Forrester’s government clients struggle with integration. It’s difficult, and it’s even more so for the public sector, where underfunded budgets and federated fiefdoms with limited shared services are the norm. The pressure is on as digital systems demand greater connectivity, especially as customers (citizens and beyond) expect a connected government experience.

Add in major cuts in resources, and it seems even more like an impossible challenge to conquer. Or is it? As governments tighten budgets and prioritize their limited time and resources, they face an important question: Is this a problem that no longer requires spending cycles to solve? With the onset of AI, and agentic AI in particular, our government clients are asking if integration is a challenging, expensive, and painful problem they will no longer have to solve in two years’ time.

Will agentic AI bridge the gap between siloed systems? Will it finally deliver the seamless integration government agencies have been waiting for, thus making interoperability a problem of the past?

No, AI Makes Your Integration Problems Worse

Sorry, it’s not good news.

AI is not magic fairy dust. How is AI supposed to access your data and digital operations without integration? The fact is AI amplifies the urgency to do integration well. Agencies that have cut corners on integration will find executing on AI will be more difficult than those that built a robust integration foundation.

For AI to be more than a chatbot, it needs events and APIs:

  • Events notify AI agents when a decision is needed and keep them up-to-date on the latest state of your systems in real time.
  • AI agents evaluate what just happened and the current state across systems — and make a decision.
  • APIs are invoked by the agent to retrieve additional data when needed to aid making a decision and to actualize that decision.

Agentic AI Will Fail With Traditional Integration Architecture

The principles of good integration are not new, but AI makes them more necessary than ever. To enable AI action, you need:

  • Definitions. APIs and events must be documented with a schema definition and plain-English semantic meaning of what they do. Otherwise, AI won’t have a clue.
  • Business architecture. APIs and events must reflect your agency’s capabilities as defined by capability mapping. They should not reflect technical applications as happens in point-to-point integration commonly found in the public sector. Otherwise they will be too low-level for AI to figure out how to use them to accomplish a goal.
  • High-quality API design. APIs with schemas and error messages that confuse humans will confuse AI agents. But unlike humans, AI agents can’t spend a few hours experimenting with an API and chatting with its owner sitting in the next cubicle aisle. AI agents don’t have memory to recall the weird quirks of an API that human developers discover and remember during the SDLC.
  • Tight security. Whatever security holes you have in your APIs will be found much faster by an AI agent than by a human.

It’s Time To Stop Neglecting Integration

The facts listed above are nothing new. They were best practices before AI. They are even more critical with AI. Now is the time to pay down the tech debt of point-to-point integration. Yes, integration is difficult, but if the public sector can figure out how to send a man to the moon, it can solve its integration problems with the right will and incentives.