I keep meeting GTM teams paying for 5 different AI subscriptions, and it isn't doing anything for them.
There are 4 stages of AI in GTM, and most teams I talk to are stuck at 3:
1. Mostly manual
2. AI as an answer engine
3. AI as an action engine
4. Shared AI layer across GTM
At stage 3 they bought the AI tools and they know they're working in silos. SDRs have their own Claude skills and email them to each other. The sophisticated ones keep them in a Notion page.
Brian Carroll, who has been doing agentic development for over a year, put the problem perfectly when we talked: it's fine when one marketer uses AI for themselves. But how does a team of 10 get better from each other's outputs?
Here's what a 4 looks like, with one GTM brain shared by every GTM department:
1. The CMO updates the ICP, and it's updated for everybody
2. An SDR drafting outreach gets the updated ICP, case studies and know-how
3. Anyone can spin up automations and async work without waiting for RevOps
4. The CRO sees who's reaching out to whom, reply rates, meetings booked, and whether people follow company guidelines
The honest part: on a personal level, you can play with a harness and get to something decent. I built my own GTM system before I joined Swan AI. At the org level you're dedicating like 3 people to maintain it, keep up as the models get better, and write evals.
Unless you're a giant company, buy.
Which stage is your team at?
I booked 50+ demos last month off our GTM brain.
It has 4 layers:
1. Records: every account, every person, and the funnel stage they're in
2. Strategy: our ICP, competitors and value prop
3. Rules: what we've sent before and what we never say
4. Know-how: the play for every stage
Records are the layer that makes money. Every dot in the video is an account. Every red line is an agent pulling one a stage closer.
Two kinds of agents run on it:
Async agents work in the background. Someone engages with a post, visits the site, changes jobs.. the agent reads the account's history and takes the next step.
Sync agents are the ones you talk to in Slack. A rep says "go after this account" and the agent already knows everything we know.
Both write back to the brain, so every touch starts where the last one ended.
Ours lives in Swan AI.
Do your agents have this level of context?
The most counterintuitive thing I did for pipeline last month: I changed our scoring skill to do LESS.
I raised the bar on who counts as an MQL. Fewer people qualify, so we send less outreach.
Demos doubled.
We launched Swan AI V3 yesterday: KPI-driven agents for the whole GTM team, all running on one shared GTM Brain.
I'm an AI researcher, so I'll say it plainly: aligning agents in production is HARD. Here's why we spent over a year on it:
Building one agent is the easy part. Anyone can do it. The hard part starts when a whole team has them.
10 people, 30 agents, all live.. and each one working from its own version of your ICP, your messaging and your goals. So the outbound agent says one thing, the ABM agent says another, and nobody notices until a prospect does.
Getting all of them to work from one source of truth is a serious engineering problem.
In V3 every agent reads from the GTM Brain and writes back to it. So anyone on the team can build their own agents on the same ICP, the same positioning and the same learnings.
We're not the only ones who see it. HubSpot just started following the "customer harness" positioning.
To everyone trying to build this: good luck, it isn't easy. But when it works, you unlock the real value of AI in GTM, which is alignment.
I one-shotted this launch film with Claude Fable 5.2.
Total cost: $4.20. Render time: 50 minutes.
I told Claude what I wanted and which assets to use. Then it:
1. Wrote the code that builds the whole office in Blender: the bench, the monitors, the lamps, even the paper swans
2. Built the web pages playing on the screens inside the 3D scene
3. Used Codex for the storyboard and the Manhattan skyline
4. Animated 14 camera shots, four swans, and the light moving from dusk to night to sunrise
5. Rented 8 GPUs on RunPod and rendered it
6. Graded it, added the titles and the music
The film: one night in a Manhattan office with Swan AI, just the work getting done.
This is the first draft.
Marketing is now taste + tokens + code.
Anthropic posted a job this fall: "Staff AI Engineer, GTM Claudification." Up to $405K a year.
The company that makes Claude needs a $405K human to make Claude do sales.
I live in Claude Code all day. Here's why it can't run GTM out of the box:
Most GTM teams now have a whole AI workforce that nobody is managing.
5 AI subscriptions. Every SDR with their own Claude skills, emailed to each other.
One marketer using AI is fine. How does a team of 10 get better from each other's outputs?
254 people registered for a conference I spoke at.
Before I walked in I had the buying committee, our full history with each person and a one-line play for every hot account.
It took Swan AI and Claude one afternoon and about $4 of API credits.
Ariel Cohen (@arielco30) replaced event badge scans with a LinkedIn link and a WhatsApp voice note. A badge scan tells you who you met. It doesn't tell you they asked for a demo or said, "Talk to me next quarter." Here's how he set it up. Full episode: https://t.co/hvRnZa7XpB
THIS IS INSANE
We just launched the first KPI-driven agents for GTM teams.
You stop prompting your agents every morning and give them a number to chase instead.
They also run in WhatsApp and Telegram, stay inside your orgβs guardrails, and write everything back to the same brain, so no agent works in a silo.
300+ GTM teams already stopped babysitting their AI.
Itβs live today. Try it now!
Picture 50 people running their own AI agents with no shared rules and no oversight.
Now picture being the CRO who has to answer for it.
That's the pain senior execs kept coming back to at GTM2026.
I added $230K of pipeline in 14 days from one LinkedIn tab almost nobody opens: "who viewed your profile".
What came through:
1. 1,087 profile views
2. 435 ICP-qualified leads (40% of all views)
3. 254 Gold + Silver accounts
4. 14 MQLs, auto-scored in Swan AI and ready for outreach
About 95% of that pipeline came from people I'd never have known were interested.
Everyone obsesses over posting more while their buyers quietly raise a hand in that tab and get ignored.
The intent is already there. You're just not reading it.
We posted 23% more on LinkedIn in October and booked 2x the demos.
The extra posts don't explain it. These 3 things do:
1. Every post throws off signals and we capture all of them: likes, comments, profile views, site visits, DMs. They land in our GTM brain at Swan AI, on the person and the company they belong to.
2. The agent reaching out already knows who they are, what they engaged with, what their company is doing right now and our full history with them. That's why 30-40% of our outreach gets a reply.
3. I changed our scoring skill to do LESS. I raised the bar on who counts as an MQL, so we send less outreach.. but every message goes to someone who can actually buy.
4 of us post. 43 posts in September, 53 in October.
Posting more gets you more engagement. What you do with it gets you the demos.
The best content marketing campaign ever was run by a beer company in 1955.
Guinness printed a book to settle pub arguments and gave it away free. People started STEALING it.
They never asked "how do we sell more beer?" They asked "what are customers already doing?"