Freedom of choice: models, development direction and ways of working
The fourth and final episode of “Four questions before choosing an AI environment” starts with a question nobody can answer: which model will be best in two years? Since nobody knows, the safest decision is not about choosing the right model — it is about choosing an environment where changing models is no big event.
A decision about an AI environment has to be made today, when nobody — no supplier and no analyst — knows which model will be best in two years. In the final episode, we show how to make a decision you will not have to pay for twice.
What we discuss in this episode
Local and interchangeable models as an architectural assumption — without a catalogue of approved suppliers, somebody else’s version retirement date or a forced rebuild. Choose the model for the task: document classification does not need the most expensive AI model, while contract analysis may need it — lower costs wherever a smaller model is sufficient. We demonstrate this in PROXY:AI: matching models to tasks, configuration changes that go through tests and criteria before a decision, routing groups that preserve the process while allowing the model to change, and a national model for Polish and a local model in the organisation’s infrastructure — connected on the same terms.
Continuity independent of the supplier — software source code and training are handed over. Not so that you have to take over development, but so that you can if the need arises. This mechanism works best when it never has to be used.
Who decides the development direction — in global products, the roadmap is set centrally and the influence of a single institution is close to zero. SAIE develops on a quarterly cycle; priorities come from customer discussions, deployments and research, and the development roadmap is public. In regulated sectors, unusual requirements are the norm — this is the difference between a solution that fits and one that requires you to adapt your own processes.
Ways of working as an organisational asset — the cognitive digital twin in TWIN:DESK learns how a particular user works and lets that knowledge be shared as a pattern for the team. Documents remain when someone leaves. Their way of working leaves with them — and no document search engine solves that problem.
Five questions about independence — and 24 questions to close the series
The episode closes with five questions about independence — for every supplier, including us: from the cost of changing models and the right to use local and national models, to what remains on your side when the relationship ends.
Together with the questions from the previous episodes, they form a complete checklist: 24 questions for an AI environment supplier — 6 about cost, 7 about data and code, 6 about compliance and 5 about independence. You can download it and work through it with any supplier. Including us.
The best architecture keeps you in control: you can change the model, supplier and development direction without rebuilding everything from scratch. We design SAIE around this assumption.
Four questions before choosing an AI environment
- How much does it really cost?
- Where are the organisation’s data, code and competitive advantage?
- Do you receive compliance as a ready recipe, or must you build it as a separate project?
- Will you still be free to change the model, supplier or way of working in two years?
VIDEO · EPISODE 4
Freedom of choice: models, development direction and ways of working
Checklist: 24 questions for an AI environment supplier
Episode 1: How much does an AI environment really cost?
Episode 2: Where are the data, the code and your competitive advantage?
Episode 3: Compliance as a ready recipe, rather than a project to build