Will Washburn and Khaliq Gant
Will Washburn and Khaliq Gant Agent Relay

AI has spent the past few years making individual agents more capable. The next challenge is making those agents work well together.

Engineering teams are already running multiple agents across the same projects, with each system handling a different part of the workload. Speed has improved, yet a new problem has emerged alongside it. Agents can move quickly enough to duplicate work before teammates even know what is happening.

Agent Relay, co-founded by Khaliq Gant and Will Washburn, is building infrastructure around that problem. The company is developing a communication layer designed to let multiple AI agents coordinate with each other across workflows while keeping human teammates involved. The company is currently participating in Y Combinator's 2026 Fall batch as it develops its technology for teams adopting multi-agent systems.

Gant encountered the problem firsthand while working with multiple agents as an engineer. He found himself moving between separate conversations and trying to track what each system was doing. The same issue surfaced in conversations with engineering teams, where agents were increasingly operating in isolated environments.

"People had a single agent in their siloed experience, where engineers were working with a terminal on their computer and moving really fast with that one agent," Gant explains. "But then they didn't know what their other teammates were doing with their other agents. The problem we're trying to solve is that communication and coordination problem."

According to Gant, Agent Relay's design partners are already using the system to coordinate development work. Some teams had been chatting with seven or eight agents simultaneously, creating a fragmented workflow that required people to manually track tasks and progress. Agent Relay allows a user to communicate with one agent that can manage a wider team of agents, giving the human a single point from which to understand the work underway.

The need becomes particularly apparent when agents begin operating at a pace that makes duplication expensive. Gant recalls speaking with teams that had built the same feature twice because their agents lacked a way to communicate across teammates. Agent Relay is designed to reduce those wasted cycles by allowing agents to see what other agents are doing and coordinate around shared work.

One design partner described the tool as "indispensable," according to Gant. The value lies in giving teams visibility into work that previously remained isolated on individual terminals.

Trust remains a significant consideration as agents gain greater autonomy. He explains that recent incidents involving agent access and sandbox environments have made companies more cautious about giving AI systems broad permissions. Gant argues that autonomy needs to develop alongside appropriate restrictions.

"It's like an intern," he says. "You give an intern a task to get a cup of coffee, and then next thing you know, they're getting the entire breakfast. You have to start small and grow those responsibilities, giving the intern or the agent a path to success by taking on increasingly larger tasks as time goes along."

Security becomes especially important as agents begin communicating across organizational boundaries. During a cross-company experiment, Agent Relay worked on connecting its communication infrastructure with another company's authorization layer. The exercise, he notes, highlighted the practical complexity of getting separate systems to communicate while preserving control over access.

Gant remarks, "You give the agent the least amount of access possible so that it literally can't give away some of the data because it doesn't have access to it."

Agent Relay's larger ambition reaches beyond today's engineering workflows. Gant expects a future in which agents could outnumber humans across the internet, creating a need for infrastructure capable of supporting machine-to-machine communication at scale.

"Just because you have humans in a room doesn't mean they know how to work together as a team," he says. "You have to actually teach them how to talk and work together as a team." That idea is closely tied to Gant's background in basketball. He played the sport from childhood and later experienced a serious spinal injury while playing in 2006. Doctors told him he would not walk again, but he recovered, with his teammates supporting him throughout the process. The experience left him with a lasting understanding of how much coordinated effort can accomplish.

He now applies that same principle to AI systems. A single agent can handle substantial work, yet specialized agents can divide responsibilities and coordinate toward a larger objective. The challenge is creating the communication infrastructure that allows those individual capabilities to function as a team.

"The companies that we are speaking to are seeing this problem when they are deploying 40-50 agents across their organization. We feel that the current solutions that exist in the market today don't adequately solve this coordination problem and that's the hole we're trying to fill."

The need is personal for Gant and his co-founder, Will Washburn, too. Gant is normally based in Oslo, Norway, while Washburn is in Boston, leaving a six-hour time difference between them. Their own agents have helped coordinate work across those gaps, giving them a practical test case for the system they are building.

Agent Relay is betting that the next stage of AI will depend on this kind of coordination. As companies deploy more agents, the question will increasingly shift from what a single agent can accomplish to how effectively a group of agents can work together.