Vision and Roadmap

Multi-agent systems and assistants represent the next step in artificial intelligence, creating a potential path toward AGI. These systems often consist of a central coordinating agent (the "assistant") that orchestrates interactions between users and multiple specialized "agents." Each agent performs specific tasks and communicates via shared protocols ranging from simple API calls to sophisticated emerging standards like MCP or A2A.

The Current Landscape

Today, most agentic systems are built by single companies (OpenAI, Manus.im, Agent.ai) who are either free or charge end users monthly subscriptions. More importantly they run all their agents β€œin-house”, paying for infrastructure and services themselves. While protocols exist to build interoperable networks of agents, developers have little incentive to interconnect their systems due to a critical missing piece: easy payment mechanisms that allow agents to pay each other.

Agents vs. Assistants

Image without caption
To understand the architecture more clearly:
  • Agents function like APIs – backend, non-user-facing systems that receive requests from assistants or other agents. Their sophistication ranges from simple stateless "tools" to autonomous, self-learning LLM-powered systems.
  • Assistants function like UIs – frontend, user-facing software that receives user prompts, understands intent, identifies appropriate agents, creates plans, delegates tasks, and delivers results back to users.
This distinction mirrors the API revolution that accelerated in 2000 and gave birth to companies like Zapier.com.

The Problem with Payments

The user experience with assistant aggregators (like ChatGPT that gives user access to multiple CustomGPTs ) is already broken due to payment issues. Users must sign up for multiple 3rd party services, manage numerous subscriptions, and navigate complex payment flows just to try basic functionality. This problem becomes exponentially worse when scaling to true agent-to-agent systems with multiple layers of delegation and specialization. This is because traditional payment methods are fundamentally incompatible with autonomous systems because they:
  • Require human initiation of transactions, specially with new parties
  • Don't support microtransactions cost efficiently
  • Have complex geographic and regulatory constraints
  • Lack programmability for machine-to-machine payments

Open Innovation

Our vision contrasts sharply with closed ecosystems where only big tech companies can innovate. By equipping agents with crypto wallets, we enable permissionless innovation where anyone – from solo no-code developers to startups to enterprises – can build assistants that access the same ecosystem of specialized agents.
Instead of forcing every company to recreate every capability, we create an open marketplace where the best specialized agents can thrive regardless of who created them, democratizing AI development.

The Economic Model

The tokenomics are straightforward:
  1. Users buy token using any payment methods storing them in their wallet but - ideally - dont need to understand more manage cryptocurrency directly
  1. Users grant their assistants limited, secure permission to access their wallet to make payments and transactions
  1. Agents set their own pricing (per token, per use, or other models) and assistants (and later other agents) automatically pay as needed, without requiring user approval for each transaction
Its noteworthy that cryptocurrency enables us to go beyond traditional API payment wrappers. By leveraging proven DeFi concepts to create payment mechanisms that are more flexible, secure, and cost-efficient than fiat solutions.
Image without caption
Furthermore wallets enable powerful new economic models impossible with fiat. Wallets can be owned and managed by individual creators/owners or can be distributed across token holders (as pioneered in Virtuals.io) as a result joint governance and revenue incentives can emerge naturally. Agent creators and users can collaborate on building better agents through aligned incentives.
This creates entirely new business models for AI development that align incentives between creators and users in ways that credit card payments simply cannot support.
Grindery has developed the necessary secure, permissionless smart wallet technology that works with agents built using any framework or tool, on top of any LLM, and interacting with users through any client. This gives the system secure financial autonomy while maintaining user control.

The Currency of AI

Imagine planning a Hawaiian vacation with your AI travel assistant. One message sets in motion a network of specialized agents for flights, hotels, and activities - each receiving micro-payments from your pre-allocated funds without requiring your approval for each transaction.
Our ecosystem starts with simple virtual credits but will evolve to leverage GX token's unique capabilities:
  1. GX already functions as a universal gas token across multiple blockchains, eliminating the need for users to hold different tokens for different networks.
  1. GX will power agent-to-agent transactions as a cost-effective Layer 3 solution, enabling frictionless microtransactions where traditional payment systems fail.
  1. Popular agents can issue their own tokens backed by GX, creating specialized economic models with shared ownership between creators and users.
By providing this fluid payment layer, GX becomes the essential currency for AI - enabling an open marketplace where the best agents thrive regardless of who created them.

Building the Minimum Viable Economy

Our long term goal is to create an agent economy that accelerates the development of more capable, more specialized, and more useful AI systems – ultimately benefiting both creators and users. Our immediate objective is not just to build technology but to create what we call a Minimum Viable Economy (MVE) - a functioning agent ecosystem with real economic activity that can grow organically without our constant intervention.
The MVE strategy consists of these key elements:
  1. Focus on Speed and Accessibility: We're using no-code tools like FlowXO to build assistants and agents rapidly. This approach allows us to demonstrate the concept faster and makes the technology accessible to more developers.
  1. Start with Telegram: We'll deploy our first assistants through Telegram bots where we already have users, enabling us to quickly demonstrate the complete flow from user to assistant to agent with real economic transactions.
  1. Demonstrate Component Functionality: Through a series of progressive demos, we'll show each component working - from building assistants to processing payments to cross-agent coordination.
  1. Enable Others to Build and Connect: We'll create documentation and tools that make it easy for other developers to build their own agents and assistants and connect them to our payment ecosystem.
  1. Target Initial Use Cases: Rather than trying to build a comprehensive system immediately, we'll focus on specific use cases that demonstrate the value of agent-to-agent payments, such as AI-driven design assistants that can access specialized tools and services.
The MVE is achieved when transaction volume increases organicall, new agents and assistants join the ecosystem without our direct involvement and - as a result - the ecosystem grows through its own incentive structure
This initial economy may be small in monetary terms, but it will prove its value - that agent-to-agent payments create a more vibrant, innovative AI ecosystem than closed platforms can achieve. From there, we can expand to incorporate more advanced protocols, specialized AI frameworks, and enterprise use cases.
By focusing on building this MVE as quickly as possible, we're providing a concrete demonstration of our larger vision - creating the currency for AI and enabling a thriving ecosystem of specialized AI capabilities that can be composed into increasingly powerful solutions.

Powered by Notaku