This summer Consumer Reports (CR) started prototyping against a set of defined use cases for consumer-authorized AI agents. The use cases spell out a complete customer journey from research through purchase, support interactions, and, ultimately, return. We kicked off development on the “submitting feedback to a company” use case due to its relative simplicity and unauthenticated nature. We partnered with Genesys, a global cloud leader in experience orchestration, to deliver the prototype.
The prototype in action
Imagine a consumer decides they want to submit feedback to a company proactively; their case doesn’t require remediation, they just want the company to know more about their experience with the product. It would be very rare for a customer to initiate communication like this with companies today, considering the initiative required to identify the right support channel and make the feedback known. However with the help of an AI agent, we may feel more willing to tell companies what’s on our minds when it comes to the products we purchase.
For example, a consumer might want to convey to the company that they didn’t like the packaging their product arrived in. In this case, perhaps the consumer would say to their agent, “Tell AcmeShop their packaging was hard to recycle.” From that single instruction, the agent might perform the following tasks:
- Translating. Converting free-form consumer intent into the structured fields a support system expects.
- Clarifying. Prompting the consumer to fill in missing details when necessary.
- Delivering. Relaying the complete payload into the company’s system so it can be logged, routed, and actioned.
Success means that the consumer’s feedback arrives in a format the company can ingest, without the consumer needing to navigate confusing forms or call trees. The agent serves as a liaison, bridging the consumer’s free‑form language with the company’s structured requirements.
We built this prototype with Genesys, whose system processes requests asynchronously. Support systems rarely operate in real time: tickets are routed, checks are run, and sometimes queues build up. Our prototype reflects that reality; agents must gracefully acknowledge, wait, and resume when responses come back. Just as human callers need to listen to hold music to pass time, agents experience latency in their interactions with both the consumer and the support platform.
Our “submit feedback” prototype is currently unauthenticated, meaning feedback is not linked to the consumer’s account or history. In our next iteration, we’ll enable identity so that feedback can be tied to the customer’s profile, and interactions can become more personalized and better documented over time.
While identity is one next step, discovery is another open challenge. How will consumer agents reliably find the right endpoint inside a company’s systems? In this prototype, we used tool calling to allow our agent to connect directly with specific Genesys APIs. In the future, mechanisms like MCP servers or standardized registries may be essential to scale these interactions across many companies.
Why it matters
Customer support can be frustrating with routing issues, long hold times, and repetitive requests. Sitting on hold with support is when I feel most hopeful about a future where AI agents handle mundane tasks on our behalf. In a world where consumers could authorize AI agents to act in our stead, we could leave it to our agent to enjoy the hold music while we spend time elsewhere.
Having an agent do the work for me would of course make my life as a consumer more convenient. However, it could also empower consumers broadly, providing a means to greater representation within the systems where decisions are made.
Every year, millions of consumer complaints vanish into unstructured, hard-to-route channels. Agents that can translate consumer voice into structured, actionable data could reduce frustration – and in doing so, make it more likely that consumers share their experiences with companies. Having an agent liaise on a consumer’s behalf may help rebalance power, ensuring consumers are better represented and heard in the systems companies use to make decisions.
We see this “submit feedback” prototype as a single node in what will someday be a much broader network of agent‑to‑agent coordination. Many believe the likely end state will be “the agentic mesh,” a world where consumer agents, company agents, and third‑party services all interact in dynamic webs. We look forward to layering agent-to-agent communication features into future prototypes so we can more deeply understand the mechanics and implications of these multi-agent futures.
One inevitable challenge will be designing for loyalty and trust when agents are embedded in dense networks with nodes representing different and sometimes conflicting interests. We’ll be tackling questions like this in our Loyal Agents research initiative with Stanford.
If you’re contemplating the future of customer support, experimenting with the agentic mesh, or exploring what it means for agents to be loyal to consumers, we’d love to hear from you. Reach out to the Consumer Reports × Stanford team via innovationlab@cr.consumer.org.
Many thanks to Genesys for collaborating with us on this prototype, and to Andrew Bunyea and Kennion Gubler for their design and development contributions.
