- Own orchestration: become the place where customers view, authorize, and manage every agent.
- Own a trusted control point: even if someone else owns the shell, identity, reputation, routing, procurement, and settlement can still hold value.
- Become the easiest product for important shells to call: CLI tools, APIs, and agent interfaces become the distribution layer. Taylor says this may be the right answer for more companies than many founders think.
He adds that systems of record such as Salesforce and core banking platforms are unlikely to give up their existing wedge easily. A system of record can become the shell, a specialist can own a control point, and a product can distribute through all of them. The prize, he says, may be the control plane itself, or it may be the one indispensable component every control plane needs.
Returning to the three founders he spoke with, Taylor says the answer kept coming back to the same thing: the job to be done. Positioning follows product.
Find the jobs AI creates next to the value you already deliver. Build interfaces for those jobs. Then decide how they spread. A company can own the shell, own a trusted control point inside it, or become a product that every shell can call elegantly.
That, in his framing, is what reinvention means. Start by using AI to run yourself. Use it to do old jobs better. Then climb toward the jobs that did not exist until models created new problems for customers.
Taylor’s closing point is that Ramp did not reach a $44 billion valuation simply because it built the best expense management product. It got there because investors believe AI will keep creating new demand next to its core business, and because Ramp has the speed to capture it.
He also adds a footnote on growth. By financial-company standards, he writes, Ramp’s expansion has been extraordinary: TPV was up 170% year over year through March, the fastest pace in three years, while the business had already scaled 20x.