Why A Dedicated Service Layer Is Critical
Evan Schwartz is the Chief Operations Officer at AMCS Group.
AI is evolving beyond simple text-based interactions with humans. The next stage, agentic AI, envisions self-directed systems capable of spawning specialized agents that carry out tasks autonomously. While generative AI provides an excellent framework for humanlike interaction, agentic AI demands a robust back-end ecosystem where AI-to-AI communication can thrive without the overhead of human-friendly protocols.
Generative AI was revolutionary because it understood conversational context and nuance. However, it was designed for humans. When AI interacts with other AI systems or data sources that don’t require a human-friendly interface, a more streamlined approach becomes essential.
The Problem With Domain-Specific Agents
We’re seeing a surge in AI products that excel in narrowly defined tasks—healthcare billing, insurance claims, schedule management, personal digital assistants, etc. The logic behind specialized agents is sound: You train them on a specific dataset, optimize them and reduce error rates. However, these specialized agents are siloed and rarely “talk” to other systems easily.
That siloing stands in direct opposition to the promise of agentic AI: frictionless data exchange and task automation on a global scale. Specialized agents should operate at Layer 2 or Layer 3. Layer 1, or agentic AI, needs to be creative, plugged in and capable of accessing data. The deeper layers—Layer 2 and Layer 3—serve as specialized agents designed for specific functions. These layers function as service layers within the overall architecture.
Toward A Dedicated AI Communication Layer
Agentic AI needs an underlying framework akin to the “WSDL in SOAP” for web services but reimagined for autonomous AI consumption. Let’s call this hypothetical technology the “Axiom Protocol.” Axiom would feature:
• Discovery Mechanism: A distributed ledger (e.g., a blockchain) where data or service providers register their endpoints, usage rules and cost structures. Each entry is both human-verifiable and machine-readable, promoting trust and transparency.
• Service Contract Schema: A lightweight, binary or semi-structured format that AI can read, interpret and transform without the overhead of verbose JSON or XML. Contract definitions that specify data requirements, allowable methods, response formats and cost in tokens.
• Tokenized Payment System: Each service endpoint defines a microtransaction cost for usage, payable in digital tokens. When a coordinator AI spawns an agent, it loads that agent’s wallet with enough tokens to pay for the required service calls.
• Dynamic Code Generation: Agentic AI can automatically generate code stubs or specialized micro-agents for each service, bridging the gap between service contracts and the AI’s own process flows. After the task is completed, ephemeral agents can be terminated, freeing compute resources.
Coordinator AI And JITA (Just-In-Time Agents)
An efficient agentic AI ecosystem might include a higher-level “coordinator AI” that manages tasks. For any given request, the coordinator AI:
1. Checks the Axiom Protocol’s ledger for compatible services.
2. Spawns specialized agents, providing them with permissions and tokens.
3. Monitors progress as the agents execute tasks, calling on whichever services they need.
4. Aggregates and returns results before decommissioning the micro-agents.
This is where the human interface exists and where the work done in generative AI is maximized: the human interface point.
AI Evolution Through Service Usage
Continuous feedback and evolution are among the biggest perks of a system like this. Each time an AI agent uses a service, it can learn more about its constraints, performance or new ways to optimize calls. Over time, it can suggest updates to the service contracts, leading to a self-improving ecosystem where interface definitions refine themselves in real time.
While this model introduces novel attack surfaces—such as malicious or fraudulent endpoints on the ledger—it can also streamline compliance. Every transaction is logged on a blockchain, enabling robust auditing. Providers can enforce access controls, while AI can confirm service reputation through on-ledger trust scores.
Embracing The Infrastructure Era
Agentic AI will never reach its full potential through isolated pockets of specialized systems. We need an infrastructural backbone built around dedicated AI communication protocols, decentralized discovery and token-based transactions. By shifting our focus from siloed mini-agents to a cohesive, global network, we unlock AI’s real promise: universal, automated collaboration that can evolve at the speed of intelligence.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?