AI tools each work well, but they stay disconnected—leaving employees to bridge them by hand. The AI worth having carries that invisible work itself.
Laila Almounaier
VP of Connected Experiences
Workday
Early in my career, I was leading mobile strategy at a global retailer when we noticed a disconcerting customer habit. Customers were filling their carts in our app, then closing it without buying. There were no bugs, and the checkout system worked. We were worried. Why didn't customers check out? We quickly realized they were pre-shopping: browsing on their phones and saving what they liked, coming into the store to look for what they knew they wanted, touch fabric, and check the fit. The disconnect wasn't in our technology. It was between the digital and the tangible customer experience. We just hadn't connected the two of them yet.
So we did. We built tools for customers to save items and try them on in store, or buy online and pick it up in store, and even helped me navigate the store to find the items they had viewed online. It not only worked, but it improved sales, customer loyalty and satisfaction. Once we understood behavior, we could create an environment and an experience that catered to that.
That's where we are with AI right now. Tools are good individually, but collectively disconnected — and people are doing the bridging by hand. Your store's app can tell a customer an item is ready for pickup, but it still leaves an associate to track it down on the floor. It's never been this easy to share data, but our tools are still focused on singular transactions and missing the broader context.
I think about that retail story a lot these days. The shape of the problem feels familiar: technology moving fast, people adapting faster, and the systems around them struggling to catch up.
Report
Recently, I was hiring for a new role — nothing unusual, just the ordinary work of building a team. Not long ago, that same week would have turned into a disjointed project of me using every AI tool I had, all of which worked well. But I'd have to stitch the pieces together by hand.
This time was different. My requisition tool picked up the headcount number the moment the planning tool had generated it — no copying, no reconciling. The candidate's interview notes moved straight from my AI notetaker into the agent drafting follow-up emails, already framed the way that agent needed them. Each tool didn't just do its own job well, it also handed off what it knew to the next one.
That's a small thing when it happens once. It's a different way of working when it happens across every hire, every meeting, every handoff a company runs — when the connective work nobody put on anyone's job description simply isn't there to do anymore. Most companies aren't there yet.
A study Workday published this year found that 82% of employees spend significant time acting as translators between AI tools and systems, manually copying and reconciling information the tools should have shared on their own. One in five are losing more than seven hours a week to it, according to our research. This isn't isolated to one team or one function. It is showing up in HR, finance, IT — wherever people pick up good AI tools and are left to introduce them to each other.
Giving an AI agent access to a system's data is not the same as giving it access to the rules that govern that data.
Giving an AI agent access to a system's data is not the same as giving it access to the rules that govern that data. Most agentic AI treats the two as interchangeable, which is exactly where it falls short of its promise.
Take a manager approving a colleague's time off. The accrued balance is data. Whether a blackout period applies, whether the manager's delegation is currently active, who else on the team is already out, whether this particular request would violate a policy nobody wrote down explicitly but everyone enforces — that's judgment. And it lives in the rules a company has spent years encoding into its systems of record.
An agent that can see the balance but not the rules will still answer confidently. It just won't be right, and it won't know that it's wrong.
An agent that can see the balance but not the rules will still answer confidently. It just won't be right, and it won't know that it's wrong. At the scale of one approval, that's a minor headache. At the scale of every approval, exception, and unwritten policy an enterprise runs on, it's the difference between an AI program employees trust and one they quietly learn to double-check.
If I were evaluating an AI vendor right now, I'd ask two questions.
● Can it see the rules that govern a decision, not just the data behind it?
● When it hands a task to another system, does that context remain intact, or does a person have to rebuild it on the other side?
In every vendor conversation I've had this year, the first question gets a confident answer. The second, almost every time, gets a pause.
Most vendors aren't being evasive here. Accurately tracking whether context survives a handoff is a genuinely hard engineering problem, especially at the daily volume of tasks IT, finance, and HR teams run. But that difficulty doesn't change what it means for the person using the tool. An agent that can see data but not rules will still act, and act confidently, which is worse than an agent that visibly hesitates.
A tool that pauses at least tells you where its limits are. Whereas, a tool that answers anyway, using rules a person supplied from memory, is quietly asking employees to keep doing the same invisible work — and quietly asking the business to absorb whatever that work gets wrong.
This isn't a new problem. Every generation of technology has needed someone to do the connecting work it didn't yet know how to do itself — whether between a mobile app and a store shelf, or between one AI tool and another. What's different now is that we already know the fix, because we've built it before: judgment has to travel with the work, not get reassembled by whoever happens to be holding the pieces.
Judgment has to travel with the work, not get reassembled by whoever happens to be holding the pieces.
We didn't close the gap between a phone and a store shelf by making the app better. We closed it by connecting the digital with the tangible experiences: reserving inventory, confirming pickup, syncing what a shopper saw online with what a store had on the shelf. The result? None of this became the customers' problem to solve.
That standard is worth holding enterprise AI to as well. Run the math on that earlier statistic — seven hours a week, multiplied across a workforce of thousands — and the invisible work stops looking like a personal inconvenience and starts looking like a line item: hours an AI investment was supposed to give back spent reassembling what the tools should have shared on their own.
AI with real value is the one that carries that invisible load instead of leaving it to the workforce.
AI with real value is the one that carries that invisible load instead of leaving it to the workforce — the one whose judgment travels with the work, across every handoff, every approval, every system it touches. That's the real test of an AI strategy, and it's worth applying before the next vendor demo: ask the two questions above, and see whether the answers hold up once the work actually starts moving.
Report
A gap between effort and impact is the real ROI problem. Discover the solution.