Agentic artificial intelligence startup Canyon Code stepped out of the shadows today after closing on a $5 million pre-seed funding round.

Its mission is to give enterprises the granular controls required to successfully manage and optimize a new generation of “multi-agent” intelligent applications.

Cota Capital led the round, with participation from Newbuild Ventures and Blackhorn Ventures. The funds will help to accelerate the startup’s research and build out its new “workflow intelligence layer,” which is a specialized software stack that helps to govern how AI agents interact with each other and the models that power them.

AI agents are quite a bit different from the generative AI chatbots most people are familiar with today. Unlike chatbots, which can only respond to questions and generate images and text, AI agents are designed to work autonomously on behalf of people, freeing them up to focus on more creative work. These agents can reason about problems, use third-party tools and collaborate with other agents to get work done. But as enterprises look to move beyond experimental pilots and deploy more complex multi-agent applications at large scale, they’re experiencing some major problems.

One of the most significant challenges they face is that the existing infrastructure their agents are deployed on was primarily built for the model serving layer, rather than the actual agentic workflow. That means they lack visibility into what their agents are doing, the actions they take, the decisions they come to and so on.

This lack of clarity is the main reason behind the spiraling costs of agentic AI and inconsistent performance of agentic workloads. According to Canyon Code co-founder and Chief Executive Ravikiran Gopalan, what enterprises need is a special workflow intelligence layer that can help them to see and manage the collective behavior of their agentic applications.

That’s what Canyon Code provides. It has built an enterprise-grade orchestration layer that sits above the model serving layer and gives companies insights into what their agents are actually doing. The critical component of this layer is Canyon Code’s dependence graph, which helps users to monitor the dependencies and interactions between different AI agents.

This graph enables companies to track the real-time progress of each AI agent and provides much-needed context that’s fed back into the model serving layer. Using these insights, teams can orchestrate and schedule their model calls with greater efficiency.

For example, if one agent’s output is a prerequisite for three other agents, it will prioritize that agent’s call to reduce system latency. The agent will do its job first, so the others have the context they need to fulfill their own tasks. In addition, Canyon Code also manages the contextual memory to make sure that every agent has access to the information it needs, when it needs it, without bloating prompts and increasing the costs.

With this level of optimization, enterprises can start setting more specific policies, Gopalan said. For instance, a company could set things up so customer-facing support agents prioritize lower latency, while back-office data analysis agents would instead put a priority on the accuracy of its outputs and cost-efficiency. This can be done even if the two agents are running on the same underlying model.

Gopalan is a three-time founder who has already successfully scaled up one agentic AI startup, while his co-founder Aditya Akella is a respected researcher that has previously authored papers on topics machine learning and operating systems.

“Enterprises are crossing the dependability thresholds with agentic systems and are beginning to deploy more and more multi-agentic apps at scale,” Gopalan said. “However, they don’t have an easy way to set policies of behavior for these apps on a per-app and per-persona basis. Canyon Code’s technology will allow them to do exactly that.”

Image: Canyon Code