As AI spreads across frontline software, who’s actually accountable for what it sees?

Opinions expressed by Digital Journal contributors are their own.

Ask most CIOs whether they can name every AI system with access to employee data, and the honest answer is usually no. A recent Deloitte survey found 95 percent of Canadian leaders confident in their organization’s AI compliance, while 64 percent admitted that some AI activity runs outside approved tools entirely. That gap, between confidence and actual accountability, is the starting point for how Tushneem Dharmagadda thinks about AI in workforce technology, and why he built his workforce experience and operations platform, HubEngage, around consolidating it rather than adding to it.

A governance problem wearing an efficiency costume

Most of the AI governance conversation happening at the board and CISO level right now focuses on customer-facing systems and core enterprise infrastructure, where the question of who’s accountable for an AI system’s behavior has at least become a standard part of the review process. Frontline workforce technology, the tools that manage scheduling, communications, engagement surveys, and day-to-day operations for hourly and distributed employees, has largely stayed outside that scrutiny, even though it touches data just as sensitive: location, performance feedback, communication history, scheduling patterns. Nobody is asking the accountability question there yet, mainly because nobody has been asking whether they should.

Dharmagadda has watched this play out from inside the industry. Organizations frequently end up running a separate AI feature inside each individual tool, a chatbot embedded in the communications platform, a sentiment engine inside the survey tool, a scheduling optimizer in a different system entirely, with no single team responsible for what any of them are actually authorized to see or do.

“The biggest disconnect: employees don’t hear directly from the C-suite, and the C-suite doesn’t hear directly from employees,” he says. “As companies grow, that disconnect becomes bigger and bigger.” He applies the same logic to AI oversight. The further removed leadership is from how a system actually operates day to day, the easier it becomes for that system’s data access to go unexamined, and the harder it becomes to answer for it later.

Why fragmented AI is harder to govern than consolidated AI

The mechanics of the problem are structural, not malicious. Communications tools get purchased by one department, engagement platforms by another, scheduling and operations systems by a third. Each purchase clears its own budget approval and, increasingly, its own AI feature bundled into the product. Almost none of those approval processes include a step where someone evaluates the AI’s data access against the other AI systems already running elsewhere in the organization, which means the accountability question never gets asked at the level where it would actually matter: across the whole employee population, not one tool at a time.

That fragmentation is exactly what shows up in the kind of shadow AI findings CIOs have been reporting throughout 2026: tools operating with access nobody centrally tracked, because no one purchase looked risky enough on its own to trigger a governance review. A frontline scheduling AI feature, evaluated in isolation, rarely looks like a security question. Evaluated alongside four or five other AI-enabled tools touching the same employee population, with no single owner able to answer for the combination, the picture changes.

The case for one governed layer

HubEngage’s architecture is built around a distinction the company calls native versus orchestrated: for smaller organizations, replacing scattered point tools with one platform outright, and for larger organizations already running systems of record like UKG, Workday, ADP, or Paylocity, sitting on top of them as a single frontline layer rather than asking employees and administrators to manage each system’s AI separately.

Dharmagadda frames the benefit less in terms of convenience and more in terms of who can actually answer for the system. A company running one consolidated AI layer across frontline communications, scheduling, and engagement has one place to document what that AI is authorized to access and one team accountable for it. A company running the same functions across four or five separately administered tools has four or five different answers, assuming anyone has asked the question of each one individually, which is precisely the gap the Deloitte numbers suggest most organizations haven’t closed yet.

That framing lines up with what HubEngage has been building publicly under what it calls a workforce operating system, a control layer intended to sit above fragmented point tools rather than add another one to the pile.

Where this intersects with the broader AI governance conversation

The pattern Dharmagadda describes mirrors what’s shown up in Canadian AI governance research more broadly this year: confidence in oversight consistently outpaces actual visibility into what AI systems are doing. Most of that conversation to date has centered on knowledge work and customer-facing systems, where the accountability question is now routine. Frontline and distributed employees, who make up a substantial share of the workforce in retail, healthcare, hospitality, and manufacturing, are typically managed through software that hasn’t been pulled into that same governance lens yet, in part because it’s rarely viewed as a single operational layer rather than several disconnected ones.

Dharmagadda’s bet is that this changes as AI governance scrutiny matures and extends further down the technology stack. Organizations that have already consolidated their frontline AI into something one team can document and govern will have a straightforward answer ready when someone finally asks who’s accountable. Organizations still running AI across five separately purchased tools will be doing that consolidation work under scrutiny instead of ahead of it.