Introducing Human Assisted Resolution: let your AI agent ask a colleague for input

For AI agents, getting brief human input and handing over the conversation entirely should be two different things. Today, we're launching Human Assisted Resolution (HAR), an opt-in feature that allows your AI agents to do the first without the second.

A couple of years ago, an AI agent that could complete a task for your customer was extraordinary. Today, branded agents are consistently authenticating customers, pulling up accounts, working through multi-step requests, and completing them.

However, what hasn't changed is what happens when customer-facing AI needs approval or hits a roadblock. 

For many deployments, it still transfers. The customer still waits, repeats their story to someone new, and the work the AI already did gets thrown away. And every one of those conversations is one you double paid for: first to automate, then again for a person to finish. And along the way, you frustrate your customer.

So, how do you get a person involved without escalating the conversation?

Human Assisted Resolution (HAR) is an opt-in feature within OASYS that lets your AI agent ask a colleague for input in real time, while remaining in control of the conversation. How? When it can't continue without team permission, it orchestrates to a staff colleague the same way it would to another task agent: asking them a specific question, getting an answer, and then carrying the interaction through to resolution on its own. The customer remains in the same conversation the whole time, and gets the outcome they came for.

We believe this is how human-in-the-loop is meant to operate in the agentic era.

Driverless taxis use a structurally similar model in their autonomous fleets. When a vehicle encounters a situation it cannot confidently resolve, a remote specialist can provide additional context while the autonomous system remains in control. Without that capability, vehicles would have to pull over and wait for on-site assistance—an unsafe, inconvenient constraint that would make the service commercially unviable. HAR brings the same principle to the enterprise conversations customers have with your branded AI agents.

HAR is different from escalation with better context passing, which has been the category's soft solution to date. And it isn't an agent assist feature that helps a representative once they've taken the conversation over. With HAR, a human never has to take the conversation over in the first place. They contribute one piece of judgment, in seconds, and move on to the next.

The clearest way to understand HAR is to watch it in action.

HAR in the real world

A customer messages about a refund she's still waiting on. Her chair was delivered to the wrong address and never arrived. The AI agent pulls her order, confirms she's owed a $450 credit, and reaches the one step it can't take alone: issuing a credit of that size requires authorization.

It tells her it's confirming, then sends a colleague a single request with the details. The rep reads it, and approves on the spot. The AI agent is now able to quickly tell the customer her refund is approved and in route. 

She never left the chat, never repeated herself, and never talked to anyone but the assistant. The approval is captured with the conversation, so a step that needed human accountability still has it, with greater efficiency at scale than ever before.

Human assistance becomes a property of the moment

What your AI agent does should stay consistent. However, where a person needs to get involved can change constantly.

You decide when HAR fires: when approval is required, on topics your agent hasn't learned yet, on the knowledge that lives in your best team members' heads and never reaches a document, when the agent isn't confident it heard the customer correctly, on cases that need human judgment, or on a set percentage of conversations if you want eyes on AI-handled work while you build trust in it.

Human input stops being a property of the conversation and becomes a property of the individual moment, which lets you place judgment exactly where the risk is and nowhere else. That’s how we enable scale.

What HAR changes for businesses

Resolution rates now hold up through the interactions that used to break them. Complex, sensitive, nuanced, or genuinely unusual requests are exactly where automation can still stall, while a few seconds of human judgment can carry them to completion instead.

This allows your team's capacity to stretch much further. A transfer occupies one person for an entire conversation. Assisting takes seconds, which is why a single rep can support three to five conversations at once, and why an assisted interaction keeps automation’s economics.

Nobody has to learn a new tool. The interaction appears in the desktop your staff already use, whether that's our own agent console, Salesforce, ServiceNow, or your chosen CCaaS, with the full exchange in order and one specific question waiting at the end of it.

Language stops determining who can help. Translation runs both ways in real time, so an English-speaking rep can assist a Spanish-speaking customer inside the same conversation, and your standby pool gets sized by volume rather than by coverage.

And the customer stays on the channel they chose. Voice in particular stops being the place where automation gives up and a queue begins.

Resolution without compromise

As AI agents improve, it's tempting to measure them by how much they handle alone. We think the more useful measure is how much they resolve, and your escalation rate is the ceiling on that.

By separating getting help from handing off, one AI agent can ultimately do so much more for your customers. And your staff can cover more ground than before. This is what the next generation of human-in-the-loop looks like.