By Dheeraj Toshniwal, CEO at Powerweave
The enterprise AI debate keeps asking what machines can do. The better question is what organisations should still leave to people. AI no longer just recommends actions. It increasingly takes them. So the old idea of keeping a human at the end of a workflow isn’t enough anymore. A person who simply approves an AI decision isn’t really providing oversight. The real challenge is redesigning work itself. AI should handle speed, scale and repetitive execution. People should stay responsible for context, judgement and consequences.
This shift is already happening. McKinsey’s 2025 State of AI survey found that 62% of organisations are experimenting with AI agents. 23% say they’re already scaling an agentic system somewhere in the business. But most companies are still early in this journey. They haven’t captured much value yet. That gap matters. Enterprises need to think about where humans fit in before deploying an agent. Not after something goes wrong.
Human oversight can’t be an afterthought
Oversight is usually treated as a simple checkpoint. AI makes a recommendation. A person reviews it. The process moves on. This doesn’t work at enterprise scale. People can’t review every AI decision at the speed and volume these systems operate at. If every action needs manual approval you just recreate the bottlenecks AI was meant to remove.
A better model builds human judgement into the process from the start. Enterprises need to decide which decisions are routine and can be automated. Which ones need escalation. And which should always remain human led.
Take procurement function as an example. AI can analyse supplier data, compare options and flag risks. But a procurement leader still needs to judge when a supplier relationship is strategic. Or when an exception changes the bigger picture in ways the data can’t capture. The human’s value isn’t checking every transaction. It’s owning the decisions where context matters most. That is a more useful way to think about human-in-the-loop AI. The human is not standing at the end of the process waiting to approve whatever the machine has done. The workflow is redesigned in such a way that humans and AI work together for the best outcome.
The same idea applies in finance, customer operations and HR. AI can prepare, analyse, recommend and even execute within set boundaries. That should make human ownership more deliberate, not less important.
Using AI isn’t the same as gaining value from it
Here’s something enterprises can’t ignore. Deploying AI doesn’t automatically create business value.
PwC’s 29th Global CEO Survey found only 12% of CEOs say AI has delivered both cost and revenue benefits. 33% report gains in one or the other. 56% say they’ve seen no significant financial benefit so far. PwC also found that companies with stronger AI foundations tend to see better returns. Things like responsible AI frameworks and systems built for enterprise wide integration made a real difference.
The lesson is simple. The question shouldn’t be how fast an organisation can deploy an agent. It should be whether the process is actually ready to benefit from one. Success needs to be measured by business outcomes rather than deployment numbers. If an agent speeds up a workflow but doesn’t improve customer experience, reduce risk, sharpen a decision or create real efficiency it’s hard to call that transformation.
Don’t automate a broken process
There’s an old lesson from enterprise transformation that applies just as much to AI. Technology doesn’t fix a weak process on its own. If a workflow already runs on spreadsheets, emails and manual follow ups adding AI just makes that mess move faster.
Before deciding where to deploy AI, leaders should ask simpler questions first. What outcome are we trying to improve? Which steps are repetitive? Where do exceptions happen? What data does this process depend on? Who owns the decision when something goes wrong?
The answers should shape the technology choice not the other way around. This also means companies don’t always need to replace systems that already work. AI can often act as an intelligence layer around existing systems. It can improve workflows without forcing a disruptive overhaul. The goal is better outcomes, not more technology.
Adoption happens when AI fits how people actually work
A system going live isn’t the same as a successful transformation. A platform can be launched, employees can be trained and dashboards can go up. Yet teams keep using old spreadsheets, offline approvals and familiar workarounds. The real test comes on a busy working day. When data is incomplete or an exception appears or the new workflow feels harder than the old one.
People don’t resist technology because it’s new. They resist friction. If a system doesn’t reflect how work actually happens people will find another way to get the job done. The strongest AI deployments are designed with the people who use them. Not simply handed to them.
The future is human plus AI
The organisations that pull ahead won’t be the ones deploying the most AI agents. They’ll be the ones that understand where automation creates value and where human judgement is important.
This calls for a different way of thinking about enterprise AI. The goal isn’t keeping a human standing behind every automated action waiting to step in. It’s giving the right people real decision rights at the points where their expertise matters most. While letting AI handle routine bounded work at greater speed and scale.
The future of enterprise AI was never really a choice between people and machines. It’s an operating model where each does what it’s best placed to do. AI can extend human capability. It cannot replace the responsibility that comes with making decisions in the real world.
Humans in the loop shouldn’t be a safety net stretched across everything an algorithm might get wrong. Human intelligence should be a deliberate design choice. Built into the workflow wherever context accountability and consequence genuinely require it.
That is what people led AI enabled transformation looks like. Not replacing human judgement. Giving it greater reach.