You Can't Manage an Agent’s Job if You Couldn’t Do it Yourself
I run marketing at a startup, and AI agents are part of my daily workflow. They sweep research, draft content, pull analytics, and flag competitive moves while I'm doing something else. The leverage is real, and anyone who tells you otherwise hasn't tried it seriously.
I recently read a post that raised a question I can't stop thinking about: should a startup's marketing lead stay hands-on, or should they manage a team of AI agents?
It's a fair question.
But I think the question is poorly framed. It assumes "hands-on" and "managing agents" are two different jobs, and that “managing agents” is a promotion away from being “hands-on”. They're the same job. You can't manage agents whose job you couldn’t do yourself.
The Irony
In 1983, a cognitive psychologist named Lisanne Bainbridge published a short paper called "Ironies of Automation." She was studying industrial process control, the same world of plants, operators, and control rooms we spend our days protecting at BlastWave. Her finding sounded counterintuitive: the more you automate a system, the more skilled the human supervising it needs to be, not less. As she put it, "The more advanced a control system is, so the more crucial may be the contribution of the human operator."
The logic is simple: Automation handles the routine, and the human steps in only when something goes wrong, at the exact moments that require the deepest understanding of how the system works. This means that if the human operator never does the work by hand again, the skills they once had may erode to the point of uselessness, right when they're needed most.
Plant operators have learned this lesson the hard way over decades. Those of us in Marketing will learn it in about eighteen months.
What "Managing Agents" Really Looks Like
When people imagine a marketing leader who "just manages agents," I think they picture someone reviewing dashboards and approving outputs. Click, approve, ship. It looks very different in practice.
You write the brief. Average-in produces average-out. An agent is only as good as its instructions. Telling an agent to "write a blog post about OT security" gets you something that smells like a blog post. Telling the agent who the reader is, what they already believe, which claims can and can't be made, what vocabulary a water utility operator uses versus a CISO, and what the reader should do next? That requires knowing the audience and the craft. Nobody can write that brief without understanding the content output.
You judge whether the output is good or just plausible. The big one! AI output is fluent by default. It reads well, it's confident, and sometimes it's wrong in ways that only an expert would catch; a statistic that's slightly off from the source, a claim that drifts past what the product actually does, or a positioning statement that sounds great but quietly concedes ground to a competitor. Sounding fluent and confident isn't the same as being accurate, and only someone who knows the difference can spot the gap instantly.
You know when the agent is confidently wrong. Agents don't raise their hand when they're out of their depth. They just keep going. The operator who's done the work by hand recognizes when something’s fishy. The operator who hasn't, ships it.
I've built my own workflows around this reality. We run an agent that sweeps the OT security research landscape twice a month, pulling new statistics we might cite in our content. It's enormously useful. But it's governed by one hard rule: every figure gets verified against the source text, and nothing enters our stat library without my approval. Same with our monthly competitive review. The agent does the sweep and rates what changed, and then a human with context decides what it means. There have been many times when I asked for multisource attribution, and the agent came back with “You are right, my bad”.
Each of those gates exists because I know enough to see where the agent can go wrong. That knowledge didn't come from managing agents; it came from doing the work.
The Tradecraft Lesson
I spent years at the NSA, and I saw this long before anyone talked about large language models. The tools available to analysts were extraordinary. The analysts who got the most out of them weren't those who simply ran the tools, but those who understood the underlying tradecraft well enough to know what the tool was missing, what it was overweighting, and when its output didn't pass the scratch-and-sniff test.
Good tools don’t replace sound judgment but amplify it. An expert with a powerful tool becomes dramatically more effective. A novice with the same tool just produces mistakes faster with a more confident swagger.
I believe this is exactly where marketing is heading. AI agents are the most powerful amplifier our profession has ever had. The crucial question is: what are they amplifying?
If you run security for a plant, you're next. Someone is going to ask you to approve autonomous agents with reach into your network, and the same question applies. Whoever signs off on what an agent can touch has to know that system well enough to run it by hand. Bainbridge wrote her paper about your control room first.
What This Means for Startups
If you're a founder hiring your first marketing leader, here's my unsolicited ten cents’ worth of advice: Don't hire a manager of agents – hire a player-coach who uses agents for leverage.
At a startup, the marketing leader is often the entire marketing team, with a budget that wouldn't cover a single agency retainer. AI has made that person capable of output that used to take five or six people. That's a compelling story, right? But that kind of leverage only works if the person has the ability and experience to do every job the agents do. They need to write the copy, read the analytics, build the campaign, and understand the buyer, because the lion’s share of their job will be catching the moments when agents trip over their own shoelaces.
A hands-off marketing leader in an AI-driven team isn't a strategist but a single point of failure with a fancy title.
Higher Floor, Higher Bar
I'll close with a theme I keep coming back to in OT security, just in a different building.
AI raises the stakes of having the fundamentals in place. In OT, AI didn't create new vulnerabilities; it just made old ones dramatically easier to exploit. In marketing, AI didn't eliminate the need for craft; it made the consequences of lacking craft show up faster and at greater scale.
AI raises the floor for everyone. Anyone can now produce competent-looking content. But it also raises the bar for whoever is accountable for the output, because when everything looks competent, the only differentiator left is judgment.
So should a startup marketing leader be hands-on or manage agents? Yes. Both. The second is impossible without the first.
– Cam Cullen, CMO, BlastWave
Frequently Asked Questions (FAQs)
1. Can AI agents replace human marketing teams?
AI agents can automate research, content creation, analytics, and competitive intelligence, allowing small marketing teams to accomplish significantly more. However, they cannot replace the expertise, strategic judgment, and accountability of experienced marketers. Effective AI-driven marketing requires human oversight to ensure accuracy, relevance, and quality.
2. Why do marketing leaders need hands-on experience to manage AI agents?
Marketing leaders must understand the tasks they delegate to AI agents so they can write effective instructions, evaluate results, identify inaccuracies, and correct mistakes. Without hands-on expertise, leaders risk approving convincing but misleading content, unsupported claims, and ineffective marketing strategies.
3. What is the irony of automation, and how does it apply to AI agents?
The irony of automation, described by cognitive psychologist Lisanne Bainbridge in 1983, is that increased automation can make human expertise more important rather than less. Automated systems handle routine tasks, leaving humans responsible for complex situations requiring deeper knowledge. This applies to AI agents in marketing, industrial operations, and cybersecurity.
4. How can startups use AI agents effectively in marketing?
Startups can use AI agents to scale marketing activities without proportionally increasing headcount. Experienced marketing leaders can delegate repetitive research, content drafting, analytics, and competitive monitoring while retaining responsibility for strategic decisions, fact-checking, brand positioning, and final approvals. The ideal startup marketing leader acts as a player-coach rather than a hands-off supervisor.
5. What are the biggest risks of relying on AI agents without human oversight?
AI agents can produce inaccurate statistics, unsupported product claims, misleading competitive comparisons, and content that appears credible despite factual errors. Without knowledgeable human review, these mistakes can spread rapidly across marketing campaigns. In operational technology (OT) environments, autonomous agents with inappropriate network access can introduce additional security and operational risks.
6. How does AI agent management relate to OT cybersecurity?
Both AI agent management and OT cybersecurity require knowledgeable human oversight, clearly defined responsibilities, and appropriate access controls. Just as marketing leaders must understand the work their agents perform, OT security professionals must understand the systems autonomous agents can access and the consequences of their actions. Human expertise remains essential when evaluating and approving autonomous operations.
7. Should startup marketing leaders remain hands-on or focus on managing AI agents?
They should do both. Effective marketing leaders combine practical skills in copywriting, analytics, campaign development, and customer understanding with the ability to direct and supervise AI agents. AI increases productivity, but the quality of the results ultimately depends on the experience and judgment of the person managing the technology.
How To: Manage AI Agents Effectively Without Losing Human Expertise
AI agents can dramatically improve marketing productivity, but successful implementation requires experienced human supervision. Follow these steps to integrate AI agents into startup marketing workflows while maintaining accuracy, accountability, and strategic control.
Step 1: Maintain Hands-On Knowledge of Marketing Tasks
Before delegating work to AI agents, ensure you understand how to perform the underlying tasks yourself. Maintain practical experience in content writing, campaign development, analytics, audience research, and competitive positioning.
Why it matters: You cannot reliably evaluate an AI agent's output without understanding what accurate, high-quality work should look like.
Step 2: Create Detailed, Context-Rich Instructions
Provide AI agents with specific instructions that identify the target audience, business objectives, industry terminology, product capabilities, prohibited claims, and desired outcomes.
For example, instead of asking an agent to write about OT cybersecurity, explain whether the audience consists of plant operators, CISOs, or industrial security engineers and identify the technical details relevant to that audience.
Why it matters: Clear instructions reduce generic outputs and help agents produce content aligned with business objectives.
Step 3: Verify AI-Generated Research Against Original Sources
Require every AI-generated statistic, technical assertion, and competitive claim to be verified against its original source before publication.
Establish an approval process in which research findings cannot enter an organization's reference library until a qualified human reviewer confirms their accuracy.
Why it matters: AI-generated content can sound authoritative even when its claims are inaccurate or unsupported.
Step 4: Establish Human Approval Gates
Identify activities that require human review before an AI agent's output can be used.
These should include publishing marketing content, approving competitive positioning, accepting research findings, and making consequential strategic recommendations.
Why it matters: Human approval checkpoints help prevent automated errors from becoming public-facing mistakes.
Step 5: Use AI Agents for Research and Monitoring
Delegate repetitive activities such as competitive intelligence gathering, industry research, analytics collection, and preliminary content drafting.
For example, schedule AI-assisted research reviews twice monthly and competitive analyses monthly, as described in BlastWave's marketing workflow.
Why it matters: Automation frees experienced professionals to focus on interpretation, decision-making, and strategic execution.
Step 6: Apply the Same Oversight Principles to OT Security
Before authorizing autonomous AI agents to interact with operational technology networks, ensure responsible personnel understand the systems involved and the potential operational consequences.
Define which systems an agent may access, establish appropriate access restrictions, and require qualified human authorization for consequential changes.
Why it matters: Autonomous operations in industrial environments demand appropriate safeguards, technical expertise, and accountability.
Step 7: Continuously Exercise Human Skills and Judgment
Regularly perform essential marketing tasks manually, review AI-generated results critically, and investigate unexpected findings.
Treat AI as a tool that amplifies professional expertise rather than a replacement for developing or maintaining that expertise.
Why it matters: The more organizations depend on automation, the more important it becomes to preserve the human skills needed to recognize and correct failures.
Key Takeaway
Effective AI agent management requires both automation and expertise. Startup marketing leaders and OT security professionals should use AI to expand their capabilities while maintaining hands-on knowledge, rigorous verification processes, and human accountability. The goal is not to remove humans from the workflow but to make skilled professionals more effective.
American Water’s cyber incident shows why water utilities need stronger OT isolation to stop IT breaches from threatening critical infrastructure operations and availability.
Explore the complete analysis of 23 OT attacks that defeated firewalls, VPNs, and air gaps.
.svg)
.svg)
