Press
The People Behind the Code
Last week we announced $5M in pre-seed funding to build the Workflow Intelligence Layer, the missing infrastructure layer that gives enterprises visibility, optimization, and governance over their multi-agentic AI applications. The round was led by Cota Capital, with Newbuild Venture Capital and Blackhorn Ventures co-leading alongside.
We started Canyon Code because we kept running into the same wall. Enterprises are deploying multi-agentic workflows at real scale, in sales, recruiting, contract review, customer support, finance. The agent-building platforms have made that easier than ever. But once those workflows are running, the infrastructure to understand them, optimize them, and govern them simply does not exist. Engineering teams are flying blind on cost. CFOs are getting bills they cannot explain. And the controls that would let an operator say “this workflow gets low latency, this one optimizes for cost.” They are nowhere.
That is the gap Canyon Code is built to close.
The Journey: Observe, Optimize, Govern
We think about enterprise AI maturity in three stages. At the first, enterprises finally get visibility: not $/token, but $/workflow. A cost spike traces to a specific workflow for the first time, and the right questions can finally be asked. At the second, the picture deepens: workflow costs hide GPU scheduling waste that token metrics will never surface. At the third, optimization gives way to governance, with per-workflow, per-persona policies that let operators allocate AI capacity as a strategic resource, not an unmanaged bill.
“Having deployed multi-agentic applications at enterprise scale, I know firsthand what the infrastructure gap looks like from the inside. Enterprises can build workflows. What they cannot do is see what those workflows are actually costing them, where the waste is hiding, or govern AI capacity as a strategic asset. That is the gap Canyon Code is built to close.”
Built on Deep Research
Canyon Code is not a thin wrapper on top of existing infrastructure. The core insight, that workflow-aware serving can eliminate the GPU scheduling waste that sits invisibly inside every AI deployment, comes from years of foundational research in ML systems. That is why I am building this alongside Aditya Akella as Chief Scientist and Co-Founder.
“AI-native enterprise workloads are revealing significant gaps in observability, resource management, and control across modern infrastructure stacks. At Canyon Code, I'm excited to guide the technical and scientific direction as we build foundational infrastructure. Our goal is to elevate agentic workflows to first-class system entities. We aim to provide the necessary visibility, scheduling, and governance mechanisms to effectively manage these workflows at an enterprise scale. This will not only enhance efficiency but also drive better decision-making and operational success.”
The team translating this research into product is led by Saurabh Agarwal as Chief Architect. Saurabh brings the systems depth to turn ML research into infrastructure that enterprises can actually run.
“Building the infrastructure for agentic workflows is one of the most compelling challenges in systems today. At Canyon Code, I am excited to be developing a solution that bridges the gap between rapid development and robust operations, empowering developers to seamlessly create agentic applications, while equipping deployment and security teams with the rigorous control they demand.”
Why Our Investors Bet on This
The round was led by Cota Capital.
“Canyon Code goes beyond simple GPU optimization to build application-aware GPU optimization, understanding how models, agents, and infrastructure interact across a multi-agentic application to optimize the whole rather than the parts. We led because the next durable layer in AI infrastructure will be built around workflow-aware execution.”
“Enterprise AI compute is heading toward a fundamentally heterogeneous landscape: multiple vendors, multiple architectures, no single dominant stack. That fragmentation creates the opening for a standardized intelligence layer that sits above the hardware and makes sense of the whole. And as AI spend scales, the FinOps teams already responsible for governing cloud costs will need exactly this kind of workflow-level visibility to do their jobs. That is what Canyon Code is building, and it is why we co-led the round.”
“Canyon Code's breakthrough platform unlocks the efficiency needed to make agentic AI commercially real: their combination of performance, observability, and developer experience is exactly what's needed to build and monetize the next generation of AI-native applications for industry at scale. We've known Ravi for more than a decade through Stanford GSB, and we're proud to support his remarkable team in its next chapter of growth.”
What Is Next
We are working with enterprise design partners today. If you are deploying multi-agentic workflows and want more control over your AI infrastructure, we want to talk.
We are also actively hiring. If you live and breathe ML systems, distributed systems, or AI infrastructure and want to work on one of the most consequential problems in enterprise software, we would love to hear from you. See our open roles →