We Built Antares for a standard that didn't exist. It just arrived.

When we started building Antares two years ago, the question that shaped almost every architectural decision was what the second agent would need to know.

If a sales agent has been running thousands of prospect conversations for six months, it accumulates a body of knowledge about which objections appear most often, which leads convert from which source, how enquiries from a particular gym type tend to behave. Now introduce a retention agent into the same business. Should it start from scratch, blind to everything the sales agent has learned? We thought the answer was obviously no, and that conviction drove us toward a platform architecture rather than a collection of standalone agents deployed from different vendors with no shared memory and no shared context.

The other reason was operational. Gym operators are already running complex businesses. Asking them to manage separate applications, separate knowledge bases, separate vendor relationships, and separate points of failure for every AI capability they wanted to deploy was adding a second job, a management overhead that would quietly compound every time they added a new capability. The promise of AI-assisted operations should not come with a new category of administrative drag attached to it.

So we built Antares as a platform from the start, with a shared intelligence layer, shared toolset, one contract, one admin surface, one place to see what is happening across the entire agent estate. Five agents are live today with three more before the end of Q3. Every new activation inherits what the platform already knows.

We have never been naive about the limits of that model, though. There are operator needs too specific for a platform of our scope to address natively. A climbing gym that wants an agent fluent in route-setting data and climber grade progression. A competitive swimming club with specialist performance data it needs to make conversational and actionable. These are real needs, and the fact that we would not build them natively does not mean operators should have to manage them as separate systems, without observability, without shared context, accumulating exactly the overhead we spent two years designing out.

That gap is what the Agent2Agent protocol closes. Announced by Google in April 2025, it reached version 1.0 in April 2026 under Linux Foundation governance, backed by more than 150 organisations and in production at Google Cloud, Microsoft Azure, and AWS. We tracked it from the earliest public discussions and completed the technical work ahead of the 1.0 release.

Antares now has full A2A compliance.

For an operator, it means any A2A-compatible specialist agent, whether bought from a niche vendor, built by a developer, or assembled in a no-code tool that supports the standard, can be integrated into Antares rather than deployed alongside it. The specialist agent runs inside the platform environment on the same terms as our native agents: visible through the same operator dashboard, subject to the same governance model, with access to the same intelligence layer.

The reason this matters beyond the technical detail is what it means for improvement over time. Pulse, the intelligence layer underneath Antares, learns from everything it sees, and that learning shows up in operator results. The platform it powers delivered nearly 300 additional tours in a single month for CLUB4 Fitness. With A2A in place, self-optimisation across the platform is no longer constrained to the agents we built. It applies to the whole system, including specialist agents that came from elsewhere. One view of what every agent is doing. One surface for catching problems. One environment in which performance compounds rather than fragments.

The operator who chose Antares for a sales agent now has a platform that can run five agents natively, integrate specialist agents from any A2A-compatible vendor, and maintain a single coherent view of all of them. That is a different kind of infrastructure decision than buying a use-case tool, and it becomes more valuable with every agent added to the estate. If you are thinking about where AI sits in your operational infrastructure over the next two or three years, book a platform walkthrough and we will show you a specialist agent running inside Antares, sharing the same intelligence as the native estate.