Software design and its relation to Crypto AI
In general, software design follows 2 primary approaches: horizontal and vertical
• Horizontal design: modular, flexible, and reusable, providing general-purpose functionalities that can be applied across multiple domains. It focuses on interoperability and scalability, making it easy to integrate with various platforms.
• Vertical design: domain-specific and deeply integrated, optimising all components for a particular workflow. While highly optimised, they are less flexible and harder to repurpose beyond their intended scope.

These architectural approaches are now playing a crucial role in shaping the rapidly evolving landscape of AI agents in the crypto. AI agents - autonomous software entities powered by large language models (LLMs) that can perform tasks on behalf of users - have transformed from simple conversational interfaces (text) to potentially powerful tools for navigating the complex world of DeFi and crypto dApps.

Evolution of AI Agents
When I first covered AI Agents, popular agents such as GOAT had limited functionality and developer flexibility to them. These agents were generally LLM-based reply guys (usually on Crypto Twitter) and had closed-sourced codebases which made it hard to ascertain the AI capabilities. Since then, we have seen an influx of AI frameworks that focus more on open-sourced development and thus, development flexibility.

Current Web 2 frameworks like Agentverse, Anthropic's Claude Functions, and LangChain have gained popularity by offering modular components that developers can assemble for various use cases. These platforms allow developers to integrate multiple LLMs, connect to various blockchain networks, and provide generalised tools for transactions, data analysis, and user interaction. While powerful in their breadth, they often lack the specialised depth needed for specific crypto verticals.

Within frameworks, there is also a distinction between horizontal and vertical architecture. So far, most of the crypto AI frameworks we have seen are designed horizontally. These frameworks are generally LLM-agnostic, offer a wide range of modular on-chain tooling and are broadly applicable across crypto. However, the AI framework market has gotten rather saturated and stale and token prices have also come down quite considerably. A new form of innovation is required to reignite the spark within the Crypto AI sector.

The Case for Vertical Frameworks
Right before the recent market downturn, the hottest vertical within Crypto AI revolved around DeFAI agents, or AI DeFi assistants. DeFAI agents were envisioned to be autonomous agents that could carry out DeFi activities for you on a whim. They were pitched as AI agents that could optimise money management, from parking funds in the best yield farms (across multiple chains), to finding the next 100x gems, etc.

However, the current state of these DeFAI agents and applications are far from that ideal. At present, many of such applications are limited in functionality and are only able to perform simple tasks such as swaps, user-to-user transfers, very basic analysis of tokens etc. When tested, "Perp AI" might struggle to execute profitable trades amidst complex liquidity positions and macro conditions, or "OmniAI" might fail to execute multi-step transactions across different protocols and chains. The UX of such applications leaves more to be desired, often requiring users to manually verify and approve actions that should be automated. Of course, this niche sector is still in its infancy and it is understandable that they have their limitations.

Flipping the perspective on this, I think that DeFAI agents is a sub sector that would benefit greatly from vertical frameworks or agentic development. DeFAI is a more specific use-case as compared to the AI Agents. However, it is still a rather large domain by itself. Ideally, DeFAI agents will be able to carry out tasks across the DeFi verticals, from swaps, to liquidity provision and yield farming, lending and borrowing etc. However, this is a difficult ask for an agent built on a horizontal framework.

By leveraging vertical frameworks and development, developers will be able to create specialised DeFAI agents in this case, perhaps even DeFi vertical specific ones or protocol specific ones. For instance, a DeFAI agent that is specialised in yield farming, expertly navigating from one 1000% APR yield farm to the next. Or perhaps, a DeFAI agent built on top of a protocol such as Pendle, that allows users to optimise their yield by buying or selling Principal Tokens (PTs) and/or Yield Tokens (YTs) at opportune moments.

Where Does Vertical Design Lead Us?
The future of vertical AI agent architecture in crypto points toward specialised, interconnected systems that excel at specific tasks while collaborating to solve complex problems. This approach is already beginning to take shape in the form of agent swarms.

Swarms are somewhat the foundation of vertical agentic architecture. An agentic swarm refers to a decentralised network of AI agents that autonomously interact with blockchain ecosystems to execute tasks, make decisions, and coordinate activities. These agents leverage smart contracts, on-chain data, and decentralised protocols to operate collectively with minimal human intervention. In a vertical agentic architecture, it is likely that it features different AI agents with specialised roles. Swarms enable the distribution of tasks among these agents based on their expertise, ensuring optimised execution.

For example, a DeFi swarm might include:
• A risk assessment agent that evaluates protocol security and market conditions
• A yield optimisation agent that identifies the best farming opportunities
• A gas optimisation agent that determines the ideal transaction timings
• A transaction execution agent that interfaces directly with blockchain contracts

Each agent is specialised for its task, but they work together seamlessly to achieve the user's financial goals. This division of labor allows for much deeper expertise in each domain than a single, general-purpose agent could achieve.

The concept of specialised agent swarms extends beyond just DeFi applications. Robotics, which has been a largely unexplored sector within crypto, might benefit from vertical agentic architecture as well. While seemingly disconnected from cryptocurrency, robotic swarms represent a physical manifestation of the same architectural principles and could eventually integrate with blockchain systems for autonomous coordination and economic incentives. Robotic swarms can enable coordinated, autonomous systems that operate efficiently at scale. For instance, in logistics, robotic swarms can streamline warehouse automation and last-mile delivery, while in agriculture, they can optimise crop monitoring, precision spraying, and harvesting. Their scalability, fault tolerance, and adaptability make them a powerful foundation for the future of autonomous infrastructure, smart cities and much more.

Concluding thoughts
The shift toward vertical development in crypto AI represents progress in the sector. We are moving from general-purpose frameworks to specialised systems that excel at specific functions.

Virtuals Protocol has recently unveiled their Agent Commerce Protocol (ACP) which exemplifies this vertical approach. ACP is designed to enable secure, autonomous transactions between AI agents and humans through a specialised four-phase process—Request, Negotiation, Transaction, and Evaluation. What makes it unique is its focus on a specific vertical: agent-to-agent and agent-to-human commerce. Unlike general frameworks, ACP isn't trying to be everything to everyone; it's optimsing specifically for commercial interactions. The protocol includes specialised evaluator agents that verify transaction quality and earn a fee, creating a decentralised quality assurance market. Successfully tested in a multi-agent simulation (a lemonade stand business), ACP demonstrates its potential for fully autonomous digital economies, where AI agents can negotiate, execute, and regulate transactions without human intervention.

This is a big leap forward and I think more frameworks will be specialising in specific domains going forward. By moving from horizontal to vertical design, there is a possibility that a framework might lose some developers and users. However, this allows a framework to create a niche and a moat in a particular sector. For developers in the space, this suggests that focusing on solving specific problems rather than building general-purpose solutions may yield more valuable and differentiated products. For users, it means we can look forward to AI agents that truly understand our needs in specific domains rather than being jack-of-all-trades but masters of none.

I am excited to see what innovations Crypto AI continues to build as vertical development enables AI agents to move from interesting experiments to truly useful tools for navigating the complex world of decentralised finance and blockchain applications.

*Disclosure: the information provided in this article is for general informational purposes only and does not constitute professional nor investment advice.

Thank you for reading!