Blackhorn Ventures participates in Canyon Code’s $5 million pre-seed financing to help enterprises observe, optimize, and govern multi-agentic applications at scale.
Blackhorn Ventures is proud to announce our investment in Canyon Code, which is building the workflow intelligence layer for enterprise AI. The company’s $5 million pre-seed round was led by Cota Capital, with participation from Newbuild Venture Capital and Blackhorn Ventures.
Agentic AI is entering a new phase. Enterprises are moving beyond individual copilots and chat interfaces toward systems in which multiple agents collaborate, call models and tools, retain context, and act across business processes. That shift creates a new infrastructure challenge: the software stack that serves models was not designed to understand the end-to-end workflow that creates the business outcome.
Today’s infrastructure can count tokens, allocate containers, and schedule model calls. But it often cannot answer the questions enterprise operators now need to ask: Which workflow is driving cost? Where is it stalling? Which agent is blocking the rest? Should a customer-facing workflow prioritize latency, while an overnight research workflow prioritizes cost or accuracy? Canyon Code is building the layer that makes those questions measurable and actionable.
Agentic applications need workflow-level intelligence
A multi-agentic application is not a single model request. It is an evolving web of agents, model calls, tool invocations, and shared context. The path can change at runtime based on previous outputs. When infrastructure treats each call as isolated, it loses the dependencies that determine which work is urgent, what state must move with it, and where capacity is being wasted.
Canyon Code observes the dependencies between an application and the large language model calls made by individual agents. Its dependence graph tracks progress across the workflow and gives the model-serving layer additional context for scheduling, orchestration, and contextual-memory management. If one agent’s output unlocks several downstream tasks, the system can prioritize it. If a workflow can tolerate more latency in exchange for lower cost, the enterprise can govern it accordingly.
This is a significant architectural shift: optimize the outcome-producing workflow, not merely the component calls.
From $/token to $/workflow
The Canyon Code team argues that the workflow is the right unit of enterprise AI productivity. A cost-per-token metric can show aggregate model spend, but it cannot tell a business whether its contract review workflow, recruiting agent, or customer-support system is delivering value efficiently.
Measuring at the workflow level connects infrastructure behavior to a business process. It allows engineering, finance, and business owners to see the same unit: the total AI spend and performance associated with a completed workflow. That visibility is the foundation for optimization and governance.
“The durable control plane for enterprise AI will be built around workflows, not isolated model calls.”
Observe. Optimize. Govern.
Canyon Code’s product journey is organized around three capabilities that build on one another:
As agentic systems proliferate, a one-size-fits-all infrastructure policy becomes less useful. The right target depends on the workflow and the user. A customer-facing sales agent may need speed; contract review may favor accuracy; back-office summarization may optimize for cost. Canyon Code gives operators the context and controls to make those tradeoffs deliberately.
Compute efficiency is becoming a strategic advantage
At Blackhorn, we invest in digital infrastructure that expands productive capacity while using physical resources more efficiently. AI compute is rapidly becoming a critical input for enterprises and industrial systems. Meeting demand cannot rely only on building more data centers and buying more accelerators. It also requires software that converts a greater share of each GPU-hour, watt, and dollar into useful work.
Workflow-level visibility can expose idle time, scheduling bottlenecks, redundant context movement, and workloads that are optimized against the wrong objective. We believe Canyon Code can help enterprises get more out of existing infrastructure, accelerate experimentation, and lower the compute required for each completed business outcome.
This is why Canyon Code fits squarely within Blackhorn’s compute-efficiency thesis. The company is building a layer that can improve both economics and resource efficiency as agentic workloads scale.
A team built for the systems challenge
CEO and founder Ravikiran Gopalan is a three-time deep-tech founder. Before Canyon Code, he served as CTO at Aira Technologies, where he deployed multi-agentic applications into enterprises, after a decade of systems engineering at Qualcomm and Bell Labs. His experience gave him a direct view into the gap between agentic applications and the infrastructure beneath them.
Chief Scientist and co-founder Aditya Akella is an ACM Fellow and Regents Chair in Computer Sciences at the University of Texas at Austin. His research spans machine-learning systems, operating systems, and cloud networking—the disciplines required to build a new workflow-aware runtime layer.
Together, Ravi and Aditya combine enterprise deployment experience with foundational systems research. That founder-market fit, and their focus on a technically difficult but increasingly urgent problem, were central to our conviction.
Building the control plane for enterprise AI
Canyon Code will use the financing to continue developing its platform and expand its research and development team. The company has emerged from stealth at a moment when enterprises are beginning to demand not just more capable agents, but dependable, observable, and governable systems.
We are excited to partner with Ravi, Aditya, and the Canyon Code team, alongside Cota Capital and Newbuild Venture Capital, as they build the workflow intelligence layer for enterprise AI.
Learn more at canyoncode.ai.
Financing reference: DLA Piper transaction announcement.
