Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

As the AI experimentation era ends, a manager’s work begins

Prepare to reimagine your career.

[Photo: Getty Images]

Profile photo of Ee Lyn Khoo

EVP and Chief People Officer · Avalara

Many leaders are feeling pressure to show fast return on investment (ROI) on agentic AI initiatives. The instinct is to point an agent at an existing process and declare victory: “Now it’s faster!”

But workflow orchestration still matters, even when a machine performs it. This moment is about reimagining work end to end and understanding who owns the outcome when a person-agent combination produces it together.

AI’s boundless experimentation era ends

For the past couple of years, “we’re experimenting” was a legitimate response when asked where a company stood in its AI journey. It bought people and HR some time, and we learned a lot about where AI can potentially do work better.

However, that era is ending. We’re shifting from optionality to obligation, and that raises the stakes. According to PwC’s 2026 Global CEO Survey, 56% of CEOs say they have seen no significant financial benefit to date, either revenue gains or cost reductions.

How can that be? It means a lot of companies bought tools and skipped the work.

First, the tools aren’t free. In fact, around two out of three companies are spending at least $1 million on AI technology, according to a 2026 study commissioned by Writer. Token costs eat into your returns, along with security and governance reviews, data cleanup, and training.

Second, we’re too often layering a faster tool on top of processes designed for humans. When we narrowly focus on faster task completion, we miss the chance to redesign the work. A better approach allows humans to focus on higher-value work.  

Third, some savings were never real. Someone gets a draft of a report in 30 seconds and spends an hour correcting the output and feeding more prompts. Call it bot-sitting.

For savings to reach the profit-and-loss line, someone has to decide what to do with reclaimed capacity: take on more volume, redeploy a team to deferred work, or stop doing something entirely. That’s a management decision, not a technology one, and if nobody makes it, the time just vanishes.

Manage the human-agent combination

Making those decisions well requires grasping the human-agent combination. If you only review the person, you can’t tell whether good outputs reflect strong judgment or a good tool is covering weak judgment. You certainly can’t tell how much capacity actually opened up.

Instead, make the human-agent combination the unit of accountability. It forces a named owner to answer two questions: What did the human-agent system produce, and what did we do with the capacity it freed up? It also forces measurement onto outcomes like speed to resolution or customer impact.

Managers can’t outsource this to IT or a vendor. Today, when an agent doesn’t deliver, the instinct is to escalate to engineering. But if it’s producing your team’s work, it’s your outcome.

That means managers need judgment skills they’ve never needed before: When to trust an agent’s output and when to interrogate it, when a team is spending more time supervising a system than doing the work, and where the human has to stay in the loop.

Reimagine work, don’t redesign it

Of course, ownership without reimagining the work is just accountability for a mess.

I use that word deliberately: reimagining, not redesigning. Redesigning a process makes it better. Reimagining treats the outcome as fixed and the process as entirely negotiable. Reimagining is where the biggest payoffs from AI live.

Take product discovery. Redesigning might use AI to summarize support tickets and partner calls into a nicer dashboard of themes. Reimagining starts with understanding the customer correctly, as early as possible. Reframing that way changes everything. You might skip the requirements document and use AI agents to turn customer and partner feedback directly into working prototypes. We’re doing exactly that.

The caveat is that you can’t reimagine everything. It’s disruptive and takes real leadership attention. Pick a small number of processes where the outcome truly matters and clearly reflects how work used to get done, like batch work that could run in real time.

Don’t slow down, just know where you’re headed

The pressure to show returns from AI investments is real, because it’s a genuinely disruptive and potentially transformative technology. But the return comes from deciding what the work should actually look like now, naming the outcome owner when a person and an agent produce it together, and being honest about the difference between a process that gets faster and a process that gets better.

Companies that skip that step will keep buying capabilities that never convert to ROI. Those that do the work will end up with something considerably harder for the competition to replicate than anything they purchased.

Ee Lyn Khoo is the executive vice president and chief people officer at Avalara.