The first time I fired up an MCP (Model Context Protocol) server plugin, “Agent,” I was excited to see it registered in Claude Desktop but immediately annoyed by the errors that popped up. I didn’t expect a smooth experience in my encounter with the future of Agentic AI, but I found many configuration tweaks, clunky debugging tools, and broken dependencies along the way. It was a stark reminder that we’re in the early days, and there’s a lot of ground to cover before Agents become seamless collaborators.
Autonomy is Hard
Once I had a demo MCP-enabled Agent up and running under Claude Desktop, it became clear that autonomy would not be easy as the interplay between myself and the model continued. The handoffs felt awkward, with the model frequently pausing to ask for confirmation and review before continuing. At one point, it even obscured the information about the action it needed guidance on, content I needed to review behind a popup, like someone standing in front of the whiteboard while you’re trying to solve a problem.
It became clear that future Agents must generate custom UI/UX as part of their operation. Integrating a set of Agent plugins to solve unique problems with a generic UI doesn’t seem workable. Claude Desktop points in the right direction, but it won’t scale to more complex activities.
Seeing the Potential for SaaS Disruption
In some ways, these tools already shine in vertical applications like customer support or order management, where workflows are predictable and well-defined. We’ll continue to see SaaS providers using these AI tools to improve their products, but I don’t think Agents will be a SaaS killer soon.
For someone like me, and probably many other knowledge workers who must hop between projects and improvise solutions on the fly, Agents are more like a collection of loose parts than a fully formed toolkit. There is potential, but the vast majority of more “horizontal” applications will require a vast improvement in execution capabilities.
In addition to improving agents’ capabilities, entirely new and mature ecosystems need to be in place to support the effective deployment and operation of Agentic AI networks. The SaaS providers have invested in robust ecosystems, which will not easily be replaced or updated.
Bridging the Gap
That gap between tools that work for specialized tasks and those flexible enough for day-to-day experimentation and use might be the biggest challenge these systems face. The idea of “horizontal integration” sounds great in theory, but in practice, the tools still demand a lot of customization to fit into diverse workflows. Take monitoring, for example. Robust monitoring is a must-have for scaling these tools, especially in professional settings. During one experiment, I used MCP Inspector to visualize the capabilities of an MCP server. It was a step in the right direction, but even that felt like a workaround rather than an integral part of the system. For these tools to take off, monitoring and diagnostic capabilities must be baked in, not bolted on.
Is the Agentic Future Here Yet?
I see many challenges and opportunities in the next year. On one hand, we’re still in the experimentation phase, and the tools are rough around the edges. The key will be simplifying the interactions, refining the interfaces, and improving the tools while also delivering on the ecosystems, allowing users to monitor and manage large Agent networks easily. The dream is a set of AI tools that don’t just augment workflows in specific domains but empower users across the board to create and innovate on their own terms.