All insights

Perspective10 November 20262 min read

We don't build models. We build with them.

Foundation models are becoming the most powerful general-purpose infrastructure ever built. The winning strategy for almost every organisation is to use them brilliantly, not to compete with them.

People sometimes ask whether Sukarma trains its own AI models. The answer is no, and it is one of the most important decisions we have made.

The economics are clear

The leading model developers (OpenAI, Anthropic, Google, Meta, Mistral and others) invest extraordinary sums in compute, data and research. Each new generation arrives months, not years, after the last, and each one makes the previous generation cheaper. Competing with that is not a sensible use of a public body's budget, and it is not a sensible use of ours.

What is scarce is the ability to turn those models into working, accountable services: understanding a policy area, designing for real users, integrating with twenty-year-old systems, proving accuracy, handling personal data lawfully and keeping the whole thing running at a predictable cost.

That is the work we do.

Model-agnostic is a procurement principle

The Technology Code of Practice asks government to avoid lock-in and use open standards. The same principle applies to AI. If your service is hard-wired to a single model provider, you inherit that provider's pricing, availability, data terms and roadmap.

Every system we build has a thin, well-defined boundary between the application and the model. In practice that means:

  • Provider abstraction. Prompts, tools and evaluation suites are portable across models.
  • Continuous evaluation. Each candidate model is scored on your tasks, so switching is a measured decision rather than a leap of faith.
  • Open interfaces. We use open protocols such as the Model Context Protocol (MCP) to connect models to your systems, rather than proprietary plug-ins.
  • Cost routing. Simple tasks go to small, cheap models; hard reasoning goes to frontier models. As prices fall, your bills fall too.

What this means for clients

You get the best model for each job today, and the freedom to adopt better ones tomorrow. You own your prompts, evaluation data and integration code. And you never have to bet a public service on the fortunes of one AI company.

We think that is what "using AI responsibly" looks like in practice: not just safe outputs, but sound architecture and sound commercial decisions.

Keep reading

Related insights