Enter AX: The Agentic Experience Beyond User Experience
MCP, A2A, and agent-native payments are reshaping how software gets discovered, negotiated, and consumed.
Your app users are no longer only people. They are AI agents that interact through protocols instead of pixels.
The deeper shift is that the customer itself is changing. The customer is no longer a person making slow decisions through dashboards and pricing pages. Increasingly, it is an agent choosing services instantly at machine speed.
The Companies that optimize for this transition early will win the same way TikTok won the last decade by engineering a doom-scrolling loop so addictive that users spend hours in a single session without ever making a conscious decision to stay.
Welcome to AX, the Agentic Experience.
This is not about building better UIs. It is about designing for an entirely new class of consumers who never read, never click, and never wait. Agents discover your service through search or MCP servers, negotiate through protocols, and move on. If your product is not designed for that interaction, you become invisible to an entire class of users.
I have only been in the industry for five years, and even in that short span I watched product teams obsess over a single question: how do we build the best experience for the person sitting at a screen? Every company benchmarked against the best UX in the room. We’ve already seen with Blinkit and Zepto, racing with new UX experiments all the time.
The platforms that understood UX on this level won the last decade. The same shift is happening now, but the user this time is an AI.
The Protocol Stack Reshaping Everything
A protocol stack is emerging and growing faster than any previous adoption curve I have tracked. Three layers are converging at once: tools, communication, and interface.
MCP: The Tool Layer
MCP, the Model Context Protocol (modelcontextprotocol.io), started at Anthropic and is now under the Linux Foundation. It has crossed 110 million SDK downloads per month, outpacing React in the same timeframe of its early adoption. That is not a niche experiment. That is infrastructure forming in real time.
MCP gives every agent a standard way to discover and use your tools. Build it once, and any compliant agent can connect automatically. No custom integrations for each provider. The real shift is that your service becomes addressable by any agent on the network. Integration costs largely disappear. The bottleneck now moves from wiring things together to how well you define your tools.
A2A: The Communication Layer
Above MCP sits A2A (a2a-protocol.org), Google’s Agent-to-Agent protocol backed by over fifty partners, including Salesforce, SAP, and MongoDB. If MCP connects agents to tools, A2A connects agents to each other. Think of it as TCP for the agent internet. It is a standard way for autonomous systems to discover, negotiate, and coordinate.
Originally developed by Google and now donated to the Linux Foundation, A2A provides a standard language for agent interoperability. An agent built with one framework can discover, negotiate with, and delegate tasks to an agent built on another. The protocol handles authentication, capability negotiation, and task status tracking out of the box.
This matters because no single agent will solve everything. The future is a mesh of specialized agents. One for data analysis, another for customer support, another for payment processing. All coordinated through a standard communication layer. A2A is the protocol that makes that mesh possible.
But if these agents can negotiate a service, how do they get paid? We’ll get to that in a bit.
The Interface Layer: A2UI and AG-UI
Then there is the interface layer. A2UI (a2ui.org) lets agents describe UI components they want rendered, which makes it more interactive than a text-based chat. AG-UI(ag-ui.com) is the runtime that renders them.
You have seen this in action without realizing it. When a customer support agent detects a refund request and dynamically generates a form with amount fields, reason dropdowns, and a submit button instead of typing back and forth, that is A2UI at work. The protocol lets agents decide that a specific task is better handled through a clickable interface than a text conversation.
The Payments Problem Nobody Solved Yet
The unresolved piece is payments. How does an agent pay for a service?
The traditional model assumes a human with a credit card, an API key, and a monthly subscription. That model collapses when agents are making autonomous, real-time purchasing decisions. The industry is responding with protocols like x402and MPP, designed for agent-to-service payments at machine speed. Stripe recently launched Link for Agents, a payment product explicitly designed for autonomous spending.
In India, Razorpay has been making significant moves in this space. At FTX 2026, they launched the Razorpay Agent Studio, built on Anthropic’s Claude, enabling businesses to deploy AI agents for recovering abandoned carts, resolving disputes, and fixing failed payments. They also introduced UPI Reserve Pay, which lets users pre-approve spending limits so AI agents can trigger payments seamlessly within conversations.
The deeper implication is what keeps me thinking. When agents can discover, negotiate, and pay for services without human approval, the economics of SaaS change fundamentally. When services are chosen instantly at machine speed, traditional SaaS pricing models begin to feel outdated.
What I Am Watching Next
The pieces are moving fast now. Protocols are converging, payment rails are being built, and we are already seeing companies like Swiggy experiment with MCP integrations. I think this is the beginning of user-facing products preparing for AI agents as actual consumers of software.
What feels most important is how discovery itself is changing. The internet was built around SEO, clicks, and human attention, but agents do not behave like humans. Agentic Experience is not a feature to add. It is a new design discipline, and it requires rethinking how your service presents itself. They care about structured access, reliability, latency, and execution quality, which means the next moat may come from being agent-friendly, not just user-friendly.
If you found this interesting, you can subscribe with your email below. I’ll be sharing more deep dives and breakdowns on infrastructure, AI systems, cloud architecture, and the kind of tech that usually stays behind the scenes.