What actually happened
Runlayer, a two-year-old startup founded by repeat entrepreneur Andrew Berman, who previously built Nanit and Vowel, builds a gateway that sits between AI agents and enterprise systems, enforcing role-based access and logging every request an agent makes for company data. Rippling, the HR and IT platform, tested that gateway for more than a year with close engineering collaboration between both teams, but never signed as a paying customer despite the depth of the technical relationship. When a Rippling employee texted Berman that the company was building its own version, internally described as a clone of Runlayer's product, Runlayer sued for breach of the testing agreement both companies had signed.
Rippling countersued on patent grounds, a move Runlayer's side characterized publicly as pressure to force a settlement rather than a genuine infringement claim. On August 20, both companies dropped their cases after roughly three weeks of discovery, with no money changing hands and no public admission from either side about who, if anyone, was in the wrong. Rippling marked the moment by shipping its own MCP gateway product almost immediately afterward, the same category of infrastructure at the center of the dispute it had just settled with the vendor it evaluated.
Why an MCP gateway is worth fighting over
Model Context Protocol gateways are becoming the control point for agentic AI inside the enterprise, the layer that decides which internal systems an AI agent can query, what it can write back, and what gets logged for audit. As agent adoption accelerates, this layer is turning into one of the more strategically important pieces of infrastructure a platform team will buy or build in the next two years, because it is where shadow AI use gets detected and where compliance teams get their evidence trail.
That is exactly why the category is attracting both dedicated startups like Runlayer and horizontal platforms like Rippling that already sit inside enterprise IT stacks. A gateway vendor's value proposition depends on being trusted with visibility into agent behavior across many systems, which means the technical bar to build a credible competitor, once you have watched a real one operate inside your own environment for a year, is lower than most procurement teams assume.
The uncomfortable lesson for enterprise buyers
The framing that should worry a CIO here is not who was right. It is that a lengthy, hands-on technical evaluation, the kind enterprises routinely demand before trusting a startup with access to sensitive systems, created the exact conditions for the prospective customer to build a competing product. Coverage of the dispute put it directly: in an era where building new software has become close to trivial with AI assistance, a company's next competitor might be a prospect currently running its proof of concept.
That risk is not unique to Runlayer, and it is not really about bad faith on Rippling's part either. It is a structural feature of how enterprise AI infrastructure gets evaluated right now: long free trials, deep engineering access, and internal teams that are themselves fluent in AI-assisted development. The traditional multi-month technical shoot-out was designed for a world where replicating a vendor's product took a competitor a year or two, not a team with modern AI tooling and a few sprints.
What governance teams should change
Procurement and legal teams overseeing AI infrastructure evaluations should treat extended technical trials as a distinct risk category, separate from ordinary vendor due diligence. That means explicit contractual language restricting internal teams from building functionally similar systems during and immediately after a trial period, clearer carve-outs around what happens to evaluation data and design documentation once a trial ends without a purchase, and shorter default trial windows for infrastructure that touches sensitive internal systems like agent gateways rather than the open-ended arrangements common today.
It also means being honest with startups about intent earlier in the process, since ambiguity is what let this particular evaluation run for more than a year without resolution. A year-long free evaluation with no purchase decision is itself a signal worth acting on, either by converting to a paid pilot with real commercial terms and defined milestones, or by ending the relationship outright, rather than letting an open-ended trial continue indefinitely while internal engineering teams quietly absorb the vendor's architecture and design choices along the way.
Where the MCP gateway market goes from here
Rippling now competes directly in the same category as the vendor it evaluated, and other horizontal platforms with existing enterprise footholds, from identity providers to observability vendors, are watching the same opportunity. Expect more incumbents to build or acquire gateway capability rather than partner with point solutions, since the defensibility of a pure-play MCP gateway startup depends on staying ahead of platforms that already have the customer relationship and can bundle the feature for less.
For CIOs, the practical takeaway is to treat agent gateway selection as an architecture decision with a real switching cost, not a feature to bolt on later. Decide early whether this control plane belongs with a specialist vendor or with your existing platform provider, put a real timeline on the evaluation, and write the contract as if the vendor's product could become a line item in your own platform's roadmap, because in this category it increasingly can.
