When implementing new software systems, most companies have not had the luxury of investing in solutions that fit the business 100%. Usually, it’s a 70% fit, with customisation layered on top. AI tools and integration platforms have changed this story. Companies can now build their own applications around their own rules and logic and place them comfortably on top of the systems they already use.
For JJ Milner, MD of Global Micro, this works in two ways. “You have a team that’s now equipped to build integrations at tremendous scale and speeds that were never possible before,” he says. “On the other hand, you can develop integrations and use cases that weren’t possible before AI.”
That said, AI can fail. Prototypes often get to 70% rapidly, with the remaining 30% evolving over the next two to three weeks, but then they become brittle and fractured when they need to change. “There is a spaghetti underneath the application that you didn’t build,” says Milner. “This is tolerable for a form or for a minor integration, but if you’re using AI to handle critical systems and data like invoicing, stock, patient records or credit checks, 99.9% isn’t good enough, and brittle systems are not acceptable. You need AI tooling and integrations that have 100% fidelity.”
South African companies are solving this discrepancy between the speed of AI and the reliability of AI by pointing AI tools at deterministic platforms and systems of record that are 100% reliable, then building human experiences around them. “If a client questions a credit card transaction, for example, you can’t go and edit the AI’s code; you need a system that has logs and records,” says Milner.
Governance sits at the heart of this new wave of AI integration and intelligence. It is a layer that adds critical oversight and insight to AI’s potential and, says Milner, it is essential to treating AI agents as identities with scoped permissions, just like employees. Sometimes the agent needs more access than the user, such as checking timesheet hours without exposing the underlying records. Sometimes it needs less. “If I am working on planning an event, do I want to give my AI my credit card or do I just want to make the booking myself,” says Milner. “This type of thinking is what should define how you run your agents, the permissions they have, the data they process and the systems they can access.”
Milner leaves companies with three priorities. “The first is to be systematic and to make it safe to fail,” he says. “Then analyse and document the business so effort goes where it counts, and finally, just because you can build something, it doesn’t mean you should, so prioritise governance and relevance. That way, you can build into that 30% in ways that differentiate your business and your potential.”
Share
