Foundation Capital identifies AI's trillion-dollar opportunity in context graphs, arguing that enterprise value is shifting from traditional systems of record to systems of agents. These context graphs are described as living records of decision traces, stitching together information across entities and time to make precedent searchable. However, Greg Ceccarelli points out a critical oversight: Foundation Capital focuses on capturing decisions at execution time, missing the crucial first mile where decisions originate in conversations and collaborative discussions.

The problem lies in the loss of context as decisions move from conversations to systems of record. Decisions made in customer interviews, engineering debates, and management calls often lack documentation, leading to a game of telephone where the original intent is lost. This context loss is further exacerbated in data pipelines, where the reasoning behind data transformations and decisions is rarely captured, causing significant challenges in data observability and quality.

SpecStory's Intent product addresses this by recording every exchange between developers and coding agents, capturing the decision-making process before it turns into code. Their approach includes three layers: Capture, Arena, and Repo, which automatically record agent prompts, extract real-time decisions, and version decisions alongside source code. This method aims to bridge the gap between conversations and code, preserving the intent behind decisions.

The implications for AI agents are profound. Without captures of upstream intent, agents struggle with edge cases and lack the contextual reasoning that humans use. The next generation of platforms will need to capture intent upstream, keeping it close to where it matters and making it available for both humans and agents to understand the 'why' behind decisions.