When AI Runs the Workflow, Mentorship Becomes an Operating Requirement
The human-AI partnership is very much a partnership, but you cannot simply say you are going to augment a process without knowing how the original process was built.
Domenico Merlino
Workforce and Talent Strategy Leader
Replacing experienced employees with AI agents can look like a clean win on the labor math. Run the FTE calculation, subtract the headcount the agent absorbs, and the savings appear immediately. What the calculation misses are two risks that surface later and cost far more: AI agents operating inside the business with no accountable owner, and the permanent loss of the institutional knowledge required to govern them. As companies cut jobs, they often inadvertently eliminate the senior talent who understand why the workflows were built the way they were, and then discover that no one left in the building can explain the intent behind the process the agent is now running.
Domenico Merlino is a workforce and talent strategy leader who spent 25 years at IBM helping build the workforce management platforms the company uses today, guiding the organization through its own human-digital work transformation. He writes and advises on AI governance, agentic accountability, and the talent architecture that supports both. Merlino notes that the industry’s current approach to AI often overlooks the historical context needed to train the next wave of managers.
“The human-AI partnership is very much a partnership, but you cannot simply say you are going to augment a process without knowing how the original process was built. You need the institutional knowledge behind the workflow to understand how AI should improve it,” he says. That knowledge is precisely what gets discarded when organizations move to cut talent first and ask questions later.
You can’t download wisdom
Merlino’s central objection to premature senior layoffs is that the judgment behind a workflow doesn’t live in documentation, but in the people who built it and broke it and fixed it. “You can’t download wisdom. As we work through the transformations, it’s critical to have senior leaders who have gone through the pain to teach the newer generation what the workflows are that we will automate or benefit by AI.”
The people entering the workforce now bring real advantages. They’re digital natives, ambitious about the technology, and spin up on new tools quickly. What they can’t acquire on their own, he points out, is the business context underneath the tooling. “That still requires the knowledge of what a business workflow does,” Merlino says. “What’s the intention of it? The outcome expectation? How do you measure success in the workflow?”
In his experience, the consequence of skipping that transfer is already visible. “I’m seeing a pattern of eliminating senior talent perhaps prematurely, and in some cases you’re seeing it now in the public domain. Some companies are regretting doing that and they’re having to hire back consultants to help build processes to understand how the workflows will be redesigned and reimagined with AI in the loop.”
Paying consultants to reconstruct knowledge you already owned is an expensive way to learn that the labor math was incomplete. It also compounds a labor market where younger workers are struggling to find entry points, because the roles where they would have learned the business are the ones being automated away before the knowledge transfer happens.
Orphan AI is a governance failure waiting for an auditor
The second risk Merlino points to is structural. Every agent making decisions inside a business needs a documented owner, and he’s blunt about what happens without one. “If that person leaves your organization and the registry is not a solid, governed registry where you have someone replace the person that owned that agent, you have an orphan AI out there. You won’t ever know who made the decision unless you have a tight registry of ownership within the workflow.”
The registry has to be more than a name on a page. It requires evidence, measurement, and succession. “You can have great policies on paper, and even before agentic, governance is challenging. How do you actually prove that an owner on paper is actually meeting their responsibilities? There has to be an audit trail.”
That distinction between an AI policy that exists and ownership that is provable is where most organizations struggle right now. Merlino ties it directly to individual accountability. “Having a measurement to each person, a key performance indicator, is critical. If you don’t measure, you’ll never know, and a person will always say, ‘Well, I wasn’t being measured on that.’ That’s a huge gap.”
The urgency is regulatory as much as operational. Under the EU AI Act and similar frameworks, the question is not whether AI participated in a decision, but whether you can produce the evidence. “When an auditor comes in and asks how you made the decision to let go of these people, and the answer is through AI, the next question is going to be, ‘Show me the evidence of how that decision was made.’ That’s a massive risk for organizations.”
Match oversight to risk, not to fear
Merlino’s governance model tiers oversight by risk tolerance so that scrutiny scales with consequence rather than slowing everything down uniformly. Low-risk agents that update logs or draft summaries need minimal auditability. Medium-risk decisions, like a lateral transfer, require manager approval. High-risk transactions demand gated human review. “You definitely need more humans in the loop around those agents, with risk tolerance thresholds in place, so you don’t slow down AI but you have a mechanism to go back and understand the decision,” he advises.
That tiering is what separates governed adoption from the aspirational rush he sees in many organizations, where the pace of AI adoption has outrun the rigor. It also connects to a broader push to build trust in AI agents, because the enhanced risks that come with agentic capability are proportional to the autonomy granted.
Transparency to employees is part of the same obligation. “Employees have the right to understand how the AI is being used against their digital footprint,” Merlino says. “Having all those areas of visibility is vital.”
The CHRO mandate: connect talent governance to AI governance
Merlino places the CHRO at the center of this, because the registry of who owns what is fundamentally a people question. “The CHRO definitely needs a seat at the table. They don’t always have one, but the rigor of their responsibility is massive. They own the registry of the people and who owns the outcomes across the whole C-suite.”
That means ensuring domain leaders in finance, HR, and elsewhere actually have the capability to govern these processes, which turns how HR defines workplace AI into an operating requirement rather than a communications exercise. It also means the connective tissue across functions has to be deliberate. “Transformation without overlap and connective tissue is not transformation. It’s siloed initiatives.”
The talent side of the mandate is rebuilding what has eroded. Merlino sees mentorship programs as the mechanism for extracting tacit knowledge that no document captures. “That wisdom comes from people in senior positions who have that tacit knowledge you can’t always document. It’s only really achieved through conversation, working side by side with colleagues and senior leadership.”
He notes that the newer generation wants this. They want accountability and want to understand how processes work. The gap is that senior talent often has no structured mechanism to demonstrate and transfer what they know, which makes evidence-based, outcome-oriented mentorship and deliberate succession planning an operating necessity rather than a nice-to-have. “Without that, the pipeline of senior leaders just starts to dry up. You can’t run a business that way.”
The organizations that get this right will be the ones with explainability: how AI transforms the workforce, how the workforce benefits, and how senior talent gets the opportunity to train the leaders who will manage the agents. Those leaders won’t learn the judgment overnight, and the internal champions who help convert skeptics are far more effective when the underlying knowledge transfer has actually happened. The alternative is what Merlino warns against directly. “Not just blindly saying AI is going to replace everybody. That’s a problem.”
Related articles
TL;DR
Domenico Merlino
Workforce and Talent Strategy Leader
Workforce and Talent Strategy Leader
Replacing experienced employees with AI agents can look like a clean win on the labor math. Run the FTE calculation, subtract the headcount the agent absorbs, and the savings appear immediately. What the calculation misses are two risks that surface later and cost far more: AI agents operating inside the business with no accountable owner, and the permanent loss of the institutional knowledge required to govern them. As companies cut jobs, they often inadvertently eliminate the senior talent who understand why the workflows were built the way they were, and then discover that no one left in the building can explain the intent behind the process the agent is now running.
Domenico Merlino is a workforce and talent strategy leader who spent 25 years at IBM helping build the workforce management platforms the company uses today, guiding the organization through its own human-digital work transformation. He writes and advises on AI governance, agentic accountability, and the talent architecture that supports both. Merlino notes that the industry’s current approach to AI often overlooks the historical context needed to train the next wave of managers.
“The human-AI partnership is very much a partnership, but you cannot simply say you are going to augment a process without knowing how the original process was built. You need the institutional knowledge behind the workflow to understand how AI should improve it,” he says. That knowledge is precisely what gets discarded when organizations move to cut talent first and ask questions later.
You can’t download wisdom
Merlino’s central objection to premature senior layoffs is that the judgment behind a workflow doesn’t live in documentation, but in the people who built it and broke it and fixed it. “You can’t download wisdom. As we work through the transformations, it’s critical to have senior leaders who have gone through the pain to teach the newer generation what the workflows are that we will automate or benefit by AI.”
The people entering the workforce now bring real advantages. They’re digital natives, ambitious about the technology, and spin up on new tools quickly. What they can’t acquire on their own, he points out, is the business context underneath the tooling. “That still requires the knowledge of what a business workflow does,” Merlino says. “What’s the intention of it? The outcome expectation? How do you measure success in the workflow?”
In his experience, the consequence of skipping that transfer is already visible. “I’m seeing a pattern of eliminating senior talent perhaps prematurely, and in some cases you’re seeing it now in the public domain. Some companies are regretting doing that and they’re having to hire back consultants to help build processes to understand how the workflows will be redesigned and reimagined with AI in the loop.”
Paying consultants to reconstruct knowledge you already owned is an expensive way to learn that the labor math was incomplete. It also compounds a labor market where younger workers are struggling to find entry points, because the roles where they would have learned the business are the ones being automated away before the knowledge transfer happens.
Orphan AI is a governance failure waiting for an auditor
The second risk Merlino points to is structural. Every agent making decisions inside a business needs a documented owner, and he’s blunt about what happens without one. “If that person leaves your organization and the registry is not a solid, governed registry where you have someone replace the person that owned that agent, you have an orphan AI out there. You won’t ever know who made the decision unless you have a tight registry of ownership within the workflow.”
The registry has to be more than a name on a page. It requires evidence, measurement, and succession. “You can have great policies on paper, and even before agentic, governance is challenging. How do you actually prove that an owner on paper is actually meeting their responsibilities? There has to be an audit trail.”
That distinction between an AI policy that exists and ownership that is provable is where most organizations struggle right now. Merlino ties it directly to individual accountability. “Having a measurement to each person, a key performance indicator, is critical. If you don’t measure, you’ll never know, and a person will always say, ‘Well, I wasn’t being measured on that.’ That’s a huge gap.”
The urgency is regulatory as much as operational. Under the EU AI Act and similar frameworks, the question is not whether AI participated in a decision, but whether you can produce the evidence. “When an auditor comes in and asks how you made the decision to let go of these people, and the answer is through AI, the next question is going to be, ‘Show me the evidence of how that decision was made.’ That’s a massive risk for organizations.”
Match oversight to risk, not to fear
Merlino’s governance model tiers oversight by risk tolerance so that scrutiny scales with consequence rather than slowing everything down uniformly. Low-risk agents that update logs or draft summaries need minimal auditability. Medium-risk decisions, like a lateral transfer, require manager approval. High-risk transactions demand gated human review. “You definitely need more humans in the loop around those agents, with risk tolerance thresholds in place, so you don’t slow down AI but you have a mechanism to go back and understand the decision,” he advises.
That tiering is what separates governed adoption from the aspirational rush he sees in many organizations, where the pace of AI adoption has outrun the rigor. It also connects to a broader push to build trust in AI agents, because the enhanced risks that come with agentic capability are proportional to the autonomy granted.
Transparency to employees is part of the same obligation. “Employees have the right to understand how the AI is being used against their digital footprint,” Merlino says. “Having all those areas of visibility is vital.”
The CHRO mandate: connect talent governance to AI governance
Merlino places the CHRO at the center of this, because the registry of who owns what is fundamentally a people question. “The CHRO definitely needs a seat at the table. They don’t always have one, but the rigor of their responsibility is massive. They own the registry of the people and who owns the outcomes across the whole C-suite.”
That means ensuring domain leaders in finance, HR, and elsewhere actually have the capability to govern these processes, which turns how HR defines workplace AI into an operating requirement rather than a communications exercise. It also means the connective tissue across functions has to be deliberate. “Transformation without overlap and connective tissue is not transformation. It’s siloed initiatives.”
The talent side of the mandate is rebuilding what has eroded. Merlino sees mentorship programs as the mechanism for extracting tacit knowledge that no document captures. “That wisdom comes from people in senior positions who have that tacit knowledge you can’t always document. It’s only really achieved through conversation, working side by side with colleagues and senior leadership.”
He notes that the newer generation wants this. They want accountability and want to understand how processes work. The gap is that senior talent often has no structured mechanism to demonstrate and transfer what they know, which makes evidence-based, outcome-oriented mentorship and deliberate succession planning an operating necessity rather than a nice-to-have. “Without that, the pipeline of senior leaders just starts to dry up. You can’t run a business that way.”
The organizations that get this right will be the ones with explainability: how AI transforms the workforce, how the workforce benefits, and how senior talent gets the opportunity to train the leaders who will manage the agents. Those leaders won’t learn the judgment overnight, and the internal champions who help convert skeptics are far more effective when the underlying knowledge transfer has actually happened. The alternative is what Merlino warns against directly. “Not just blindly saying AI is going to replace everybody. That’s a problem.”