The agent format wars have begun

Welcome back to Building AI Agents, your biweekly guide to everything new in the AI agent field!

Also reality: an “AI agent” that is just an LLM with no tools on a loop.

In today’s issue…

• 📣 SPECIAL ANNOUNCEMENT: automate your business with Orynt AI

• Google battles for control of agent integration

• OpenAI’s upcoming software engineering agent

• The state of AI in 2025 in 10 charts

• The first agent to publish a peer-reviewed paper

…and more

📣 SPECIAL ANNOUNCEMENT

Today, we’re excited to announce the launch of our agency partner, Orynt AI!

Orynt's team of software engineers and consultants works with you to identify the areas where automation can reduce costs, save time, and improve consistency in your company’s operations with AI agents.

If you're a business owner or executive ready to unlock serious growth, cut overhead, and operate smarter, schedule your free 30-minute strategy call with Orynt today!

🔍 SPOTLIGHT

Ever heard of Betamax? No? Well, that’s what happens when someone loses a format war.

Throughout the history of digital technology, a critical enabler has been the establishment of agreed-upon standards for the transfer of information between components by different vendors. Many of these, such as hypertext transfer protocol (HTTP) and universal serial bus (USB), have become ubiquitous enough to be household names—as I covered last month in my piece on the rise of Anthropic’s Model Context Protocol (MCP). But often, several different groups or companies propose different standards, and no one can agree which to use. The result? A format war.

In an typical case, two companies, each with a distinct standard, compete to have theirs established as the dominant format for a certain type of media, whether video discs, document files, device connections, or countless others. Perhaps the most famous of these conflicts was the battle between the VHS and Betamax videotape formats in the 1970s and 1980s, which was eventually won by VHS—making “Betamax” a cultural synonym for “loser format”.

When Anthropic launched the MCP for connecting AI agents to tools and data sources in November of last year, it seemed like the potential first salvo in a new AI agent format war. But with the rapid uptake of MCP by agent builders in February and March, the war became a blitzkrieg, with the other two major LLM providers, OpenAI and Google, each quickly conceding the field to Anthropic and announcing that they would adopt MCP.

Agent-tool communication, however, is only one potential front in the conflict. Of equal if not greater importance is agent-agent communication—the exchange of information between agents built using different frameworks. Last Wednesday, Google stormed into this territory with the launch of its Agent2Agent (A2A) protocol. Prominently featured in the announcement was a massive coalition of allies: major tech companies such as PayPal, Salesforce, and SAP, and nearly all of the world’s largest consulting and accounting firms. The clear unspoken intent was shock and awe. Google wants the agent-agent communication crown, and won’t let anyone stop it.

Here, however, they could meet more resistance than Anthropic did. A separate alliance with agent heavy-hitters LangChain, LlamaIndex, and others, has put forward AGNTCY, an open-source rival—though LangChain’s allegiance could be complex, as they were also named on the Google announcement, and earlier proposed their own Agent Protocol for agent-agent interoperability. IBM, though not the giant it once was, has its own Agent Communications Protocol (ACP). Unlike agent-tool communication, the agent-agent field won’t be claimed without a fight.

The stakes are potentially high: the ability to exert profound influence on the future direction of the AI agent field by controlling the ways agents from hundreds of different providers communicate with each other. But ego, too, is on the line. No company wants to be remembered as having the next Betamax.