The context graph is the moat — but it lives between companies
A year ago, AI agents could work autonomously for about 2 hours. Today it is 12. Capability was never the problem. Context was.
The chart everyone is sharing shows why this matters now: when agents can autonomously run 12-hour tasks, capability is no longer the excuse. The missing piece is context.
Ashu Garg’s point is that a company’s edge is not its data but the way that data gets navigated to make decisions. Agents need all of that to do real work. Agreed — but that thesis breaks down the moment a transaction crosses a company line.
Every approval, dispute, and side deal is organizational judgment that the ERP throws away. It does not recall the payment terms negotiated over years, or the exceptions granted to a specific supplier because the relationship earned it. Neither side has that context, which means no agent does either.
AP agents do not fix the handoff with the supplier’s AR agent — they just make one side of a broken exchange faster. Agent-to-agent plus a shared context graph means both agents carry the full history of the relationship: the terms, the exceptions, the patterns. Disputes that exist purely because each side is missing what the other knows simply disappear. That is the layer nobody in finance has built yet, and it is where the real moat is.