Safely manage your Zendesk from the AI assistant you already use, via the Deltastring MCP. Beacon configuration platform
← Back to news

Cohere VP says enterprise AI sovereignty requires control of the full agent stack

Cohere's VP of product engineering used VB Transform 2026 to articulate a strategic position that challenges the current vendor landscape: enterprises need end-to-end control over their AI agent infrastructure to achieve genuine sovereignty over their customer experience operations. The argument centres on a critical vulnerability in how most organisations currently deploy AI—by relying on third-party platforms that control the underlying models, data pipelines, and agent orchestration layers. For CX teams already embedded in Salesforce Agentforce or similar integrated platforms, this raises an uncomfortable question: how much of your agent's behaviour and decision-making actually belongs to your organisation versus the vendor, and what happens when vendor priorities diverge from your customer experience strategy?

The sovereignty framing matters because it moves beyond typical vendor differentiation into operational control. When a CX leader deploys an AI agent through a managed platform, they typically gain access to configuration and customisation at the surface level whilst the foundational model, training data handling, and core reasoning architecture remain proprietary black boxes. Cohere's position suggests this arrangement creates dependency risk—vendors can change model behaviour, adjust pricing, modify data retention policies, or sunset features without meaningful recourse. For mid-market and enterprise teams, the practical implication is stark: should you be evaluating whether your current platform gives you genuine portability, or are you locked into a vendor's roadmap regardless of your actual business needs?

The full-stack control argument also exposes a structural advantage for larger organisations with dedicated AI infrastructure teams. Smaller support operations lack the engineering capacity to manage their own model fine-tuning, data governance, and agent deployment pipelines, which means the sovereignty narrative, whilst technically sound, may simply entrench existing market power rather than democratise it. The real tension for CX professionals isn't whether sovereignty matters—it does—but whether the vendors offering it will price it in ways that make it accessible beyond enterprise accounts with six-figure AI budgets.