Applied AI in Your GTM Team
Ask a GTM team today whether they use AI and the answer is always yes. Press on what that means and it usually comes down to a few people with a chat window open, pasting in emails to rewrite and calls to summarize.
That’s real. It’s also the bottom rung of something much taller.
The teams pulling away right now aren’t the ones with access to better models. Everyone has the same models. They’re the ones who have climbed higher up the ladder of what AI can actually do inside a revenue org. Applied AI is the difference between a team that dabbles and a team where AI quietly runs parts of the system. It helps to be specific about what that ladder looks like, because where your team sits on it decides what you get out of it.
This builds on work by Kyle Norton, who broke down the four levels of an AI-native team and, at his own company, centralized applied AI inside one expert group instead of scattering it across the org. What follows is our own three-level cut of the same idea.
The three levels
Level one: chat. Someone opens ChatGPT or Claude and prompts it by hand. Rewrite this email. Summarize this call. Draft a follow-up. The output is only as good as whoever is typing, nothing is saved, and nothing compounds. Useful, but it lives and dies with individual effort.
Level two: custom assistants. The team builds reusable assistants - custom GPTs, projects, saved prompts loaded with the ICP, the positioning, real call notes. Now the knowledge is shared and the output is more consistent. This is where most teams that take AI seriously top out. It feels like a destination. It is actually the middle of three.
Level three: engineered systems. AI gets wired into the machine itself and runs without anyone prompting it - Clay tables that enrich and score, n8n flows that route, and, where it counts, real code with prompt engineering and evals that test output across thousands of cases instead of eyeballing one. At this level AI owns whole steps of the funnel - research, scoring, call review, first-draft outreach - and you can prove it is working because you are measuring it. The technical range is wide, from a no-code automation to a coded application, but it is all the same move: AI that runs as part of the system instead of a tool a person opens. This is where the payoff is, and almost no team reaches it by accident.
Why most teams stall
The jump from level two to level three is enormous, and it widens every quarter as the ceiling rises.
The reason teams get stuck is simple: each level up demands a skill the one below it did not. Climbing from chat to custom assistants is easy. Climbing from custom assistants to engineered systems is a different kind of work - it means thinking in systems and then building them, with code and evals where it counts. That is a real gap in skill, not a line in a budget. Handing every rep a chat window and hoping they figure out the rest does not move a team up the ladder. It gets you a building full of people stuck on rung one, each inventing their own slightly different prompt.
This is the trap inside “we gave everyone AI.” Access was never the constraint. Capability is. A decentralized free-for-all rarely climbs past custom assistants, because the top of this ladder is not a tool you switch on. It is a function you have to build.
What “applied” actually means
The highest-value AI in a GTM team is usually invisible to the customer.
The flashy version - autonomous agents blasting personalized cold email at scale - is the lowest-value use of all this, and the one most likely to embarrass you. The real gains are internal and unglamorous. Enriching records that no database sells. Scoring which accounts deserve a human’s time. Reviewing every call and surfacing the moments a manager should hear. Coaching at a scale no manager could match by hand.
None of it works on a weak foundation. AI is a pattern machine. Feed it broken data and disorganized process and you get broken output faster. So the foundation comes first - clean data, defined process - and the AI sits on top of it, aimed at a narrow job. Give a model a tight, well-defined task and it is reliable. Give it a blank canvas with too much room to roam and it starts making things up. Constraint is a feature, not a limitation.
Where this leaves you
The point of mapping the levels is not to rank teams. It is to make the ceiling visible.
Most revenue teams are operating a level or two below what is already possible, and the gap is not budget or tool access. It is that nobody has built the higher rungs yet. The teams winning with AI did not buy something the others could not. They turned applied AI into a function and climbed.
The tools will keep getting easier, and the rungs that need an engineer today will move within reach of a sharp operator tomorrow. But the ladder is not going away. The only question worth asking about your own team is which rung it is standing on, and what it would take to climb one more.