Vendors of professional services automation software shipped context servers in the period, letting AI assistants reach data, processes and agents inside the system, with initial support for models from three different vendors.
The standard adopted is the same open protocol that received, weeks later, the largest revision in its history, with hardened authentication. The calendar coincidence isn't accidental: corporate adoption accelerated and brought the access control problem with it.
What this kind of integration changes in practice is the assistant's reach. It stops answering from general knowledge and starts consulting the company's real data, which increases usefulness and risk surface in equal measure.
Choosing to support models from different vendors from the start follows the same direction as other products in the period. Tying a corporate system to a single model vendor became a hard decision to defend.
For anyone deploying, the question list is familiar: what data the assistant sees, which identity it uses, what gets logged, and what requires human confirmation before execution.