Agentic AI: Breaking down barriers to transactions and trust

Agentic AI can simplify almost any digital task, from booking travel and scheduling meetings to ordering supplies and updating code. There’s a lot of potential, but here’s the catch: Today’s agentic tools are stuck in their own lanes.

You often need one platform, one set of logins, and separate subscriptions for each system, defeating the promise of “automation everywhere.” These silos aren’t just annoying; they limit what agents can deliver and ultimately make the end user do more manual work.

Agents are great at understanding what needs to be done, but when the time comes to take action—especially across platforms or where payments are involved—they hit a wall. Imagine asking a virtual assistant to book flights: it understands your request, but suddenly you’re dragged back in to sign up, pay, and verify your identity. That’s not automation. That’s just another to-do on your list.

To change this, we need three things:

1. Fewer walls between platforms (break down silos)
2. Smarter ways for agents to find the services they need (service discovery)
3. Trusted, seamless payments and identity checks handled in the background (automated transactions)

A brief look back: The first wave of agentic software

In the early 2000s, Tim Berners-Lee and the W3C’s “Semantic Web” project aimed to empower agents—not just humans—to directly use web content and applications. This vision led to efforts like structured metadata (RDF, OWL), XML-based API standards (SOAP, ebXML), and agent communication protocols (FIPA). These initiatives relied on symbolic AI, using formal rules and logic to enable automation.

Despite a decade of work, real-world adoption fell short. Symbolic AI struggled with flexibility and natural language understanding, making it hard for agents to reliably use web services. Creating machine-readable service definitions was labor-intensive, and inference engines couldn’t deliver on their promise.

Fast forward to today, and the game has changed. Large Language Models (LLMs) and protocols like Model Context Protocol (MCP) have largely solved interoperability challenges by enabling agents to use natural language to access APIs. However, two major hurdles remain: automated service discovery and seamless paid access. Even with LLMs, account creation and payments still require manual intervention that goes far beyond simple and much-needed approval, keeping silos and friction intact.

Why are AI agents still so siloed?

Most AI agents work in isolation. A project management bot can update your team tasks, but can’t order snacks for the office. Your AI travel planner can find flights, but can’t book the hotel with your corporate discount unless it’s already integrated. Users end up handling each “connected” service separately, each with its own accounts, rules, and limits.

There are tools in the works to help agents discover what’s out there, like growing registries of digital services (for example, Pulse). But letting agents “see” more options doesn’t matter if they still can’t actually use or pay for them. Discovery isn’t enough. Access is crucial.

The big missing piece: Letting agents pay and prove who they are

Here’s where most agentic systems fall short. Agents need to sign up, prove who they represent, agree to terms, and pay for services—automatically. The world’s best automation engine is useless if a payment screen halts the workflow.

Let’s get specific. An agent should be able to:

• Prove that it’s acting on a user’s behalf
• Accept terms with another service
• Make secure payments (and log what they were for, and how much)

And all of this should happen without breaking the user’s focus or trust.

What’s next for agentic? Two paths to smarter transactions

The industry is circling two main models—one decentralized and open, the other more like today’s big app stores.

1. The decentralized path: Smart contracts & digital IDs

Blockchain-based smart contracts offer a way for agents to make (and keep) agreements automatically. They’re programmable, reliable, and are supported well by decentralized identifiers (DIDs), so agents can prove who they represent and transact, all without a central authority. This approach could open up a marketplace where any agent can connect with any service. No gatekeepers, and no hoops to jump through.

2. The app store model: Centralized simplicity, but with limitations

Think about how you buy apps on your phone: there’s a single provider who handles your ID, takes your money, and makes sure you get what you paid for – including for subscriptions and add-ons. An “agentic app store” could do the same for agents to access AI services. But there’s a trade-off. Those store owners get to decide who plays and what gets paid. That simplicity comes with fewer choices and makes businesses even more dependent on a few large vendors.

How should business leaders pursue agentic AI?

There’s a clear fork in the road. Do we build a future where agents can connect securely and automatically to any service? Or do we accept a model where a few platforms control which connections are possible?

If you want agentic AI that truly works for your organization and your customers, you need to advocate for solutions that provide both ease of use and the flexibility to grow. Don’t settle for walled gardens if your goals require open, adaptable systems.

Take action:

• Push vendors to support open standards for discovery, authentication, and payment
• Explore decentralized options that give your business more control (and choice)
• Stay informed as new standards emerge. You don’t want to lock into yesterday’s limitations

By championing transparency, openness, and real automation, you’ll help shape a digital future where agentic AI delivers on its promise: less friction, more value, and truly empowered users.

Ready to explore how agentic AI can help achieve your goals? Learn more about AI and ML solutions at Modus Create →