We are witnessing a profound shift in productivity and development applications towards conversational interfaces. Whether we use ChatGPT, Claude or Gemini, a simple text can trigger in-depth research, design applications, or orchestrate complex tasks directly on our systems and browsers.
But the next step in agentic AI lies where users already spend their time. The messaging application is the gateway to the city of agents.
Inspired by China, built around work
In China, the Super App has brought digital services together within a messaging environment. With Weixin Mini Programs, services for shopping, travel and daily life become accessible without leaving that environment. Messaging is a natural gateway. It is where we exchange, contact one another, and where the events of the day take shape.
My intuition is that the Western equivalent of this Super App will be built within the enterprise. Around work and the interactions that make it happen. We already see activities converging towards environments where we can discuss, find information, design, code and act.
This does not exclude a similar evolution in everyday life. But in the enterprise, there is a constant need to increase productivity, to be more competitive and improve commercial performance. Workers are asked to be increasingly effective, while information remains dispersed across applications, documentation and Excel files. Even digital native companies still experience this fragmentation.
This creates a need to bring information together. Within a company, good practices can be shared and adopted collectively. The need for productivity, often driven by top management, pushes this convergence towards AI environments that bring information and capabilities into the same space.
There is also the push towards AI transformation, towards companies that become more AI native. For me, this is where the force of transformation is stronger than in everyday consumer use. This is why I see the enterprise as a natural place for the Western Super App to emerge.
The Western equivalent of the Super App is emerging in the enterprise, where agents bring work towards us.
The gravity of AI platforms
I see these applications as celestial bodies with a strong gravitational pull. Like black holes, they progressively draw in different domains: software programming, UX and UI design, mathematics, then physics and biology. Platforms and ecosystems extend their reach, even into hardware.
Interaction remains at the centre. It moves from text to voice and becomes increasingly natural. WhatsApp already brings together our written exchanges and our calls. Why would this convergence not also happen in AI environments, where conversation becomes a way to work? I experience this evolution myself, more and more naturally.
Work is pulled towards you
ChatGPT, Claude or Copilot progressively position themselves between us and our information systems. Messaging and email. Work organisation tools, such as Jira or Microsoft Loop. Wikis, files, documentation websites, architecture repositories and codebases.
The sources are disparate, but their information and capabilities can be brought into the same environment. Work is pulled towards you. You pull in the data, but also the capabilities you need to act. The AI environment becomes a centre for organising work, and agents organise that work for us.
Fill in your timesheet. Find the training you need to take. Prepare your expense report. Get a report on upcoming activities. These are the uses I see this environment moving towards, through available integrations and through computer use, which also makes it possible to interact with application interfaces.
A plugin is a functional extension. An MCP server gives access to tools. A skill brings a way to carry out an activity. These mechanisms extend the work environment. We extend planet ChatGPT with new capabilities. The mechanisms differ across platforms, but the trajectory I observe is the same: bring more work into this space.
The source applications remain places where data and business functions reside. It becomes less necessary to open them for every operation. They retain their value when an activity requires a specific environment: a claims management view, an architecture graph, a 3D visualisation. Some activities need their own representations.
With agents, fleets of agents or swarms, part of the orchestration can also happen automatically, in the background. As these capabilities progress, you stay in your environment to do your work, or let agents do it. This is how I see these applications becoming Super Apps.
Accumulating context and the compounding effect
As we move along this trajectory, the AI agent accumulates a detailed and continuous understanding of our life and work context. It develops its own memory, becoming a mirror of our public and professional activity. This creates a compounding effect: the more context the agent accumulates, the more relevant and effective it becomes.
The real leap happens when our networks of peers and collaborators share this dynamic. Collaboration becomes fundamentally more intelligent because intelligence is brought into the interactions themselves. This is the principle of workshops, which remain the ultimate form for designing, organising and architecting a system: a space where you pick up the phone, where you instantly look for the right expertise to solve a problem. Messaging, whether on Slack, Teams, WhatsApp or LinkedIn, rests on that same philosophy: initiate contact, get an answer, or find the entity capable of providing a solution.
The longer-term goal is to deploy ambient intelligence, an invisible cognitive layer in the background, capable of understanding who is needed to accomplish a specific task and proactively proposing the right agent’s intervention in the conversation.
The ecosystem battle and the need for interoperability
We already see the beginnings of this integration. Cursor lets us launch agents from Slack. Claude can also be called upon in Slack, while Microsoft integrates Copilot and agents into Teams. ChatGPT can be extended through MCP apps, depending on the plan and configuration. Each platform provider is trying to build its universal application, its Super App, to lock in its ecosystem.
Yet the future cannot belong to closed silos. Messaging and social networks hold a central place in daily exchanges: Discord, WhatsApp, X. They are also where users meet and organise their activities. The challenge is to bring the cognitive layer to where users are, in an open way.
We need an interconnected exchange network: a universal cognitive layer where agents can be called upon independently of their platform of origin. Imagine a flow where, from ChatGPT, you call on an agent hosted on Claude or Gemini. A world where your logistics sourcing agent collaborates autonomously with your personal planning agent to confirm a delivery, or organise a complex event with friends, managing everyone’s preferences and constraints in the background.
[ User interface: WhatsApp, Slack, Discord ]
│
▼
[ Cognitive exchange layer ]
│
┌─────────────┼─────────────┐
▼ ▼ ▼
[ GPT agents ] [ Claude ] [ Gemini ]Mellow Mesh: Opening up cognitive spaces
It is precisely to address this fragmentation that I developed Mellow Mesh, an experimental open source initiative written in Rust. Its goal is to provide decentralised, autonomous middleware where agents living on different platforms can communicate and collaborate.
Mellow Mesh lets agents on platforms such as Claude, ChatGPT and Codex, Gemini or Cursor communicate. They can discuss together and pass work to one another. This cross-platform communication also includes interagent messaging, so agents can exchange directly and coordinate their activities. Multi-agent and cross-platform communication already work in my prototype.
The architecture rests on several foundations:
- The open standard OKF (Open Knowledge Format), published by Google Cloud. It organises knowledge in Markdown files with YAML metadata, readable by humans and agents, and formalises Andrej Karpathy’s LLM Wiki pattern.
- The concept of a second brain, popularised by tools such as Obsidian or Notion, for managing and persisting memory.
- Knowledge graph modelling, as illustrated by Neo4j.
- With this approach, the system can learn and evolve in context, providing a standardised exchange hub through Mellow Mesh Skills for AI systems from different providers.
Perspectives
Any organisation or individual using AI at an advanced level will sooner or later need a transversal, platform-agnostic intelligence layer. Intelligence must be ambient, contextual and mobile.
The future belongs to the enterprise that knows how to build on the expertise of the people who make it up. Invest in it today.
