What a global AI alliance tells a university office about AI
PwC and Cohere just split enterprise AI into two jobs: the platform, and everything it takes to make the platform useful. Smaller organizations need the second job too.
On October 5, 2026, PwC and Cohere announced a global alliance, starting in Canada. The split of responsibilities is the interesting part. Cohere supplies the AI platform and models. PwC does the rest:
- identifying the uses worth pursuing;
- assessing risk and regulatory requirements;
- redesigning workflows;
- setting governance and controls;
- connecting AI to enterprise data and systems;
- supporting implementation and measuring results.
That split names something most organizations learn the hard way. Buying AI and benefiting from AI are two different jobs. The announcement is candid about it: many organizations have tried AI but have trouble moving from isolated pilots to everyday use with measurable returns.
The second job doesn't shrink with the organization
A Financial Aid office of twelve people has the same problems as a bank, just smaller:
- Which tasks are worth automating, and which only look like it?
- What student or financial data may the tool touch under FERPA, GLBA and campus policy?
- Who reviews the output, and how is that review recorded?
- How will anyone know, three months later, whether it helped?
A large firm answers those questions for a global enterprise. A university office or a twenty-person law firm usually answers them alone, or not at all. That is why so many AI licenses go mostly unused.
Vendor-neutral by design
PwC's alliance is with one platform. Most organizations we talk to already have more than one: a Microsoft agreement, a Google agreement, ChatGPT accounts and a few people experimenting with Claude or Grok. They don't need another platform. They need someone to make the ones they have work, safely, inside real workflows.
That is The Koru Project's job:
- find the right starting point;
- implement on the platform you choose;
- train your team on their own work and keep it current.
The same lifecycle, at the scale of one office and one workflow.
Where to start
Pick one process your people should not have to fight. Map it honestly, including the exceptions and the workarounds. Then decide what AI should do, what it should only assist with, and where a person makes the call. If that sounds like the work you need, start the conversation.