Greg Isenberg put a name on a job I think will become one of the hottest in business over the next five years: marketing engineer.
I had just read his email describing the role when I realized we had been moving toward it since ChatGPT arrived. We simply did not have the title.
A marketing engineer is not a marketer who learned a few prompts. It is someone who can hear what customers are saying, recognize an opportunity, build the page or campaign, connect the data, launch the experiment and make the next run smarter.
The person with the idea and the person who can build the idea have become the same person.
That changes how much one good marketer can do. It also changes what a good marketer needs to know.
The distance between an idea and an experiment is collapsing
For most of my career, marketing ideas were easier to produce than to ship.
A salesperson would hear the same objection five times. A customer-success person would notice why customers left. A founder would say something sharp on a call that belonged on the homepage.
Then the idea began its journey.
Someone wrote a brief. Someone else turned it into copy. A designer made it presentable. A developer found time for it. An analyst tried to connect it to the outcome. The small, specific idea competed with every big launch already on the roadmap.
Most ideas died somewhere in the relay race.
AI has not eliminated the need for taste, strategy, design, engineering or analysis. It has allowed one capable person to carry an idea through more of those disciplines without handing it off at every border.
That person can read sales-call transcripts in the morning, find a pattern, build a useful calculator or landing page in the afternoon, launch a controlled test and see what happened.
The value is not that the company produces more material. The value is that it can run more good experiments, learn faster and carry what it learned into the next experiment.
That is why I think marketing engineering will become such a valuable job.
The job is not prompting. It is building a learning system.
Isenberg describes marketing engineers as part marketer and part builder. They connect agents that listen to customer language, watch for buying signals, produce creative, reopen lost opportunities and connect activity to revenue.
His argument is bigger than automation.
The marketing engineer builds a system that gets smarter.
That requires judgment. Near the end of his podcast episode on marketing engineering, Isenberg says, “The agents are going to be a commodity at some point. Your judgment about what to point them to is the moat.”
I think that is exactly right.
Everyone will have access to capable agents. Everyone will be able to produce a landing page, an ad campaign, an email sequence and fifty pieces of content.
The advantage will come from knowing which problem is worth solving, which customer language matters, which claim is true, which experiment teaches you something and which result should change the system.
The tools make execution cheaper. They do not make judgment automatic.
Before you build the agents, build the brain
The section of Isenberg’s email that mattered most to me was his instruction to start with the brain.
He proposes a structured company folder containing customer fit, voice and founder opinions. Every agent reads it before doing any work.
This sounds obvious. It is also the step most companies skip.
They spin up agents, connect some tools and ask for volume. The agents come back with polished, generic material. The team blames the model. Then it tries a different model and gets a different flavor of the same average answer.
The agent was never the real problem. The company gave it nothing specific to believe.
A marketing machine needs a source of truth about:
- Who the company is for.
- What those people are struggling with.
- What the company believes about the problem.
- Why it solves the problem differently.
- What it can prove.
- How it sounds when it tells the truth well.
- What it will not claim, invent or publish without review.
At PitchKitchen, we document that narrative brain in a Magnetic Messaging Framework.
It is more than a brand-voice file. Voice tells the machine how to sound. Narrative identity tells it what the company believes, who it serves, what change it wants to lead and why its version of the solution deserves to exist.
Without that foundation, marketing engineering becomes noise engineering.
We learned this through three generations of the AI Brand Twin
When ChatGPT arrived, we began using it as a working partner in messaging and marketing.
The first phase was conversational. Give the model enough context, ask for something useful and refine the answer. It was impressive, but the context disappeared into individual chats. Every new session started with another explanation of the company.
Custom GPTs improved that. We could load a client’s messaging framework and standing instructions into a reusable assistant. We began calling the result an AI Brand Twin because the value was not the model. The value was the company’s own story, point of view and voice installed on top of the model.
Custom GPTs also revealed the next problem.
The company’s truth was trapped inside a vendor-specific container. The client might use ChatGPT while an agency used Claude and a developer used Cursor. Each tool carried another copy. When the company changed a claim or sharpened its positioning, someone had to remember every place the old version lived.
By the end of 2025, we stopped treating the Custom GPT as the main deliverable.
We delivered the AI Brand Twin as a Markdown document. Clients could save that document in a shared Claude Project or a ChatGPT Space and give the team access. The truth had become portable. The company could own the file instead of rebuilding the brain every time it changed tools.
That was better. It still created copies.
In August 2026, we launched the PitchKitchen MCP. The AI Brand Twin could now live in one controlled place while ChatGPT, Claude, Cursor and other connected tools retrieved it when they needed it.
The difference is simple:
A shared project gives everyone a copy of the brain.
An MCP gives every connected tool access to the same brain.
Update the source once and the connected tools work from the current version. The company no longer has to hope that everyone remembered to replace a Markdown file or open the correct project.
We now call this AI Brand Twin over MCP.
The model can change. The interface can change. The company keeps control of the source.
Our marketing department now looks more like a system
PitchKitchen currently operates with nine defined AI roles inside Claude Code.
One scans Reddit, X and Hacker News for buyer complaints. One checks what ChatGPT and Perplexity say about PitchKitchen. One identifies website visitors worth reviewing. One works my warm network. One helps turn ideas into articles in my voice.
That description can sound as if I replaced nine employees with nine robots.
I did not.
Each role has a specific job, approved sources, access limits, expected outputs and a human responsible for the result. Drafts remain drafts until someone approves them. Guesses do not become company policy because an agent wrote them confidently. Lessons do not become permanent until we review them and update the appropriate source or role instructions.
The most important part of the system is not how many agents it has.
It is that the agents do not each invent their own version of PitchKitchen.
They work from a shared narrative. They have separate jobs. Their output comes back to a human who can decide what is true, useful and worth acting on.
That is marketing engineering.
The best marketing engineers will do six things well
The job will keep changing, and the title may not survive. The work will.
The strongest marketing engineers will know how to:
- Listen. Pull useful signals from customers, sales calls, support, search behavior and the market.
- Decide. Separate an interesting pattern from a problem worth acting on.
- Document. Turn company beliefs, customer truth, proof and voice into a source the system can use.
- Connect. Give each agent the right context, tools, access and boundaries.
- Experiment. Turn signals into pages, campaigns, tools and offers quickly enough to learn.
- Govern. Decide what ships, what changes and what becomes part of the company’s memory.
This is why the job sits closer to leadership than production.
The marketing engineer is not merely making things. The marketing engineer is deciding what the company’s marketing machine should know, what it should build and what the result means.
Speed makes truth more valuable
For years, companies could survive fuzzy messaging because producing and distributing material was expensive. The friction limited the damage.
That protection is disappearing.
A company can now generate hundreds of pages, ads, emails and posts before anyone notices that all of them are built on the wrong story.
AI makes it possible to move faster than ever. That makes the source more important, not less.
The marketing engineer will be one of the hottest jobs of the next five years because every company will want faster experiments, better customer intelligence and more output from a smaller team.
The best ones will not begin with the agents.
They will begin with the truth the agents are being asked to scale.
Sources and further reading
- Greg Isenberg, “Making $$$ as a Marketing Engineer”, The Startup Ideas Podcast, August 31, 2026.
- PitchKitchen, AI Brand Twin over MCP.