Large companies now hold more AI capability than they use. Frontier models are available to every bank, insurer and energy company. For most of them, the limit is no longer the model. It is the organisation’s capacity to absorb what the model can already do.
Broad and shallow
In nearly every large company we meet, AI is already present: pilots in several functions, licences rolled out, a steering committee in place. Very little of it is in production, built into the workflows where decisions are made.
Adoption is broad. It is not yet deep.
What holds adoption back
The obstacles we see are rarely technical. Someone has to decide where AI should act first and who owns that decision. The process around it has to be redesigned, and the people who run it retrained. Governance and security rules have to be met before anything goes live, especially in regulated industries, where the status quo is the strongest competitor.
None of these is a model problem. Each is a question of decisions, processes and rules inside the company.
What we see in deployments
A European corporate and investment bank we worked with had AI pilots that performed well in testing and never reached the analysts who review corporate credit files.
The model was never the issue. No one had decided where AI should act first, and nothing had been built to run on the bank’s own servers, under its own security rules. Once those decisions were taken with the executive committee, the system went live in 14 weeks. Credit review went from 45 minutes to 12, for 320 analysts and bankers.
Why partner programmes exist
Model providers know where the constraint sits. They scale through the firms that deploy their models inside large companies. The OpenAI Partner Network is one of these programmes, and Quicksort has joined it as a registered member.
Registered is the first tier, and we treat it as a starting point. A programme gives partners access to the provider. It does not replace the experience of putting AI into production inside a regulated company.
Where the work now sits
The work that decides whether AI pays off sits between a model that performs and an organisation that uses it:
- choosing the first decision AI should change, and who owns it;
- redesigning the process and the roles around it;
- building it into the systems the company already runs, on its servers or its cloud;
- meeting its governance and security rules from day one.
This work is learned deployment by deployment. It is what forward deployment engineering exists to do.
Model providers set the pace of the models. How fast large companies absorb them depends on the people doing that work.