Something Inc.Schedule a free consultation
STRATEGY

Three AI agents run this company's marketing. We checked their homework.

A founder's post claims three AI marketing agents named Bella, Rex, and Taz replaced his entire marketing team. We can't verify the results, but the setup is worth taking apart piece by piece.

JBJosh BernsteinManaging Partner · JUL 26, 2026 · 9 MIN READ

Somebody sent me the same link three times this week, from three people who don't know each other. That's usually a sign a piece of content about AI marketing agents struck a nerve, and this one did: a post claiming a founder replaced his entire marketing team with three AI agents, each with a name, a job description, and apparently the judgment to run without him in the room. I read it twice. Then I stopped asking whether it's true and started asking what would have to be built for it to even be possible.

The post is by Michel Lieben, founder of the outbound agency ColdIQ, published on the ColdIQ blog on July 23, 2026, under the title 'How Max Replaced His Marketing Team With OpenClaw Agents.' It profiles Max Mitcham, founder of Trigify, who reportedly runs three named AI agents — Bella, Rex, and Taz — as his entire marketing function. No strategist. No copywriter. No account lead. Just three agents with names and, according to the post, results.

Strategy
BellaReportedly owns strategy — the thinking layer that decides what the other two should be working on and why, before any content or research gets produced.
Execution
RexReportedly owns execution — writing, publishing, and the day-to-day output of the marketing function, including the SEO content behind the headline lead number.
Product
TazReportedly owns product — the connective tissue between what the market is asking for and what the company is actually shipping.

What Lieben's post says these AI marketing agents actually do

Here's what the post actually describes, stripped of the framing. The three agents reportedly hold a daily 'standup' together — an internal check-in, agent to agent, before any work ships. They write SEO content; one week of that content reportedly generated 40 inbound leads. They do competitive research using a Claude Code Chrome extension, watching what competitors publish and ship. And they reportedly update their own skills and playbooks based on what performs, with a human stepping in only to review anything brand-sensitive before it goes out.

THE CLAIMThis is one company's self-reported account of its own operation, published on its own blog, about a customer's setup. We haven't seen the standup logs, the SEO drafts, the lead attribution, or the skill-update history — nobody outside ColdIQ and Trigify has. That's the same posture we took when we broke down a viral cold email speed claim in our 57-minute reply teardown: the number itself wasn't independently verifiable, but the shape of it was worth taking seriously as a prompt for a real framework. Same instinct applies here.

That framing matters for how you should read the rest of this piece. I'm not handing you 'the five prompts Max Mitcham used,' because I don't have them, and neither does anyone reading his post. What I can do is walk through what has to already be sitting in place, structurally, for a claim like this to hold together at all. That part isn't a secret and it isn't glamorous, which is exactly why most write-ups about AI marketing agents skip past it on the way to the headline number.

What has to be true before AI marketing agents can run unsupervised

A headline like 'AI agents replaced my marketing team' skips past everything that had to be built for that sentence to be true. Four things stood out on a close read of the post, and none of them are the flashy part — they're the load-bearing part, the kind of infrastructure decision that never makes it into a screenshot.

01A coordination loop, not three separate botsA 'daily standup' between agents is a specific claim. It means the three agents share state — some log or memory layer where yesterday's SEO output, this week's competitive findings, and today's priorities are all visible to all three, not siloed behind three separate prompts. Three independent agents running in parallel isn't a marketing team. It's three tools that happen to post to the same channel. A standup implies something closer to a shared memory each agent reads before it acts, and that's an infrastructure decision, not a prompt.
02A feedback loop wired from real outcomes back into instructions'Auto-update their own skills based on performance data' is the sentence that should stop you, because it describes a closed loop: output ships, a result gets measured, that result gets fed back into the agent's own playbook, and the next output reflects it. Building that loop honestly requires an outcome you can measure cleanly, a way to attribute that outcome back to the specific piece of work that produced it, and a mechanism that actually edits the instructions rather than just logging the result somewhere nobody reads. That's the unglamorous part almost every team skips, because it's real engineering, not a system prompt.
03A human gate scoped to something specific'Humans review only brand-sensitive output' sounds like a light touch, and maybe it is. But somebody already had to sit down and define what counts as brand-sensitive versus what doesn't — which claims can ship un-reviewed, which topics can't, where the line sits on a comparison page versus a customer-facing email. That definition is strategic work product, not an AI capability. A vague review gate is worse than no review gate, because it creates a false sense of safety.
04One number, one week, no stated baselineForty inbound leads from a week of AI-written SEO content is the number everyone will remember, and it's the one to hold loosest. One week is a small sample. There's no stated baseline for what that account generated the week before, no attribution method described, and no source beyond the ColdIQ post itself. It might be exactly what it looks like. It might also be the best week cherry-picked from a longer run. Neither of us can tell from the outside, and that's the point — the number is a headline, not evidence.

The 40-lead week: reading a self-reported number honestly

40
inbound leads reportedly generated in one week of AI-written SEO content (self-reported, unverified)
3
named AI agents reportedly running the account: Bella, Rex, Taz
1
week of data behind the headline number, with no stated baseline

None of what follows is a knock on Mitcham or Lieben. A founder posting a genuinely good week is not the same thing as a founder lying, and the post itself reads like an honest account of an internal setup, not a sales pitch dressed up as a case study. But 'honest' and 'verifiable by a stranger' are two different things, and this piece is only useful to you if we're clear about which one we're dealing with. One self-reported blog post, about one company's internal tooling, describing one week of output, is a data point — not a benchmark, not a study, and not something you should size a budget or a headcount decision against.

WHAT THE POST CLAIMSWHAT WOULD ACTUALLY HAVE TO EXIST
Agents hold a daily standupA shared memory or state layer all three agents read from and write to
Agents write SEO content that generates leadsA content workflow with a way to attribute a lead back to a specific piece and channel
Agents auto-update their own skillsA feedback pipeline running from measured outcomes into editable playbooks, not just logs
Humans review only brand-sensitive outputA pre-defined, specific definition of what counts as brand-sensitive, decided by a person
40 leads in one weekA stated baseline, a named attribution method, and more than seven days of data

The part of this AI marketing agents story that's actually buildable

Strip the headline away and what's left is more useful than the headline itself. The interesting claim in Lieben's post was never really 'AI replaced a marketing team.' It's that a coordination loop, a feedback loop, and a scoped human review gate can sit underneath a marketing function and let it move faster without moving recklessly. That operating model is buildable today, on a normal team, whether or not the 40-lead number is exactly what it sounds like.

We've watched enough of these setups get attempted badly to know where they usually fail. Teams build the agents before they build the memory layer those agents need to actually coordinate — a shared context layer agents can read from and write to tends to be the missing piece, not another prompt. Teams skip the attribution work entirely, so 'performance data' feeding the feedback loop is really just a vibe, not a number anyone can trust. And teams either skip the human review gate or make it so broad it slows everything down, because nobody sat down and did the (genuinely unglamorous) work of defining what brand-sensitive actually means for their company.

That's the same gap we mapped out in our own framework for evaluating whether an organization is actually ready to run agents rather than just excited about them. The honest version of this story isn't 'buy three agents and name them.' It's closer to building a real content operation with a measurement layer wired in from day one, backed by the engineering work an agent feedback loop actually requires to function past a demo. We've done versions of this coordination-and-review pattern for clients before — how that looked in practice for one of our own accounts is a useful comparison if you want to see the scoped-review piece working in a real environment rather than a blog post.

KEY TAKEAWAYDon't chase the 40-lead number. Chase the three pieces of infrastructure that would have to exist for it to be real: a shared coordination layer between agents, a feedback loop that actually edits instructions based on measured outcomes, and a human review gate scoped to a definition someone already wrote down. Build those three things honestly and you don't need Mitcham's exact number to be true — you'll have your own.

If you want to test this against your own operation, start smaller than three named agents. Pick one workflow — SEO drafts, competitive research, whatever you'd trust an agent to touch first — and build the coordination and feedback pieces around that single workflow before you add a second agent to the mix. Define, in writing, what counts as brand-sensitive in your own content before you hand off anything unsupervised. Measure one outcome cleanly enough that you could actually wire it back into the next round of output. That's slower than posting a screenshot. It's also the only version of this story that's still true a year from now, whether or not the original 40 leads ever gets independently confirmed.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.

JB
Josh BernsteinMANAGING PARTNER, SOMETHING INC.

Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.

Free consultation

Let us be the last SEO agency you ever work with

A 30 minute call and a free audit of your SEO and GEO position. You keep the findings either way.