The Future of Agentic AI:
Insights from the A2A Summit NYC

Last week I attended the A2A Summit in New York City, joining thought leaders, researchers, and technology practitioners to discuss one of the most transformative trends emerging in enterprise architecture today: agentic AI.

The event was a clear signal that we’re at the beginning of a new era of intelligent, interconnected agents that will reshape how organizations build, integrate, and operate systems. From foundational definitions to practical implementations, each session contributed to the understanding of how “agentic” architectures are evolving — and where they’re headed next.

Here are my thoughts on the talks that I attended…

Making Connections: Navigating the Multi-Agent Landscape

Gary Olliffe, Research Director, Gartner

Gary opened with a foundational exploration of what AI agents are and how they fit within enterprise ecosystems.

According to Gartner, AI agents are autonomous or semi-autonomous software entities that use AI techniques to perceive, make decisions, take actions, and achieve goals in digital or physical environments. At their core, agents are services — they expose well-defined interfaces, operate within clear boundaries, and can exist on any platform.

He categorized agents into several types, including:

• Task automation agents — designed to execute predefined workflows.
• Information discovery agents — or what many now call “chat with your data” systems.
• Task partners — that assist users dynamically.
• Open tasking agents — capable of broad reasoning and problem-solving, though these remain challenging due to complexity and security concerns.

While agents can support a range of enterprise functions, Olliffe noted that 45% of active implementations focus on customer experience use cases, followed by customer service, R&D, and software operations.

His guidance for enterprises beginning their agent journey was clear:

“Build the simplest agent possible —
limit the surface area. Don’t boil the ocean.”

Gartner’s data shows that 58% of organizations are in exploration mode, 29% are in pilot, and only 8% are in production. It’s a fast-evolving market where early experimentation is key.

A particularly insightful part of the talk centered on the parallels between agents and APIs. Just as APIs required specifications, registries, ownership models, and lifecycle management to mature, so too will AI agents.

Today, A2A gateways already exist, while agent registries are still maturing — expected given the relative newness of the A2A specification.

He also forecasted a clear trajectory: by 2026, the industry will move from single-agent applications toward multi-agent systems — where agents call, coordinate with, and even supervise other agents. This evolution, he cautioned, must be guided by lessons from the microservices era: without governance and orchestration, complexity can easily spiral.

He closed with five practical recommendations for enterprise adoption:

1. Classify your agents — not all are created equal.
2. Separate interfaces from implementations.
3. Adopt platforms that both publish and consume agent interfaces.
4. Compose multi-agent systems based on standardized interfaces.
5. Govern your agents with the same rigor applied to APIs and services.

Keynote Panel: AI Agents in the Enterprise – Learning So Far

Featuring Rory Blundell (Gravitee), Cindi Howson (ThoughtSpot), Guy Duncan (OVO)
Moderated by Sara Fischer (Axios)

The keynote panel reinforced a common theme: start with friction points. Enterprises are under pressure to “do something with AI,” but the most successful initiatives begin by addressing where and why friction exists. Whether it’s accessing data, developing dashboards, or accelerating decision-making, the objective should always be to remove bottlenecks and empower users.

As one panelist summarized:

“The biggest challenge is the friction of deployment.
Start where it matters — solve real problems first.”

This focus on purposeful AI — not just novelty — resonated across sessions throughout the day.

How Blue Yonder is Building an Agent-Powered Supply Chain

Chris Burchett (SVP GenAI, Blue Yonder), Linus Håkansson (CPO, Gravitee)

Chris shared how his team at Blue Yonder is transforming supply chain management with agent-powered solutions, leveraging Gravitee’s Agent Mesh to help customers orchestrate, customize, and deploy AI agents within their ecosystems.

Their use case illustrated how A2A can simplify complex integrations and accelerate the adoption of AI-driven decision-making.

In observing their approach, it’s clear that a combination of Gravitee’s API-driven Agent Mesh with Solace Agent Mesh and Orchestrator could provide enterprises with the ideal foundation for building scalable, real-time, and interoperable multi-agent systems.

Gravitee offers strong connectivity and governance for agent APIs, while Solace provides high-performance data movement and dynamic orchestration — together enabling a robust and adaptive agentic ecosystem.

A Journey to Using A2A: Implementing Solace Agent Mesh

Jonathan Schabowsky, Field CTO, Solace

In my own session, the full house was a testament to the growing interest in practical, real-world agent implementations.

I began by explaining the Solace journey to building our initial in-house agentic AI system, drawing a parallel between the early evolution of monolithic applications and today’s early-stage agent designs. Many of the same pitfalls — limited interoperability, poor observability, and fragmented governance — risk being repeated.

The focus must be on future-proofing architectures with A2A standards that support interoperability across vendors, ecosystems, and domains.

During the session, an audience member asked an important question:

“Why would I interact with my data
using AI instead of a custom dashboard?”

My answer was straightforward: dashboards are static; agents are exploratory.

A dashboard reflects the questions a developer anticipated. A well-designed AI agent allows users to discover new insights dynamically — in natural language — without prior assumptions.

To demonstrate this, I showcased Solace Agent Mesh in action, streaming live FAA air traffic data in real time. Using Solace’s event-driven backbone, we integrated:

• FDPS (Flight Data Publication Service) — FAA’s system for distributing real-time enroute flight data such as aircraft positions, routing, and flight plans.
• STDDS (Surface Data Distribution System) — provides surface operations data from airports, including aircraft movements on taxiways, runways, and gates.

Both data streams were combined with contextual FAA documentation and CONOPS materials, allowing a language-model-powered agent to interpret queries naturally. Participants could ask questions such as:

• “Tell me all the inbound flights to JFK.”
• “Compare Flight X with Flight Y and report on their flight profiles.”
• “List all flights currently on approach into JFK.”

This demonstration highlighted the power of “chatting with your data” — dynamically exploring live operational data without needing specialized dashboards or technical knowledge.

Parting Thoughts

The A2A Summit made one thing clear: the agent era has arrived, and enterprises are rapidly moving from experimentation to orchestration. The conversations reinforced how critical governance, interoperability, and real-time data flow will be to ensuring this transition succeeds.

As multi-agent systems evolve toward an eventual “Internet of Agents,” it will be the orchestration frameworks, event-driven fabrics, and open specifications that determine how intelligently — and responsibly — they operate together.

At Solace, we’re excited to help shape that journey.