Members of the AI Agent Congress gather in San Francisco, CA on March 31, 2025.
This week in San Francisco, Consumer Reports (CR) joined the second AI Agent Congress hosted by Jeremiah Owyang and Blitzscaling Ventures. The event brought together nearly 40 AI agent founders, working across consumer and enterprise.
CR attended the first AI Agent Conference in November. This time, we were invited to open the session with a lightning talk on agent trust. We explored the difference between bots and agents (agents are trusted), the iron triangle of agency (between principal, agent, and counterparty), and how CR is working to shape standards for agents that truly serve consumers.
Three discussion sessions followed our kickoff – see below for our takeaways from each.
Discussion 1: Agent Loyalty
Who are AI agents loyal to today? Founders shared their perspectives, with answers ranging from developers, to the large language models they call, to “whoever holds the purse strings.” People hoped that agents would one day be loyal to their users, which is certainly CR’s perspective as well!
One speaker posited that we’re likely to see a new kind of insurance market emerge for AI agents. Just as PayPal and credit cards offered guarantees and recourse, agents too may need assurance frameworks to earn trust at scale. The group raised the idea of referee or judge agents mediating interactions between parties, something CR has also written about.
Next the conversation shifted to trust-building strategies: agents might win loyalty by learning quickly—from users, and from their own mistakes—and by resisting the urge to overpromise. One recurring theme was that agent trustworthiness doesn’t just stem solely from technical performance: it’s also deeply tied to clear expectations, meaningful feedback loops, and systems of recourse.
Discussion 2: What’s Working?
What’s working today? Attendees emphasized that simplicity beats cleverness, and code written with clear purpose tends to outperform “vibe coding.” Summarization, data analysis, note-taking, and operating legacy SaaS tools were all cited as valuable use cases for LLMs and for agents in the market today.
Delegates agreed that conversational interfaces are helping clarify user intent, and there’s growing momentum in voice interfaces and voice agents. The group also agreed that consulting and advisory services in AI are booming, perhaps even outpacing adoption of agentic AI by enterprises. Finally, one attendee pointed out that the U.S. seems more bullish on agents than many emerging markets, perhaps because many countries are seeking to preserve or grow employment – and can do so more affordably than we can in the States.
Discussion 3: Multi-Agent Systems & Interoperability
In the final session, we discussed the protocols and infrastructure required to make multi-agent systems work. Delegates acknowledged that Anthropic’s Model Context Protocol (MCP) is gaining traction: it’s open source, seeing rapid adoption, and more flexible than alternatives like OpenAI’s AgentSDK or CrewAI.
The hurdles standing in the way of a multi-agent world are numerous: agent reliability, discoverability, trust, communication standards, and infrastructure were all named. Perhaps for this reason we still see more single-agent workflows today. However, the dream of agents that specialize, coordinate, and collaborate is alive.
In a related breakout session, a smaller group discussed computer use patterns. We agreed these were trust-building patterns, as they let users see what the agent is doing step by step – even in cases where navigating a GUI may not be the most efficient way to complete a task. Many believed that computer use was a transitional design (or “a patch”), but that it could be useful in enterprise settings like QA or software testing where there’s value in seeing the visuals. There was also cautious excitement about next-gen models that shift away from pure token prediction toward more reasoning-oriented architectures.
As with the first Congress, this event left us energized and full of questions. If you’re thinking about how AI agents can better serve people—and want to help define that future—we’d love to hear from you. Reach out to us anytime at innovationlab@cr.consumer.org.
