The broader implications: why context packs change everything

The end of context switching

Context switching costs the average knowledge worker 25 minutes per interruption. With MCP servers and packs of contexts, this overhead disappears entirely.

You're no longer learning interfaces but having conversations about your work. The cognitive load shifts from remembering where buttons are to clearly articulating what you want to accomplish.

Complex workflows become simple?

Today only power users can create complex automations between applications. They need to understand APIs, write scripts, or configure integration platforms. MCP democratizes this power.

Any user can now create sophisticated workflows simply by describing them in natural language. "When a high-value lead comes in, research them on LinkedIn, schedule a demo, and brief my team" becomes a simple conversation, not a complex automation project. With context packs this can also be bundled into a neat package and delegated over to the team or even autonomous AI Agents.

Traditional UIs are only for admins, users converse

This is the beginning of the end for traditional application interfaces. When you can accomplish everything through conversation, why would you need buttons, menus, and forms?

We're entering an era where technology fundamentally transforms how we interact with software applications. New employees won't need training to learn complex application interfaces because they can simply communicate their needs through natural conversation. This allows applications to focus purely on functionality rather than UI/UX considerations. Manual workflows that once required navigating through multiple systems become streamlined through simple conversational commands. The constant context switching between different applications, which costs knowledge workers significant time and cognitive energy, disappears entirely as everything happens within one unified interface.

This transformation represents a complete reimagining of how humans interact with digital tools. Instead of people adapting to software interfaces, the software adapts to natural human communication patterns.

MCPs will increase the surface area of possibilities

As more applications expose their functionality through MCP servers, the value of this approach grows exponentially. Each new MCP server creates new possibilities for combining tools in ways that were previously impossible.

When your design tools, project management system, and communication platforms all have MCP servers, you can orchestrate workflows that span all three with a single request. The combinatorial possibilities are limitless.

MCP servers typically run locally or in controlled environments, meaning your data doesn't need to leave your infrastructure to enable these powerful integrations. This addresses one of the biggest concerns about AI-powered workflows: data security.

Your packs of contexts operate within your security perimeter, accessing only the data and functions you explicitly authorize. The AI acts as an interface layer, not a data intermediary. Ctxpack allows you to deploy into your own infrastructure and self-host the necessary integration points between your team of humans (and agents) and the third parties that they want to integrate with.

So what's up and coming (does anyone know?)

This is happening right now. Today in 2025, early MCP servers exist for major platforms like Salesforce, GitHub, and Google Workspace. Power users and developers are building custom context packs by combining locally run MCP tools and remote servers.

Over the next year or two, major software vendors will release official MCP servers. Context packs will become standard for common roles and workflows. From 2026 to 2028, traditional application interfaces will start feeling outdated. New applications will launch with MCP-first approaches, treating traditional UIs as legacy features. After 2028, the chat interface will become the primary way most knowledge workers interact with software. Traditional application UIs will exist mainly for specialized use cases.

Or not, who knows...