Rachad Alao, vice president of product engineering at Cohere, told enterprise leaders at VB Transform 2026 that companies building AI agents must maintain control over their entire technology stack to preserve data sovereignty and avoid vendor lock-in.
Speaking at the Menlo Park conference, Alao stressed that enterprises cannot simply adopt third-party AI solutions and expect to retain control of sensitive information or the flexibility to switch providers. Instead, organizations need ownership over the full agent infrastructure, from the underlying models to the deployment layer.
The argument reflects growing tension in enterprise AI adoption. Companies want to leverage powerful generative AI systems to automate complex business processes, but they hesitate to hand over proprietary data or become dependent on a single vendor's black-box infrastructure.
Cohere, a Canadian startup, positions itself as an alternative to larger incumbents by offering enterprises the ability to run AI agents on their own infrastructure while maintaining control over model updates and vendor relationships. This aligns with a broader industry shift toward "sovereign" AI solutions, particularly among regulated industries and large corporations concerned about data residency, compliance, and strategic flexibility.
Alao brings relevant expertise to this conversation. He previously managed responsible AI and trust and safety engineering at Google and Meta, giving him credibility when discussing the security and governance challenges that emerge when enterprises deploy AI agents at scale.
The sovereignty pitch resonates because enterprise AI adoption has exposed real problems. Companies using third-party agents cannot easily audit decision-making processes, cannot guarantee their data stays within specific jurisdictions, and face difficulties if they need to migrate to competing platforms.
However, the full-stack control model demands significant technical investment from enterprises. They must hire skilled teams, maintain infrastructure, and manage model updates themselves. This creates a trade-off between convenience and control that large organizations are increasingly willing to make given the stakes involved with AI systems handling critical business functions.