What happens when the AI agent a business wants to use was not built by the customer service platform where its employees already work?
In this episode of Tech Talks Daily, I speak with Dan O'Connell, CEO of Front, about Bring Your Own Agent and the growing demand for companies to combine native agents, third-party systems, and agents they build themselves. The conversation moves beyond the question of which model is best and focuses on the operating choices that determine whether those agents can work safely and usefully around customers.
Dan describes Front as a customer operations platform that brings support and customer success teams together around customer problems. Front has its own native agents. Autopilot can complete tasks from end to end, while Copilot assists employees with escalations, handoffs, and daily service work. Yet Dan does not expect every customer to use only the agents supplied by one platform.
Some businesses are creating agents for a particular workflow. Others want greater control over cost, speed, or model complexity. A straightforward task may suit a faster and less expensive model, while a sensitive or difficult request may demand a different system. Dan believes companies will want the freedom to connect those choices with the platform where their employees and customer conversations already live.
Front says an early announcement of this approach attracted over 1,000 signups. That figure is company-reported, but it points to a practical issue many technology leaders are now facing. AI purchasing is no longer one decision made for the whole company. Different teams may select different models and agents, while employees still need a consistent place to coordinate the work.
Choice alone does not produce a coherent customer experience. An agent needs the history of the customer relationship, the current conversation, the actions already taken, and a clear understanding of what should happen next. It also needs to recognize when a request falls outside its skills and should be transferred to another agent or a person.
That handoff is where many customer experiences fail. People become frustrated when they have to explain the same issue again after moving between channels, departments, or systems. Adding agents can increase that risk if context does not travel with the work. Dan argues that the platform must preserve the information surrounding each transfer so the next participant can continue rather than restart the conversation.
The right boundary between people and agents depends on the business and the experience it wants to provide. Dan contrasts repeatable e-commerce requests, such as order tracking, refunds, or returns, with customer conversations involving health products. An agent may handle many routine retail questions from the first contact. A health-related interaction may need empathy, judgment, or a person involved earlier, even when automation could technically complete the task.
This makes automation a design choice rather than a contest to remove the largest possible amount of human work. A company must decide whether an agent should be the first contact, whether an employee should approve an action, and where people should manage escalations or coordinate several agents. The answer may vary by customer, channel, task, or level of risk.
Dan describes autonomy as a dial. A team can begin with an agent making recommendations while a person approves the final action. It can then increase autonomy when testing and operating evidence support that decision. This approach gives cautious organizations a way to introduce agents without handing over every task at once.
Observability and governance are part of that process. Managers need audit logs showing what an agent did and why it reached a recommendation. They also need testing that shows how the system responds to representative scenarios. If something goes wrong, Dan is clear that people still own the outcome. Using AI does not remove responsibility from the company deploying it.
Measurement should focus on the customer result. Resolution and deflection rates can look positive even when an interaction creates frustration or extra work. Dan recommends examining whether the customer's problem was solved and whether the experience was satisfactory, regardless of whether a person or an agent performed the work. Front uses inferred customer satisfaction across both human and AI interactions as one way to compare those outcomes.
The interview also addresses the tension between flexibility and control. Connecting several agents, models, and vendors can create another fragmented technology estate. Dan's answer is not unlimited autonomy. It is shared context, defined permissions, human approval where needed, audit trails, testing, and the ability to change the level of automation.
Looking toward 2027, Dan expects companies to select fewer core platforms while allowing several agents to operate through them. He also expects boundaries between support, customer success, and account management to continue becoming less distinct. For customers, those internal structures matter less than receiving a consistent answer from a company that remembers the relationship.
The conversation leaves leaders with a practical question. If a business gives teams the freedom to choose the agents that fit their work, can it also provide the common context and accountability required to protect the customer experience?
Would your organization benefit from bringing its own agents into a shared customer operations platform, or would the added choice create another coordination problem? Listen to the episode and share your thoughts.