how ai-native is your GTM team?! (Six-Step Ladder)

I've spoken to hundreds of companies now about their GTM and just seeing what everyone's steps are and can I tell you, some companies are using AI in amazing ways, others it probably hurts them, and some just not at all.

I call this some version of a GTM engine, and in this piece I wanted to break down how to approach thinking about how AI-native your team actually is.

So, plainly, what is a GTM engine? It's one central machine that runs your marketing and sales. Agents and humans wired together, pointed at a number at each stage of the funnel. Reach at the top, booked calls in the middle, closed deals at the bottom. One system, one direction. Most companies have the pieces, a writer here, an outbound tool there, a CRM nobody fills in, just scattered and siloed and not talking to each other. The engine is what you get when you wire those pieces into one thing that can actually move.

Here's the part to hold onto the whole way through: the GTM engine is an augmentation of human capability. The best systems know what they know and what they don't. AI can't host your event, can't be the trusted face on an enterprise call.

So I want to go over a 6-step ladder I've been using to think about this. Keep in mind this is very oversimplified to keep this from getting super long, but I'll go into more detail at the end for anyone that's really interested.

so this is Bob right now. Series A, B2B, two plans. And on paper there are channels, founder-led outbound, content, events, landing pages, calls and closing. In practice, Bob is the engine. DMing warm intros on LinkedIn, hopping on calls, closing on charisma and a genuinely good product. Content is a post when Bob remembers to post it. The landing page is a one-pager someone threw up in an afternoon. The CRM is empty or doesn't exist.

And there's revenue, that's the bit people get confused by. They think because there's revenue there's a system. But ask Bob which channel actually drove the last three deals and it's a shrug. Nothing's recorded so nothing can be learned from.

the general play here is there's not really anything to build yet. Don't let revenue trick you into thinking you have an engine when you just have a founder and some luck. Those are different things. The one useful thing you can do at this stage is start writing down what actually happened on the deals you win, so when you do add tools there's something for them to plug into.

L1, assisted

people start using AI tools to speed up individual steps. Faster emails, faster posts, faster research, faster follow-ups. The work is still mostly human-run and siloed, but AI is making parts of the process quicker.

so the next step up the ladder for Bob is really just starting to use some AI tools. Bob starts writing cold emails with ChatGPT and gets them out in a tenth of the time. The content person drafts posts with it. Someone researches accounts with Clay and Perplexity. Follow-ups go out faster. Every person on the team is suddenly quicker.

But most companies stop here and feel AI-native, and it's a trap. It's the same scattered, warm-intro-led motion.

And honestly, at this stage I sometimes see people over-automating parts of the process. AI SDRs burning through leads and making people hate your brand. It's the same on most channels, there is a world where you use AI wrong and you completely kill people's perception of your brand across Reddit.

I've seen this happen to one brand where it's a genuinely good tool I use and when I show it to my friends they say wow that's actually really good, but then immediately say they hated their Reddit marketing so much they refused to try it. That's a real problem.

the general play here is figure out what tools are actually good for, which channels and content is already working, and scale up from there. This stage requires a lot of experimentation. When you find something that works, stick with it and systemise it.

AI and GTM tools start getting connected into actual workflows. Some channels are now measurable. You might have one channel working really well, outbound, content, self-serve signups, with source tracking, CRM logging, and basic reporting around it. But this is still channel by channel, not one connected engine.

so for Bob, they stop spreading themselves thin and make one channel real. They notice events and outbound are working really well, connect it all to their CRM, and a simple dashboard shows where each person came from. They can actually start measuring what's working.

And they've probably found 2-3 tools that are really clicking.

Here they hit the first real wall. To climb out of L2 they have to build a context graph. It was fine when you were experimenting to upload your website to these tools manually, but to get past one-channel-at-a-time, the data has to go AI-ready. Every account, what's happened with them, what each channel did and what came of it. Without it there's just a lot of guessing and re-setting up, which is not ideal.

the general play is to figure out one channel completely before you touch the next. Source tracking, CRM logging, a dashboard you actually open. Trying to measure everything at once means you measure nothing. But once that's done, get your data ready.

every important channel or process now has an owner, either a human or an agent. Outbound has an owner. Content has an owner. Events have an owner. Sales follow-up has an owner. The data from those processes is captured somewhere so the company can start running repeatable processes.

so with the graph underneath them, Bob actually puts a name on every channel. Each one running a stack of specialist agents underneath, pointed at a single number.

And this is where the augmentation thing actually becomes real instead of just something I say. Take events. An agent can't host the thing, can't work the room, can't be the person someone remembers meeting. So the owner lets it do everything around the room instead, building the invite list, personalising the outreach, drafting the reminders, scoring who showed up, cutting the talk into content afterwards, running the follow-up the next morning. Bob does the one human hour. The agents do the forty hours around it.

I've also seen the failure mode here. An owner who trusts the agent stack too much and stops actually reviewing what it's doing. Content going out slightly off-brand. Outbound that's technically on target but reads like a robot wrote it, because one did. And nobody caught it because the owner assumed owned meant automatic. Owned still means owned. You're accountable for the number, not just for turning the agents on.

Then the next wall shows up. Six owned channels is six good little engines and no cockpit to see across them. To climb out of L3 they need a command centre, one place where every channel's state, its metrics, its agents, are visible and steerable at once.

the general play: give every channel that actually matters a single owner accountable for a single number. Point them at a stage metric, reach, leads, booked calls, not revenue, that sits too far downstream to steer by. And name your human edges on purpose. Protect those deliberately and wrap agents around everything else.

once everything is owned, the pieces start triggering each other. Sales call transcripts flow into Slack. Objections from calls become marketing content. Closed-won data changes targeting. Event attendance triggers follow-up. The GTM engine is no longer a set of owned workflows, it's a connected system with feedback loops running across channels.

so now for Bob the pieces actually start triggering each other. A recurring objection on enterprise calls turns into a landing page and a content angle inside a week. Transcripts flow straight into Slack. Closed-won data quietly reshapes who outbound targets next. Event attendance triggers the follow-up on its own.

This is the level people picture when they say GTM engine. Honestly, most people claiming it are still sitting at L3 with good marketing.

The last wall is the hardest. Short loops are easy, objection in, content out, same week. To climb out of L4 they have to actually master feedback loops over long horizons. The loops that only pay off over weeks or months, where the system runs a long task, you see how it landed way downstream, and it adjusts, and you trust it over that horizon without babysitting every step.

the general play: don't connect what isn't owned first. Wiring chaos to chaos just gives you faster chaos. This only works once L3 is genuinely solid. Start with the short obvious loops. Prove they work. Then trust the longer ones. And keep a human steering it weekly, connected does not mean unattended, it never should.

this is the level I honestly haven't seen built end-to-end yet.

Bob isn't here yet. Nobody I've seen is fully here. But it's worth saying what it looks like because it's where the whole climb has been pointing. The system sees what needs to happen, hands the work to humans or agents, drafts the assets, assigns the tasks, and routes Bob into only the moments that actually need judgment, taste, trust, or approval. The three enterprise relationships that need a real call. The brand decision that needs a human. The job stops being decide and execute every step and becomes steer the system, review the work, step in where it actually matters.

And the human never disappears here, even at the top. They just get concentrated. Pulled out of the busywork and dropped into the few places only a person can actually stand.

the general play: don't fool yourself about it. Spend as much effort deciding what gets routed to a human as what gets handed to an agent. And be a little suspicious of anyone telling you they're already here. Most of the time they're at L3 with good marketing.

that's the climb, and Bob is the clean version. Most real companies are sitting where Bob started, L1, maybe one channel limping toward L2, a CRM nobody fills in, stuck below the context graph without even a real spine under them. That's not a failure. It's just the rung you happen to be on.

the move is never add more agents. It's find the wall above you and build the one thing that gets you over it, the spine, then the graph, then the cockpit, then the long loops. Because this was never really about how much AI you use, it's about how wired the engine is, what one number it can be pointed at, and how well it knows its own edges, what to automate and what to leave to a human.

I see the next few years being a race through a maze of these tools full of sharp edges. there will be companies that come out covered in scars and blood. they will look behind them and see the companies that did not adapt slowly fading away, still stuck in the maze, while the ladder gets pulled from beneath them.

If you’ve got this far you should DEFINITELY subscribe to my newsletter here.

thank you.

roman