.png&w=2048&q=75)
If you're skeptical of another "make AI your GTM team" post, good. You should be. A lot of what gets promised here doesn't work beyond a couple of use cases. We're not here to talk you out of the idea.
We want to walk through what it takes to build this properly, resources needed for it; and what's still missing even then.
What it takes to build this
We're seeing this happening in the market among nimble GTM engineering teams that are DIY. Imagine a high-friction complex B2B category with a long sales cycle and a large buying committee - a situation that requires a tailored approach. Teams are coming to the conclusion that outbound email alone is close to dead, response rates are near zero, and most "signal and enrichment" tooling on the market is a solved, commoditized problem, just table stakes.
So some teams are building their own systems instead.
Here are two examples of this from recent conversations:
A Head of Demand Gen with a similar complex sales situation has quite a sound, disciplined setup. Their lead development runs through a Clay-equivalent tool that enriches every contact and factors accounts by intent. Sales Navigator fills gaps the enrichment tool misses, Apollo runs alongside it, and Claude, wired up with skills and connectors, sits at the center of it, generating the actual outreach. Their team is trained on this workflow.
The second example is another company where their GTM Engineer has built a meta-prompt running to dozens of pages, fed by intent data from multiple sources (Bombora, G2 to name a few), and first-party CRM data. It’s not vaporware - Claude generates personalized, persona-specific 6 step email sequences, which are then pushed out through execution tooling across email, LinkedIn, and voice, largely without day-to-day babysitting.
If you're the kind of high-maturity, DIY-minded team that can pull this off, genuinely, kudos. That's a hard to build system, and it's a real answer to the problem of generic sequences not working anymore. It creates contextual, multi-channel content and is not stitched together over one vibe-coding weekend. If you have the engineering resources to build, monitor, and continuously feed a system like this, it's a legitimate approach to the execution problem. But this approach is worth thinking through before going all in, because it's easy to mistake execution sophistication as the whole solution.
What it doesn’t solve
Looking closely at what a system like this actually does, it's answering two questions:
a) what should we say, and
b) how should we say it across channels.
Those are valid, hard problems, and solving them well is impressive.
But, what about the question of who deserves your attention in the first place, or whether now is actually the right time to approach them. A meta-prompt like this can make an email sequence strong, but it is assuming that the account is right for your business. It has to come in as an already decided, correct input.
Second, it treats every signal as relevant, while in reality, the majority of what is going in is noise, worth discarding. Sooner or later, reps learn to ignore signals because they are a waste of time. And chances are, the company is losing credibility in the buyers’ eyes, because the message is at best, a minor nuisance, and at worst, offensive because their offering has no grounding in your real challenges.
There's another risk in this. If the intent signals feeding your personalization engine - are also influencing which accounts get pursued through a scoring mechanism, you've created the problem we've written about before: signals doing a ICP fit scoring job. An account that seems active need not be the same as an account that's a good fit for you - it doesn't matter how good the six-step sequence is.
Not to mention intent signals, in principle, are unreliable, requiring ruthless filtering, because they infer interest and are about guessing or probabilities. Context signals, on the other hand, reflect something that has tangibly changed.
You don't need another execution tool
However sophisticated the build, a meta-prompt wired into enrichment sources and execution tools is still, structurally, an execution system. Don't mistake me - it's a good set up, built and maintained by people with real engineering chops. It answers how, but does it answer who? Is it clear on the when? No amount of personalization sophistication fixes those fundamentals.
This is not a case against Claude, or against using it inside your GTM workflow. It's a case against expecting a team to stitch together the strategy, tools, agents and execution in a short window. It's a case against carrying the orchestration alone, whatever the surface.
We arrived at this take after trying to solve the same problem ourselves. Early on, we were thinking in terms of email, voice, and LinkedIn orchestration, much like the system described above. What changed our minds was looking back to our time as GTM operators - thinking about where the real leverage is: which is in clearly knowing which accounts were worth pursuing, why, and when. Especially where sales cycles are over 6 months long.
Sending more messages is commoditized - execution is getting better, and is solvable by enough engineering effort, but the decision underneath isn’t getting easier.
If you're weighing whether to build something like this yourself: it's a legitimate investment if execution quality is really your bottleneck. But check that assumption first.
The harder and less-discussed, timeless problem is knowing which accounts look tempting, but should be deliberately bypassed, and which are the ones to spend energy on.
Worth knowing: that decision layer doesn't have to mean a steep learning curve associated with new tools. It's available directly inside Claude and ChatGPT now - so you needn't worry ‘will my sales team adopt a new tool’. The who-and-when call is accessible to them where they want, on-the-go.
There's a second cost too, which is just coming to surface. The human who has to run it, monitor it, catch what it gets wrong, decide what to trust and what to double-check. Instead of doing the work like they used to, they’re now working hard to manage tools. That arguably causes real cognitive fatigue, and leads to a feeling of ‘haziness’. More on that later.
So if your latest AI initiative can execute almost any GTM motion you design, the scarce resource is no longer execution. It's knowing who was the right fit in the first place, what to execute and then making it happen using the right combination of humans, agents, and tools.
FAQs
Can AI replace an entire GTM team?
AI can automate a significant amount of GTM execution, including research, enrichment, personalization and outreach. But it does not eliminate the need for decisions about ICP fit, timing, account prioritization and buying context.
What is the biggest limitation of AI-powered GTM?
Most AI GTM systems are very good at executing a motion once an account has been selected. The harder problem is deciding which accounts deserve attention in the first place.
What is the difference between ICP fit and buying intent?
ICP fit tells you whether an account resembles your ideal customer. Buying intent or signals tell you whether something is happening that may make the account worth pursuing now. Neither should substitute for the other.
Do GTM teams need another AI execution tool?
Not necessarily. If your existing stack can already research, enrich, personalize and execute outreach, the bigger opportunity may be improving the decision layer underneath it: who to target, why, and when.
