This week we dive into a subject that is becoming increasingly difficult to ignore: how the quiet emergence of standards, integrations and new design patterns in AI are reshaping everything beneath the surface, while the headlines stay focused on the shiny stuff.
It is easy to be distracted by product launches and personality-driven announcements in the AI space. But beneath the noise, something more foundational is happening. In this episode, Rob and Kev return to one of their early predictions: the role of MCP (Machine Control Protocol), and explore how it is quietly gaining traction, with potentially profound implications for how software will be built and used.
What triggered the conversation was news that Canva had launched a paid integration with Claude, powered by MCP. It might not sound like front-page news, but for those paying attention to the developer ecosystem, it is a major milestone.
MCP allows large language models to speak a universal “language” to other systems. Think of it as the USB-C of AI agents: a flexible bridge between LLMs and apps, systems or services. Claude's new integration with Canva marks the first commercially deployed use of MCP outside the development sandbox. It could be a glimpse of how software will work in the future.
Instead of fiddling with user interface components or learning a platform’s quirks, users issue natural language instructions. The AI interprets intent, connects to the MCP server, and makes the correct API calls in the background. That is not just a better interface. It is the beginning of a world where the interface disappears altogether and agents take care of the work.
This agentic model, where tasks are delegated, broken down and executed by different AI components, is becoming the real architectural story of the AI boom. While mainstream media continues to debate the intelligence of LLMs, the real change is structural.
The episode picks up on the now-notorious Apple research paper, which questioned the effectiveness of “thinking LLMs” for complex tasks. The public reacted with surprise to Apple’s claim that LLMs are merely predictive engines. But for those working closely with the technology, this was already well understood.
What the headlines missed, and Apple arguably ignored, is that agentic design sidesteps this limitation. By breaking big tasks into smaller ones and using specialised models to complete each, the overall system becomes more capable as complexity increases. This is precisely the kind of structure MCP supports.
So when Apple’s paper argued that LLMs collapse under complex reasoning, it also, perhaps unintentionally, validated why agentic design is gaining traction.
The bigger picture is not just about MCP. It is about what happens when AI becomes the default interface for software, and how that changes the nature of software itself.
Until recently, enterprise AI was focused on productivity. But we are now entering a new phase, where AI does not just help people work faster; it eliminates the need for the work to be done manually at all. Rob and Kev discuss how Microsoft’s recent layoffs, and a leaked internal memo indicating that AI usage would now be part of employee performance metrics, show that this is not about doing more. It is about doing less.
In this new reality, the apps we build must change. We no longer need tools that help users complete their tasks more efficiently. We need tools that understand the task and complete it for the user, with minimal input. The best software of tomorrow will not have better UX. It will have no UX.
This is the real disruption. The interface era is ending. The agent era is beginning.
For technology leaders, the conversation is shifting rapidly. LLMs were only the first stage. Agentic systems are the second. Standards like MCP will support what comes next.
It is easy to be distracted by the noise. But the quiet adoption of agent frameworks, the decline of rigid APIs, and the emergence of orchestration layers are the signs to watch.
If the interface disappears, what does that mean for your product? If agents are doing the work, what becomes of your team? And if AI becomes the default way to interact with services, how will your business remain accessible?
These are the questions we are exploring. Because the next thing is not on its way. It is already here.