Enterprise AI Pilots Ship When CIOs Manage Agents Like Digital Interns

The math behind autism care leaves no room for AI pilots that stall. The CDC now estimates 1 in 31 children are diagnosed with autism; a caseload that demands 300,000 to 400,000 clinicians. The US has roughly 70,000 to 80,000; an already limited workforce faces the same burnout pressures affecting healthcare providers nationwide. The result is a widening access challenge, with families often waiting months for therapy that works best when it starts early. A proof of concept that never survives contact with production is a luxury clinicians can't afford to fund.

Pawanjit Singh is the Chief Information Officer of Centria Healthcare, one of the nation’s largest providers of applied behavior analysis (ABA) therapy for children with autism and an industry leader in clinical quality and outcomes. He brings more than two decades of technology leadership across retail and healthcare, with senior roles at Family Dollar, Premier Inc., and Sevita. That background produced a simple rule for the agentic era.

"We as leaders need to think of AI agents no differently than digital employees. I treat any new AI agent as an intern within our organization. Early on, it requires a lot more oversight, and with that oversight comes a feedback loop," said Singh. Every AI initiative at Centria starts with a problem identified by its clinical and operational teams, not with the technology itself. Potential applications are then evaluated against the outcomes that matter most: improving clinical quality, expanding access to care, reducing administrative burden on clinicians and improving workforce productivity. "Putting the business value drivers first and mapping AI proofs of concept to those needs is how things move to production in a much more reliable, predictable manner," Singh said. Two early use cases set the pattern.

• Documentation that returns time: The first example Singh cites is clinical summarization that cuts the administrative load driving burnout. It shipped only after clearing internal AI governance processes that include compliance and clinical ethics, clinical quality standards, and time savings, a rising bar as ambient documentation tools draw legal scrutiny.

• Intake without the bottleneck: The next use case is AI-enabled intake. Hand-converting uploaded medical records into intake forms was slow and error-prone. Technology offers a big assist. "If you use AI to read that note in a HIPAA-compliant manner and AI does the form filling, it saves significant time and captures a much richer data set that drives a better clinical context while increasing speed," Singh said.

• ROI beyond the ledger: Singh points out that ROI isn't measured only as money saved. It can also mean reducing clinician burnout, returning time to care and creating greater workforce capacity," Singh said. The ultimate measure is whether Centria can expand access to high-quality care without compromising compliance, or the standards families and clinicians expect.

Scaling those wins took governance anchored by Centria’s AI Governance Committee and a Clinical Ethics sub-committee, the discipline enterprises now need for digital coworkers. "Not every use case is the right use case for AI," Singh said. Each candidate fell into one of four lanes: human-only, human-led, human-in-the-loop, or AI-led.

• Autonomy is earned: Direct patient care stayed human-only, administrative work ran human-led, and only repeatable back-office tasks went AI-led. New agents started under heavy supervision, with accuracy measured over weeks, and manager job expectations were modified to cover agent oversight, turning written policy into enforceable practice.

• New model, same probation: Upgrades, like swapping interns from different schools, restarted the clock. "Your intelligence level changes as you shift from one model to another. But you still need the same level of explainable testing to make sure it has not hallucinated and has not deviated," Singh said. His agentic orchestration platform, Workato, baked identity, permissions, and guardrails into that testing.

Underneath it all sat architectural discipline. Centria engineered workflows around the economics of every token, routed repeatable tasks to deterministic models, and logged explainable outputs for audit. "You need an architectural mindset to know when to use a rule-based engine and when to use an LLM reasoning engine, such that your output is more compliant, cost-efficient, and as deterministic as possible," he said.

None of it works without people willing to learn alongside the technology. "This technology is so transformational that you have to unlearn the way you worked before and develop the art of relearning with AI," Singh said. At Centria, that doesn't mean abandoning the expertise that already exists. It means pairing AI with engaged clinical and operational teams who help train, test and refine the tools around the realities of their work. That continuous feedback loop helps turn early AI applications into solutions that are better aligned to the needs of the people using them.

His parting advice inverts the usual complaint about unused AI rollouts. "Find the 10 or 20 percent who are using it and look at how they are using it. Those are the micro innovations happening in your field day in and day out. Use that data to create more scalable innovations you can launch," Singh said.

For Centria, success isn’t measured by how many AI agents are deployed or processes are automated, but by what those innovations make possible: less administrative burden on clinicians, greater capacity, and ultimately more families receiving the care and support they need. Already one of the nation’s largest ABA providers, Centria is emerging as an innovation leader by using technology to help close the distance between families searching for care and the clinicians ready to provide it.