See How a New Approach Works for Your Specific Use Case — Explore the Test Drive >
In my last article, What I've Learned About Business Transformation After Watching Companies Recreate the Problems They Were Trying to Solve, I explored something that had been bothering me after watching several organizations make decisions that seemed contradictory from the outside.
They understood what wasn't working. They could explain why it wasn't working. In some cases, they even had remarkable clarity around what needed to change.
And then they made decisions that recreated some version of the problem they were trying to solve.
I came away from that experience thinking less about the decisions themselves and more about the assumptions underneath them. The assumptions weren't necessarily irrational. Many of them had been shaped by experience, past success, resource constraints, or operating models that had worked before.
The problem was what happened when those assumptions stopped being questioned. That observation has stayed with me because, at the same time, we've been introducing significantly more AI into the GTM systems we operate. AI-assisted dialing. AI-generated content. AI agents supporting LinkedIn outreach. Automation helping identify signals, prioritize activity, and determine where human attention should go next.
For a long time, I thought about most of these capabilities primarily through the lens of capacity. How much more could one person accomplish? How much repetitive work could we eliminate? How much faster could a team execute? Those are still important questions. And after actually putting these capabilities to work, I've seen enough to believe the potential is enormous. But I've also started asking a different question.
What exactly are we giving all of that capacity to?
One of the things I find interesting about the current AI conversation is how often the technology itself becomes the center of the discussion. Can an AI agent do outbound? Can AI write the content? Can it research an account? Can it manage a conversation? Can it replace an SDR?
Increasingly, the answer to many of those questions is yes. But I'm not sure capability is the most important question anymore. Organizations were making assumptions about their markets, buyers, messaging, qualification criteria, sales processes, and channels long before generative AI showed up. They were also getting some of those assumptions wrong. What AI changes is not necessarily the assumption.
It changes how much can happen because of it.
A questionable list that once took a team months to work can now generate enormous amounts of activity in a fraction of the time. Messaging that has never really been validated can suddenly be personalized across thousands of accounts. A process that relies on weak qualification logic can now execute that logic consistently without getting tired, distracted, or overwhelmed.
That is incredible leverage when the underlying assumption is right. It is still leverage when the assumption is wrong.
Over the past several months, we've been deploying more AI across a live GTM program. Not as a demonstration. Not as a hypothetical use case. Inside an actual demand system where the technology has to interact with real data, real buyers, real messaging, real salespeople, and ultimately a real client's reputation.
Some of what we've learned has reinforced exactly what I hoped AI could do. Our AI-assisted dialing capabilities are a good example. The technology dramatically expands the amount of activity a human can manage. Work that would have required several people can increasingly be supported by one. But the important part, at least in the way we're using it, is that the technology isn't being asked to replace the human judgment that matters most.
When the conversation happens, a person is still there. The machine creates capacity around the conversation. The human still has to understand it.
I've started thinking about that distinction a lot because our experience with AI-generated content has been very different. We were using AI to help create content inside the same demand system. The reasoning was straightforward. Content requirements increase as the system expands, and producing everything manually eventually creates a capacity constraint.
So we introduced AI into the process. The content was created. It went through review. It was approved. Some of it eventually reached the market.
And then we ran into a problem.
The content wasn't necessarily inaccurate. It wasn't grammatically broken. In isolation, someone might even have called it good. It just didn't sound like the client.
Certain words were wrong. Certain phrases felt artificial. The writing technically communicated the idea while somehow losing the person and organization it was supposed to represent. That forced us to stop and reconsider what we were actually trying to scale.
This was probably the part of the experience that changed my thinking the most. It would have been easy to blame the AI. We didn't. The technology had largely done what we asked it to do. That was the problem.
We had created a process capable of producing acceptable content faster than we had created a system capable of defining what acceptable actually meant. So we pulled AI back. We returned to human-written content and started documenting the things humans close to the business recognized almost instinctively. Vocabulary. Phrasing. Context. Industry language. The subtle differences between something that technically says the right thing and something the organization would actually say.
Then we started building those lessons back into the process. We're now comparing completely human-written content with versions where AI plays a lighter role. The goal isn't to prove that humans are better than AI or that AI is better than humans. We're trying to understand what the right amount of AI involvement actually is before we scale it further. That's especially important because the next phase of the program is designed to operate at greater scale.
That experience has changed the question for me.
It is no longer simply, "Can AI do this?" It is, "Do we understand this well enough to let AI do more of it?"
I think this is where some of the current AI SDR conversation gets oversimplified. A job is not one thing.
An SDR researches. Prioritizes. Dials. Writes. Follows up. Interprets. Qualifies. Builds familiarity. Recognizes context. Makes judgment calls. Decides when something deserves escalation and when it should be left alone. AI can increasingly participate in almost all of those activities.
That doesn't mean it should participate in all of them in the same way.
We've seen tremendous value in using technology to remove capacity constraints around dialing while keeping a human in the conversation. Content required a different boundary. And now, as we introduce AI agents into LinkedIn outreach, we're encountering another one.
An AI agent helping identify a prospect or surface a signal is assisting someone's work. An AI agent communicating under someone's identity is representing them. Those aren't the same level of responsibility.
The closer automation gets to a moment involving context, trust, creativity, ambiguity, or reputation, the more consequential the underlying rules become. That doesn't mean a human needs to perform every action. It means "human in the loop" is probably too broad a concept.
The more useful question may be where the human belongs in the loop.
Why guess if a solution will work? Our hands-on Test Drive is a focused pilot engagement designed to pressure-test your entire GTM motion turning a small investment into definitive proof.
Validate your strategy with real data and scenarios, receive a customized pilot plan and ROI assessment, and eliminate uncertainty before a full-scale rollout.
Recently, we encountered an organization that wanted to create movement across several markets. The urgency made sense. But the environment underneath that urgency was complicated. GTM channels were siloed. The data needed work. A new CRM was barely out of implementation. A previous paid-media relationship had recently ended. And there were roughly 80,000 contacts without a clear understanding of how all of them had entered the system.
The instinct was to move quickly. Start dialing. Start running ads. Generate activity. Find traction. None of those tactics are inherently wrong. But together they raised a question that has become increasingly important to me. What have we actually validated?
Now add AI to that environment. Suddenly the organization can research more accounts, create more content, personalize more messages, initiate more conversations, and work through more contacts than its previous operating model could have supported.
The capacity problem begins disappearing. The uncertainty doesn't. And that is where I think AI introduces an interesting paradox. The easier execution becomes, the easier it may become to skip the learning that should have happened before execution.
We can move so quickly that movement itself starts feeling like validation.
I've seen that pattern before. In an earlier installment of this series, I wrote about how activity can create confidence faster than evidence. The campaign is running. Engagement is increasing. Meetings are happening. Pipeline appears to be moving. The activity is real. The conclusion we draw from it may not be...
AI doesn't eliminate that problem. It can multiply it.
A GTM system exists to execute, but it also exists to learn. Which buyers respond? Which problems create urgency? Which messages change the conversation? Which signals actually predict movement? Which channels work for which audiences? Which assumptions were right? Which ones need to change?
If AI helps an organization execute ten times faster without helping it answer those questions any faster, I'm not sure we've created the advantage we think we have. We may have simply created more activity to interpret. That is why some of the AI work we're doing now has started to look less like automation and more like experimentation.
Human-written content against AI-assisted content. Rules refined from actual client feedback. Signals captured and fed back into the system. Automation expanded where the evidence supports it and constrained where judgment still matters. We're not trying to decide whether AI works. We're trying to learn where it works, how it works, and what needs to be true before we trust it with more.
That feels like a much more useful question.
For years, one of the biggest constraints inside GTM organizations was capacity. There were only so many people. Only so many hours. Only so many calls a BDR could make. Only so many accounts a seller could research. Only so much content a marketing team could create. AI is beginning to remove some of those constraints. That is the opportunity.
But capacity wasn't the only thing limiting execution. Sometimes friction forced organizations to make choices. It forced prioritization. It exposed weak processes. It created time for feedback. It limited how far an untested assumption could travel before someone had to reconsider it.
As that friction disappears, I think the quality of the system underneath the execution becomes more important, not less. The organizations that benefit most from AI may not be the organizations that automate the most. They may be the organizations that understand their systems well enough to know where automation creates leverage, where it creates risk, and where they still need to learn before they scale.
I'm starting to believe that is the real AI readiness question.
Not how much AI an organization can deploy. How much of its operating model it actually understands well enough to accelerate.
In the last installment, I wrote about organizations recreating problems they already understood because the assumptions underneath their decisions remained largely unchallenged. AI adds a new dimension to that problem. It gives those assumptions leverage.
That isn't an argument against AI. Our own experience has made me more convinced of its potential, not less. I've seen what happens when technology removes work that never required human judgment in the first place. I've seen the capacity it can create and what becomes possible when people are able to spend more of their time where judgment, context, creativity, and trust actually matter.
But the more we've experimented with where AI belongs in the system, the more I've started thinking about the other side of that equation. Where do the humans belong? For years, when organizations needed more capacity, the answer was often straightforward. Add people.
More sellers to work the accounts. More marketers to create the content. More operations resources to manage the systems. More managers to coordinate the people doing the work. Sometimes that was exactly the right answer. But I'm starting to wonder how often we added human capacity because the underlying system couldn't scale.
AI is forcing us to make decisions about work that we probably should have been making all along. Which work requires human judgment? Which work should be handled by the system? Which work shouldn't exist anymore?
Maybe the opportunity isn't simply figuring out how much work AI can take away from people. Maybe it's finally getting clearer about the work we actually need people to do. Because if we're going to question which problems deserve automation, I think we also have to question which problems we've been solving with headcount.
And I'm beginning to suspect there are more of those than we think.
