The next frontier in AI customer experience isn’t speed or scale. It’s accountability — and Sendbird is making the case that AI can own it.
Sendbird, the AI customer experience company powering conversations for more than 300 million people every month, has launched Agent Steward, an AI agent designed to take full, end-to-end ownership of complex customer issues — and to hold that ownership across systems, channels, and time without requiring human coordination at every step.
The launch is accompanied by two platform updates — Trust OS 2.0 and Voice 2.0 — that together represent a significant expansion of Sendbird’s Delight.ai platform, pushing it from reactive AI support toward autonomous improvement and proactive customer engagement.
The Problem With Reactive AI
The case for Agent Steward begins with a structural critique of how enterprise AI currently operates. Despite rapid adoption, AI agents remain largely channel-specific and reactive — waiting for a prompt, handling the immediate query, and handing off to human teams when anything more complex arises. Mistakes are identified and corrected after the fact, if at all.
Consumer expectations reflect the gap. According to new data from Sendbird, 57 percent of consumers say the ability for AI to correct its own mistakes and reverse decisions would increase their trust. Nearly two-thirds — 59 percent — say the ability to stop or override an AI agent is very important to them. The signal is clear: what customers want from AI is not just competence. It is accountability.
Also Read: The CIO Who Says Governance Can Actually Speed Up AI
Agent Steward: One Owner, Start to Finish
Agent Steward is built specifically for long-horizon, multi-step workflows — the category of customer issues that current AI systems handle poorly. Rather than routing a problem across multiple agents and teams, Agent Steward acts as a single point of accountability from first contact through to resolution, coordinating across backend systems, chat, SMS, and email while escalating to a human only when genuine judgment is required.
The practical implications are significant. A customer who receives a damaged product from a third-party logistics partner currently faces a resolution process that can span days, involve multiple handoffs, and result in a chargeback. With Agent Steward, the issue is identified proactively, owned by a single agent throughout, and resolved within hours — with human involvement triggered only at decision points that require it.
“Most AI systems are effective at handling routine queries, but break down when it comes to complex, multi-step issues,” said John Kim, co-founder and chief executive of Sendbird. “The goal isn’t to replace human agents, but to elevate them — freeing them to focus on judgment and exceptions while AI owns coordination, follow-through, and continuous improvement.”
Trust OS 2.0: AI That Fixes Itself
The launch of Trust OS 2.0 introduces what Sendbird calls Zero-Touch Improvement — a capability that enables AI systems to identify, diagnose, and correct their own errors in real time, without requiring manual intervention from engineering or operations teams.
The contrast with traditional AI management is stark. Under conventional frameworks, identifying a model error, rewriting the relevant rules, testing the fix, and deploying it can take days or weeks. Zero-Touch Improvement compresses that cycle to real time, shifting AI governance from a supervisory burden to an autonomous function.
Also Read: Is Your CRM Making Your Customer Service Worse?
Voice 2.0: Reaching Customers Before They Ask
Voice 2.0 extends the platform’s capabilities to proactive outbound engagement, allowing companies to initiate contact with customers when a potential issue is identified, rather than after it has already created a problem. The capability supports more than 100 languages, enabling global deployment without localization overhead.
The use cases are immediate and practical. An airline can notify a traveler of a disruption before they reach the airport. A financial services company can send a payment reminder before a deadline is missed. The shift from reactive support to anticipatory engagement represents a meaningful change in how AI interacts with customers across the full relationship lifecycle.
Agent Steward, Trust OS 2.0, and Voice 2.0 are available now as part of the Delight.ai platform, following their debut at Sendbird’s annual customer experience event, Delight Spark.