Ambient Effort™: Agents on the Org Chart — Who's the Boss?
- carolmastrofini
- Apr 29
- 17 min read
Updated: May 8
The Human Roles Agentic AI Actually Requires
Article 3 of 3 — Ambient Effort™: The Agentic Series
A note before we go further.
This article looks at what happens when organizations ask a question the technology industry has been slow to answer honestly: not whether we can build agents, but who manages them once we have. It is written from the same vantage point as Articles 1 and 2 — observational, not prescriptive. There is much we don't yet know. What follows is what I'm seeing, what the research is showing, and what the organizations navigating this well are doing differently.
Breaking and broken are not the same condition. One is recoverable. The other compounds.
On Tuesday morning, April 22, 2026, Sundar Pichai took the stage at Google Cloud Next in Las Vegas — the world's largest enterprise AI conference — and said something that should stop every organization in its tracks:
"The conversation has gone from 'Can we build an agent?' to 'How do we manage thousands of them?'"
He said it as progress. And technologically, it is.
But that is not a technology question. It never was. It is a human question — about roles, accountability, judgment, and the organizational infrastructure required to hold what autonomous systems produce. And for most organizations deploying agents right now, the honest answer is: we don't know yet. We haven't designed for it. We haven't trained for it. We haven't even named the roles that would make it possible.
Thomas Kurian, Google Cloud CEO, said it from the same stage: "You have moved beyond the pilot. The experimental phase is behind us. And now the real challenge begins: How do you move AI into production across your entire enterprise?"
That challenge is not technical. It is human. And the research confirms the gap is real: nearly 74% of companies plan to deploy agentic AI within two years — yet only one in five reports having a mature model for governance. Only 11% have implemented governance frameworks of any kind, despite rapid deployment growth.
The agents are already here. The humans who govern, design, build, and manage them are not.
This is the final article in the Ambient Effort™: The Agentic Series. Articles 1 and 2 named the exhaustion and the structural failure. This one answers the question that follows naturally from both:
What does readiness actually look like — and who are the humans that make it possible?
A Framework the Industry Needs But Hasn't Named
What follows is not a survey of established roles. These titles do not yet exist on most org charts. There are no standard job descriptions. Most of the people doing this work are doing it without the authority, the clarity, or the organizational recognition that would make it more effective
This is a proposed framework. It is grounded in research, informed by what the organizations navigating this well are actually doing, and offered as a more precise and more effective approach than what the industry is currently using.
The industry has begun describing a pre-deployment phase for agents as "onboarding". Treating this phase, as Harvard Business Review argues, the way you'd onboard a new employee. That instinct is correct. Agents do need structured preparation before they go live: defining what the agent will do, its scope, its boundaries, its success criteria, and its escalation paths — and then building it.
What's important to understand is that for the majority of agent builds today, this doesn't require significant technical expertise. Tools like Google's Agent Designer, announced at Cloud Next this week, have made agent building increasingly accessible to domain experts — people who understand the workflow deeply, not engineers who build systems. One person with the right domain knowledge can both define and build the agent.
For more complex deployments — multi-system integrations, regulated industries, high-risk enterprise environments — the technical demands may warrant separating these functions. That's when dedicated technical expertise becomes necessary, working from specifications provided by a domain expert.
The framework this article proposes reflects that reality: one design role for the majority of builds, two roles when complexity genuinely demands it.
The Four Human Roles Agentic AI Actually Requires
1. The Agent Policy Director
This role sets the legal, ethical, and strategic framework for how agents operate within the organization. It doesn't have to sit in the C-suite but it advises the C-suite, and its decisions shape everything downstream.
Legal: The Agent Policy Director understands and manages the liability that comes with agent deployment. The EU AI Act's core enforcement framework becomes broadly operational on August 2, 2026 — though a potential delay is under consideration via the EU's Digital Omnibus package. The Colorado AI Act takes effect June 30, 2026. The UK's Competition and Markets Authority has already ruled that the organization deploying the agent bears full legal responsibility for what it does — not the platform, not the model provider. The organization that pressed deploy. Most organizations have no one explicitly holding this responsibility.
Ethical: What are agents permitted to do — and what requires human sign-off? What values guardrails govern how agents interact with customers, employees, and sensitive data? What happens when an agent produces output that is technically correct but ethically questionable? These decisions get made whether or not someone has been designated to make them. The Agent Policy Director makes them deliberately, before deployment, rather than reactively after something goes wrong.
Strategic: How does the organization communicate what's changing — to its people, its clients, its stakeholders — before the change happens rather than after resistance has formed? What is the sequencing of agent deployment? Which workflows get agents now and which stay human? The communication strategy that determines whether the organization absorbs the technology or fractures under it lives here.
The Agent Policy Director is the role most consistently absent from agent deployments right now. And its absence is not just a governance gap. It is a growing legal liability arriving on a specific and imminent timeline.
2. The Agent Designer
For the majority of agent builds, this is the person who both defines what the agent will do and builds it.
The industry has been calling this phase "onboarding" — and the instinct is right. But what the industry often misses is that for most builds, this doesn't require a technical specialist. Google's no-code Agent Designer and similar tools have made agent building genuinely accessible to people who understand workflows deeply but don't write code. Domain expertise is the primary requirement. Technical fluency is secondary — and increasingly, the tools handle what once required engineering.
The Agent Designer answers the same questions you'd answer before hiring anyone: What is this agent's role? What are its boundaries? What does success look like? When does it escalate to a human, and what context does it pass along? Then — using the available tools — they build it.
This is the practical reality for most organizations right now. One person, deep domain knowledge, sufficient comfort with modern no-code tools. The Agent Designer role is where most of the pre-deployment human work happens.
For complex deployments — multi-system integrations, regulated industries, high-risk enterprise environments — the technical demands may warrant separating the design function from the build function. In those cases, the domain expert defines the role and a dedicated technical specialist builds it. But this is the exception, not the rule. And treating every deployment as if it requires that separation adds cost and complexity that most organizations don't need.
3. The Agent Role Architect
For complex deployments — multi-system integrations, regulated industries, high-risk enterprise environments — the design and build functions may need to be separated.
In these cases, the Agent Role Architect takes on the definition work: what the agent does, its scope, its boundaries, its success criteria, and its escalation paths. This is pure domain work — no technical expertise required. The Agent Role Architect knows the workflow deeply enough to translate it into precise specifications that a dedicated technical builder can work from.
The distinction matters in complex environments because the stakes of a poorly specified agent are higher. A vague brief that works fine for a simple customer service routing agent can produce real damage in a financial compliance workflow or a healthcare triage system. When the cost of getting it wrong is significant, separating the domain expertise from the technical build produces better agents and clearer accountability.
For most organizations, most of the time, the Agent Designer handles both functions. The Agent Role Architect is the exception the Agent Designer becomes when the work demands it.
4. The Agent Manager
This is the role Harvard Business Review formalized in February 2026, that Salesforce has already deployed at scale, and that Google built an interface for this week — the Agent Inbox, a centralized dashboard for monitoring, guiding, and managing agent activity, with notifications categorized as "needs your input," "errors," or "completed jobs."
HBR's framing is precise: agent managers are to AI what product managers were to software. They don't build the technology. They define what it should do, monitor whether it's doing it, and intervene when it isn't.
Zach Stauber, a support agent manager at Salesforce, described his day plainly: "Data, data, data. I start and end my day in dashboards, scorecards, and agent observability." His job looks less like a software engineer's and more like a floor manager's — except the floor is digital and the team is made of agents.
The six competencies HBR identifies for effective Agent Managers:
Translating business goals into agent instructions — what does good output look like for this workflow?
Performance monitoring — reading dashboards, scorecards, and observability tools
Escalation design — knowing when the agent should stop and route to a human, and ensuring the handoff carries the right context
Process documentation — making sure the workflow is defined clearly enough for the agent to follow consistently
Performance accountability — owning the business case, tracking what the agent fleet actually produces
Continuous improvement — operating in weekly test-deploy-learn cycles, adjusting as the agent and the workflow evolve
What this role does NOT currently require: coding. Deep technical expertise. Years of domain knowledge built over decades. The best Agent Managers come from operations, project management, and business analysis backgrounds. They understand the workflow. They know what good looks like. They make judgment calls about when something needs human attention.
This is a generalist role. A project manager type. Someone who manages workflows, reads data, makes triage decisions, and escalates appropriately.
It is also explicitly transitional. Managing agents already exist in multi-agent architectures — coordinating workflows, invoking sub-agents, reconciling outputs autonomously. The human Agent Manager is building what the autonomous system will eventually need: documented workflows, governance frameworks, institutional knowledge, and the organizational trust that authorizes the transition. Organizations that skip this phase don't just have a gap today. They have no foundation for the autonomous future they're building toward.
Average salary: $103,000. Already on job boards. The most accessible of the four roles — and the most important entry point into the agentic organization for the humans who will shape what comes next.
Ambient Effort™ and Closing the Expertise Gap
The expertise gap doesn't just create organizational risk. It creates Ambient Effort™ — the ongoing energy people expend trying to stay competent in a field that is moving faster than any training program can track. For the people being asked to manage agents without adequate preparation, that gap is felt daily: in the anxiety of not knowing if the output is right, in the vigilance of watching for drift without knowing exactly what to watch for, in the weight of being accountable for something they were never fully trained to hold.
Closing that gap isn't just a training problem. It is a human readiness problem. And it requires the same honesty we've brought to everything else in this series.
Only 14% of US workers report having adequate AI skills for the agentic moment. 62% of employers cannot find candidates with the skills they need. And most of what organizations are calling "AI training" is solving the wrong problem.
The critical distinction almost no organization is making: the difference between standard upskilling and what researchers are calling Agentic Fluency.
Standard upskilling teaches people how to use a tool. Agentic Fluency teaches people how to steward a digital workflow — how to read agent behavior, catch drift before it compounds, know when technically correct output is organizationally wrong, and make the escalation judgment calls that keep autonomous systems operating within their intended boundaries.
What actually works: training embedded directly into real workflows, not standalone events. Role-specific practice in sandbox environments where people can make mistakes without consequences. Mentoring relationships rather than self-directed modules. And — critically — psychological safety established before training begins.
You cannot build Agentic Fluency in a workforce that is too afraid to fail. The fear has to be named and addressed before the curriculum begins. Organizations that skip this step find their training programs generating no traction — not because the content is wrong, but because the humans receiving it are not in a condition to absorb it.
What consistently fails: one-time events. Generic AI literacy programs that explain what large language models are rather than how to manage an agent in a specific workflow. Check-the-box training videos. And any program that ignores the emotional reality of the people in the room.
The timeline is uncomfortable. Leading organizations are targeting six months to close meaningful gaps. Most organizations haven't started.
The Gen Z Moment — Honest About All of It
Let's be honest about what Gen Z is actually facing.
They entered a workforce that was already closing its doors. Entry-level roles automated away before they arrived. Internships eliminated. The traditional pathway from early-career learning to organizational contribution — demolished before they could take the first step.
And now agentic AI arrives. And the story the headlines are telling, the story the org chart changes are confirming, is that the roles they were hoping to grow into may not be permanent either.
That fear is not irrational. It is a reasonable reading of what the workplace has already shown them. And they deserve honesty — not reassurance that papers over what is genuinely uncertain.
A recent survey of 2,400 workers found that 44% of Gen Z employees admit to actively working against their organization's AI rollout. Not because they don't understand the technology but because, as Fortune named it precisely, it is less a sign of technophobia than of workers who feel unprotected and are acting accordingly.
Here is what's also true: the technology that threatened them created something it didn't intend to. A new role. A new on-ramp. The Agent Manager role didn't exist before agents arrived. It wasn't automated away because it can't be yet — it requires human judgment, workflow literacy, and comfort with evolving technology that Gen Z carries as a baseline.
No coding required. No decade of domain expertise required. Already appearing on job boards. Early data suggests salaries averaging around $103,000.
That's a real opportunity. Right now.
But it needs to be named honestly — including its limitations. The Agent Manager role is transitional. Managing agents already exist in multi-agent architectures, and the autonomous management layer is coming faster than most expect. The value of the Agent Manager role for Gen Z isn't primarily the title. It's what the role builds: the deep, contextual, organizationally specific knowledge of how a particular business actually works. Where things breaks, what good looks like, and how decisions really get made. That knowledge is the foundation for more durable roles as the technology matures. And it's knowledge that takes time to develop and that AI has not yet mastered at the level of organizational nuance it requires.
The Agent Manager role, approached with intention, is the development ground. Not the destination.
But it only becomes real if organizations do three things: formalize the role with genuine authority and a real career path, rebuild the on-ramps that were eliminated, and invite Gen Z into the conversation before the deployment, not after the resistance has already formed.
The generation most naturally equipped to manage agents is currently working against adoption in nearly half of all organizations. That's not their failure. That's an organizational design problem with a measurable cost and a clear solution.
Include them. Give them the role. Mean it. And be honest with them about where it leads.
What Readiness Actually Looks Like
This is the question the series promised to answer. Here it is — specifically, honestly, and without false comfort.
Structural readiness means choosing your org shape deliberately.
Don't let the technology make the decision for you by default. The organizations succeeding with agentic AI are not just redesigning workflows. They are redesigning the human infrastructure around those workflows. That is the harder work. And it is the work that determines whether agents amplify organizational strength or amplify organizational fracture.
Role readiness means naming the four roles explicitly.
Give them job descriptions. Give them authority. Give them career paths. Stop asking people to do Agent Policy Director work without the title, the mandate, or the organizational standing to make policy decisions stick. Stop asking people to do Agent Manager work without the training, the tools, or the support to do it well. Stop conflating the Agent Role Architect and Agent Builder into a single onboarding process that underserves both. The roles exist whether you name them or not. Naming them is how you make them work.
Human readiness means psychological safety before training.
Communication before deployment. Inclusion before adoption. The organizations navigating this well opened the conversation before the technology arrived, not after resistance had already formed. They listened to what their people were worried about and used that intelligence to design a transition that held. That is not soft people management. That is operational intelligence.
Pipeline readiness means rebuilding what was demolished.
The entry-level roles that were eliminated weren't just cost savings. They were the development path for future Agent Role Architects — the domain experts who know workflows deeply enough to design agents that actually work. Rebuild those pathways. The organizations that do will have a capability that cannot be replicated through hiring or training alone.
Governance readiness means treating agents like workers.
Agents need role definitions before they're built. They need monitoring after they're deployed. They need accountability frameworks that assign human responsibility for their outputs. They drift. They need to be caught when they do. Google built the inbox for this at Cloud Next this week. Your job is to build the human governance around it — the policies, the decision rights, the escalation paths — before you need them.
Because the organizations that build governance after something goes wrong are building it in the worst possible conditions.
What This Series Was Really About
I want to be honest with you about something before we end.
Writing this series was its own experience of Ambient Effort™.
The deeper I went into the research, the more I understood why this moment feels so overwhelming — not just for the organizations and professionals navigating it, but for anyone paying close enough attention to see it clearly. The technology is extraordinary. The pace is relentless. The human cost is real. And the honest answer to "how do we manage thousands of agents?" is still, for most organizations: we don't know yet.
That is not a comfortable place to end a series.
But I think it's the right place. Because the alternative — the clean framework, the confident roadmap, the reassurance that it's all going to be fine — would be a disservice to the people reading this. And you deserve better than that.
So here is what I actually believe, after all of it:
The opportunities are real.
The Agent Manager role exists today. It pays well. It doesn't require coding. It is an entry point that didn't exist before agents arrived. The Agent Policy Director role is genuinely important work that almost no organization is doing well — which means the people who do it well will be extraordinarily valuable. The Agent Designer role rewards exactly the kind of deep domain knowledge that takes time to build and that AI hasn't replicated at the level of organizational nuance required. These are real opportunities, right now, for real people.
The unknowns are significant.
Managing agents are already replacing human Agent Managers in multi-agent architectures. AI is developing domain knowledge capabilities faster than most predicted. The roles we're describing today may look different in two years — or may not exist in the same form at all. Anyone who tells you they know exactly how this unfolds is not paying close enough attention. The honest position is: the trajectory is clear, the timeline is not, and the pace is faster than anyone expected.
The reality of this moment is genuinely hard.
We are living and working inside the most disorienting technological transition in modern history. It is moving faster than our organizations, our training systems, our governance frameworks, and our human nervous systems were designed to absorb. The exhaustion is real. The anxiety is earned. The feeling of ground moving beneath your feet is not weakness — it is an accurate reading of what is actually happening.
And the responsibility extends beyond any single organization.
The scale of this transition — the pace, the breadth, the depth of its impact on human work and human lives — is too large for individual organizations to govern responsibly on their own. This moment requires thoughtful public policy: regulatory frameworks that protect labor, establish accountability for autonomous systems, and represent the public interest at a speed that matches the technology's deployment. The early legislative efforts — the EU AI Act, state-level regulations beginning to take effect — are a start. They are not sufficient.
It also requires collective human voice. When profitability is represented by agents — when the economic incentive runs toward deploying more agents and employing fewer humans — someone has to speak for the human interest. The market will not do it. Competitive pressure makes it nearly impossible for individual organizations to do it unilaterally. The labor movement understood this a century ago, when industrialization created a similar imbalance between capital and human labor. The writers and actors who organized against AI use of their work understood it recently. The principle is the same: when individual humans cannot adequately protect their interests alone, collective voice becomes necessary.
This is not a political statement. It is a historical observation. Every major technological transition that significantly disrupted human labor eventually generated both public policy responses and collective advocacy. We are early in that process. But it is coming. And the organizations, leaders, and professionals who understand this — who participate in shaping both the policy and the advocacy rather than simply responding to them — will be better positioned than those who don't.
And it is navigable.
Not because someone has all the answers. But because the people who name what's real — who resist both the panic and the false comfort, who stay curious and honest and human in the middle of it — those people are the ones who find the path. They always have been.
In Article 1, I asked you a question: How are you doing?
I'm still asking.
Not your AI adoption score. Not your agent deployment timeline. Not your governance framework maturity.
You. The person who has been trying to lead through something nobody has a map for. The professional carrying the weight of a transition that is moving faster than anyone prepared you for. The human being doing their best inside a moment that is genuinely, legitimately overwhelming.
How are you doing?
That question is still the most important one. And the answer to it — honest, specific, and human — is where the work of navigating this actually begins.
Sundar Pichai asked from the stage in Las Vegas this week: "How do we manage thousands of agents?"
The answer, as it has always been, starts with how we manage ourselves — our organizations, our people, our anxiety, our capacity for honest conversation about what we know, what we don't, and what we're trying to build.
Readiness is a human decision.
It always has been. It always will be.
Even as the agents multiply.
The Full Conversation
This is the final article in Ambient Effort™: The Agentic Series. The complete series:
The Ambient Effort™ Series
Understanding the hidden energy required to remain effective in workplaces shaped by constant transition.
The conversation underneath this series has been building longer than this moment. My earlier work explores the human and organizational dimensions of AI transition that the technology conversation consistently skips:
And on the generation that was ready before we asked:
The Gen Z Series
Let's Talk
If this series landed — if you recognized your organization, your team, or yourself in any of these pages — I want to offer you something.
Not a framework. Not a checklist. A conversation.
I offer a free 30-minute consultation. A real conversation about where you are in this transition — as a leader, a professional, or an organization — and what readiness actually looks like for you specifically.
The technology is not waiting.
But you cannot manage thousands of agents without the humans who know how. And building those humans — finding them, naming their roles, giving them authority, and supporting them through the transition — is the work that determines everything else.
This article draws on reporting and research from Google Cloud Next 2026 (Las Vegas, April 22–24), HumanX 2026 (San Francisco, April 6–9), Harvard Business Review, PwC's 2026 AI Business Predictions, Deloitte's 2026 Global Human Capital Trends and State of AI Report, McKinsey's State of Organizations 2026, MIT Sloan Management Review, Boston Consulting Group, the World Economic Forum, Fortune, CNBC, Salesforce, Microsoft, Gartner, and Sequoia Capital's "From Hierarchy to Intelligence."
