In March I argued that the token economy is long distance minutes. That the infrastructure shift was coming and the metering model would become indefensible the same way per-minute billing did once VoIP made calls structurally free.
I was watching hardware and open weights. Local inference closing the capability gap. The tollbooth losing its structural advantage over time.
I missed a faster signal. The tollbooth operators are moving the booth themselves.
The data
Among the top 500 SaaS and AI companies with transparent pricing, there were more than 1,800 pricing changes in 2025 alone — an average of 3.6 per company. That is not an industry iterating. That is an industry in structural panic. ref: SaaS Mag
The specific move is from seat-based to usage-based and credit-based pricing. Out of 500 companies in the PricingSaaS 500 Index, 79 now offer a credit model, up from 35 at the end of 2024 — a 126% year-over-year increase. Among the new additions are Figma, HubSpot, and Salesforce. ref: Growthunhinged
The reason is not complicated. When an AI agent can perform the work of five junior employees, charging per seat punishes the customer for becoming efficient and punishes the vendor for delivering that efficiency. The seat model’s math stopped working. So the meter moved. ref: Medium
The market understood what that meant before most operators did. In February 2026, approximately $285 billion in market value vanished from software stocks in a single trading session. ServiceNow dropped 7%. Salesforce fell 7%. Intuit plummeted 11%. Thomson Reuters collapsed nearly 16%. The catalyst wasn’t a recession or a regulatory crackdown. It was investors recognizing that the per-seat subscription model powering a $300 billion industry for two decades is breaking. ref: Orbilon Technologies
Satya Nadella said it plainly: “The notion that business applications exist; that’s probably where they will all collapse in the agent era. They are essentially CRUD databases with a bunch of business logic. The business logic is all going to these agents.” ref: The Register
What the pivot actually means
The industry framing is customer alignment. Pay for what you use. Fair pricing. Value-based.
It isn’t. It is an acknowledgment that the unit of value can no longer be a human seat when agents do the work. So vendors are reaching for consumption as the new moat. Pay per token. Per credit. Per agent action. Per resolved ticket.
The tollbooth didn’t disappear. It moved to a different part of the road.
The architecture still serves the platform. The customer is still downstream of the vendor’s incentives. The dependency is structurally identical. Only the measurement changed.
Most teams building credit models admit they are “a workaround, not a long-term answer.” They are buying time. The question is what the time is for. Nxcode
When we built Orderero 2 years ago…
A few years ago we built Orderero. The premise was that chatops would transform how facilities teams work — requests, routing, resolution, all through conversational interfaces. We believed it. We shipped it.
AI generates that now. Not for one company. For companies all over the world, on demand, without a platform in the middle collecting rent. Functionality we spent months architecting, AI produces in an afternoon.
That is not a story about timing or execution. It is a demonstration of what is happening to entire SaaS categories simultaneously. Companies are moving from “one tool per task” to “one agent per outcome.” Instead of buying separate subscriptions for project management, CRM, email marketing, customer support, and analytics, businesses are building multi-step AI agents that handle entire workflows across systems autonomously. ref: Orbilon Technologies
The gut reaction from founders in this position is to become AI. Wrap the product in an LLM. Build enough surface area that customers feel switching cost before they realize the core value proposition no longer requires you.
But bolting intelligence onto platform-first architecture doesn’t change who the architecture serves. It makes the dependency stickier before the customer figures it out.
The platform-first architecture is the problem. AI features are cosmetic surgery on a structural issue.
Publicis Sapient is already reducing traditional SaaS licenses by approximately 50%, including major platforms like Adobe, substituting them with generative AI tools. The customers are drawing their own conclusions without waiting for the vendors. ref: Uncoveralpha
The gap we named
In March I described the architecture that matches a different answer. Local inference for everything personal. Stateless cloud for everything else. Zero trust as the operating model. Agent coordination instead of platform capture.
Then I wrote this:
“What’s missing is the runtime that ties it together. The open layer, owned by nobody, that a developer can stand up in an afternoon and that makes this model feel inevitable. Email had SMTP. Version control had Git. This needs its equivalent.”
We were mid-build when I wrote that.
Principal Agent Protocol
PAP is the request layer for agentic systems. We dumped MCP for PAP and if you care about security you will too. Here’s why.
Before your agents execute, PAP has already set the scope — cryptographically, at the request. A principal signs a mandate specifying the action, the disclosure scope, and the TTL. The 6-phase handshake enforces those bounds at every delegation step. A child request cannot exceed its parent’s scope. This is enforced in the protocol, not in a policy document, not in a terms of service.
Chrysalis is the execution layer. Agents register with verifiable DIDs, Ed25519-signed capability advertisements (aka passkeys), and vouch-based admission. A request without a valid PAP mandate is rejected before execution begins. An agent that requires more disclosure than the mandate permits is filtered before the session opens. Humans stay sovereign. Agents stay accountable. Data never leaves its declared scope.
The protocol stack underneath it is already standardized; WebAuthn, FIDO2, W3C DIDs, Verifiable Credentials, SD-JWT, OHTTP. We didn’t invent new cryptography. We built the runtime that ties existing pieces together into something a developer can actually deploy and be confident it works. We delivered 6 SDK’s so you don’t have to argue about language choices.
Email had SMTP. Version control had Git. HTTP has TLS. Agent coordination needs a neutral trust layer. A layer that no model provider can build without a conflict of interest, because the architecture that protects the principal is the architecture that eliminates the platform’s leverage.
That is what PAP is.
Where this lands
IDC predicts that by 2028, pure seat-based pricing will be obsolete, with 70% of software vendors refactoring their pricing strategies around new value metrics — consumption, outcomes, or organizational capability. The credit model buys time. It doesn’t change the underlying structure. ref: IDC
The answer isn’t a better meter. It’s an architecture where the tollbooth doesn’t exist because the dependency it would meter was never created.
The usage-based pivot confirms what the original post argued. Platform providers know the seat model is structurally over. They are reacting to it. Reaction is not architecture.
The organizations building for the post-tollbooth model now — local-first, agent-coordinated, zero trust by default — will be the infrastructure of the next web. The ones optimizing for credit consumption will find their model undermined by the same force that ended long distance: an infrastructure shift that makes the metering model unnecessary.
The credit economy is not the natural order of things. It is a pricing model that depends on dependency. That dependency has an expiration date.
PAP is open source. Read the protocol.
If you’re building agentic systems and want the enforcement layer that makes them production-grade, we can help with that too.