By Sagnik Bhattacharya, CEO, Rhapsody
LinkedIn: Sagnik Bhattacharya
LinkedIn: Rhapsody
The more healthcare expects from AI, the more it will demand from interoperability. Each new application introduces its own data requirements, integration dependencies and workflow considerations. It needs the right data, in the right form, from systems that may have been implemented years or even decades apart. It needs to fit into workflows that have evolved around the particular needs of an organization. None of that happens automatically.
In conversations with healthcare organizations, we hear versions of the same problem again and again: there is no shortage of technology they want to put to work. Finding the capacity to connect it all is another story. According to 2026 federal data, 91% of hospitals integrate third-party technology for at least one clinical purpose, but only 52% use standards-based APIs to integrate that data. Healthcare is clearly more connected than it once was. The day-to-day work behind those connections, however, remains complicated.
More connections are only part of the story
Standards such as HL7 FHIR and broader adoption of APIs have unquestionably improved interoperability. What they have not done is make every healthcare environment uniform. Anyone who has spent time around integration teams knows how quickly the exceptions accumulate. Specifications need interpretation. A field may be populated differently by one source system than another. An implementation may have requirements that aren’t obvious from the documentation. A connection that has operated reliably for years may need attention when something changes upstream.
There is also a growing expectation that the data moving through these connections should be useful for far more than the original use case. We hear this particularly from organizations modernizing long-standing integration environments. They have already done the hard work of establishing connectivity. Now they want that same data to support analytics, quality programs, care coordination, and AI.
AI amplifies both sides of the equation. It creates more demand for interoperability while raising expectations for the data interoperability delivers.
The integration workload extends well beyond code
This is where the conversation about AI and engineering can become overly narrow. Code generation gets attention because it is easy to demonstrate, but talk to people who build and maintain healthcare integrations and it becomes clear that writing code is only one piece of the job.
A considerable amount of time is spent understanding what needs to be built in the first place. Engineers search documentation, interpret specifications, work through security and provisioning requirements, investigate existing configurations, test changes and troubleshoot problems. They also carry institutional knowledge that rarely exists in a neat document: how a particular application behaves, why a customer needs data structured a certain way, or what happened the last time someone changed an interface.
That is where I see a more practical role emerging for AI. AI can help an engineer get to the right information faster, analyze an existing configuration, identify potential redundancies, suggest test data or take a first pass at repetitive development work. These may not be the AI use cases that attract the most headlines, but they address work integration teams encounter every day. And, the engineer remains central to the process.
One thing we’ve heard from technical teams using AI-assisted integration tools is that the quality of the result depends heavily on the context the engineer brings to it. AI may understand a specification. It doesn’t automatically understand the history of an environment, the intent behind a business requirement or the exception everyone on the team knows to account for. The expertise shifts toward applying that knowledge, rather than spending valuable time on mechanics that technology can help accelerate.
Creating capacity has consequences beyond integration
This is ultimately why integration capacity deserves more attention from health IT leadership. An integration backlog rarely stays contained within the integration department. It shows up as a delayed application launch, a longer onboarding process, a modernization initiative waiting for resources or an AI pilot that is ready to expand but cannot yet access everything it needs.
Adding people every time demand increases isn’t a particularly scalable answer. Neither is assuming AI will simply take over the work. A more useful approach is to look closely at where experienced teams are spending their time and separate the work that truly requires their expertise from the repetitive effort surrounding it.
AI gives healthcare organizations a new way to attack the latter. It also creates an opportunity to retain and reuse more of what integration teams already know. Healthcare organizations have accumulated years of knowledge about their systems, interfaces, data and workflows. Too often, that knowledge has to be rediscovered each time a new connection is built or a familiar problem appears in a different form.
As AI adoption accelerates, interoperability will have to keep pace. Creating more integration capacity isn’t about asking engineers to work faster. It’s about giving them more time to apply the expertise healthcare increasingly depends on.