Share with your CIO
Agentic AI deployments are quietly creating a second bill that most enterprise buyers haven’t priced in. Toby Crick at Bristows, a law firm with a direct commercial interest in advising on exactly these disputes, lays out the cost and licensing exposure agentic AI creates across traditional SaaS contracts. The core problem: a single agent API connection can do the work of a hundred named users while counting as one seat, and enterprise-wide flat-fee licenses never modeled the compute load an agent generates. Vendors are already watching, and renewal conversations are coming.
What this means for your business
The exposure here divides cleanly by contract type. Consumption-based contracts will show the pain first, often as an unexpectedly large invoice before anyone realizes an agent has been hammering an API thousands of times per hour. Per-seat contracts are more dangerous in a slower, legal sense: you may be operating outside your license terms right now without knowing it, because most “user” definitions were written for humans and haven’t been tested against autonomous software in court. If your organization has already deployed agents against production SaaS systems, the question isn’t whether this affects you; it’s whether your legal team has read the acceptable use policy recently.
The SAP v. Diageo precedent deserves more attention than it typically gets outside legal circles. In that case, Diageo expanded system access to more employees through an intermediary tool and a court found it liable to pay the higher per-user price for every single one of those employees, not just the original super-users who held licenses. The logic is directly portable to agentic deployments: if an agent acts as a multiplier for end-user access to a licensed system, a vendor’s lawyers can credibly argue that each human feeding instructions to that agent is an indirect user. The case was never appealed. It stands.
Vendors are not sitting still. The smarter SaaS players are already instrumenting their platforms to detect non-human usage patterns, and they will surface that data at renewal. The enterprises that get ahead of this aren’t the ones who slow down agent deployments; they’re the ones who open the vendor conversation before the vendor does. A CIO who walks into a renewal having already modeled agent-driven consumption, proposed a usage tier, and framed it as a partnership gets a very different negotiation than one who gets handed an audit finding. The renewal clause on your calendar for the next twelve months is the actual decision point here, not some future policy.
Concept deep-dive: Per-seat leakage
Per-seat leakage is what happens when a single software license credential, designed to meter one human user’s access, gets used by an agent that operates at machine speed and volume. The original pricing model assumed a person could only do so much in a day. An agent has no such ceiling. Think of it like buying one ticket to a stadium and then busing in a hundred people on that stub. Vendors priced their margins around human throughput; agents shatter that assumption without technically violating a user count.
Based on reporting from Who’s Using Who? Software Costs and Pricing in the age of Agentic AI, Toby Crick, originally published 2026-06-15 08:53:00.
