
When a carrier emails Bay Transit Co about a load, an AI agent can reply within minutes with the details or a rate request. An agent can also pull tracking information when a customer asks where a shipment is, then pass the inquiry to someone who can decide what to do. Brandon Bay, the logistics agency's founder and CEO, estimates that bringing external agents into Front saves his team at least 11 hours a week on lookups, paperwork, and first replies.
Bay's account, shared in Front's October 6 announcement, caught my attention because it gives a fairly ordinary example of how AI fits into customer operations. Customers want answers quickly, and the people responsible for the shipment need enough information to make a call when something changes. The agent fetching the tracking update and the employee handling the exception are doing two parts of the same job.
Why agent choice creates a coordination problem
A company may want a specialist agent for logistics, another for routine service requests, and an assistant that helps account managers prepare for renewals. Every time work passes between them, somebody has to decide which records each agent can see, which actions it can take, and what the next person needs to know. With several tools in play, it's easy to end up tracking those decisions in different places.
Front's survey with Wakefield Research makes that tension visible. Among 400 U.S. support, operations, and account management leaders at B2B or mixed B2B and consumer businesses already using at least two agents, 98% reported having preferred AI agent vendors and 65% were required to choose from their approved options. Front says 44% would go outside the list if the approved options fell short. Meanwhile, 77% lacked coordination across agents, systems, and teams.
In my Tech Talks Daily interview with Front CEO Dan O'Connell, we talked about how to make that choice practical. Teams need to be able to bring an agent in for a particular job. They also need the conversation history and the approval process to carry over when that job changes hands.
What Front announced
Front's Bring Your Own Agent release lets teams build custom Autopilot agents or connect external agents built with Claude, Codex, Sierra, and other tools to the same workspace. A custom agent can be configured for a brand, team, or issue type, with plain-language Playbooks that define how the work should happen. Front says its native and external agents can act across connected systems and bring a person in when an approval or escalation is required.
Compass gives teams one place to score conversations handled by either agent type, track quality and customer satisfaction over time, and review suggested changes to guidance or processes. Front also announced scheduled Copilot work, such as preparing account summaries and meeting follow-ups, and new Account Management tools designed to flag churn risks and expansion opportunities. Those additions matter when an issue that starts with support is still relevant to the person reviewing the account next week.
Decide what the agent can do on its own
Ninety percent of the leaders surveyed allowed agents to respond to customers without human intervention in at least some situations. Among those businesses, 79% reported complaints tied to agent work at least sometimes.
Only 26% of the leaders in the survey said their business measured agent performance on a formal schedule. Asked how they would most likely find out an agent had given a customer wrong information, 35% pointed to the customer. I'd take those figures into a rollout meeting and ask who checks the agent's work each week, what would trigger an alert, and who can restrict its permissions when something goes wrong.
A single approved agent can be easier to test when the work is straightforward. Adding a specialist makes sense when there's a job it can do better, such as gathering details across several carrier systems. I'd run both against the same sample of requests and include the awkward cases where the tracking data is stale or a shipment has changed hands.
Can the new agent show what it checked? Does it know when to stop and pass the case to someone with authority to act? Can that person pick up the conversation without asking the customer to explain it again? Those are the tests I'd use before giving an agent wider access.
Useful Links
Missed connections: AI agents met the customer before they met the business
The Orchestration Gap: Why half-wired AI agents leave customers doing quality assurance
Full Podcast Interview With Front CEO Dan O'Connell
