Growth Machine DNA (GMDNA) is a practitioner method developed by Dr. Arndt Schwaiger for building an AI-operated growth machine. It runs in five steps: (1) build a Single Point of Truth — the repository of brand, personas, tone of voice, value propositions, competition and market trends that every agent works from; (2) wire pipelines, rebuilding campaigns as measurable loops and opening new channels including visibility in AI answers; (3) generate content from that source, drafted by agents and curated by a human; (4) make tracking non-negotiable before any loop goes live; (5) start the loops and stay in the loop as the human orchestrator. GMDNA is the go-to-market counterpart to Dr. Schwaiger's BMDNA framework: BMDNA decodes the business model, GMDNA builds the growth machine that feeds it.
Growth Engineering
Build the Machine That Finds Your Customers
The first working version of a software product is nearly free these days. Distribution is not — and the machinery that decides who gets found and who gets chosen is increasingly automated. I help startups and scale-ups build the growth machine that runs it: from your GMDNA to the first running loops.
Why This, Why Now
Distribution Is the New Bottleneck
Three numbers frame the shift. Building got cheap, execution moved to machines, and the route to your customer changed shape — all at once.
Meta · Q2 2026
$75bn
Annual revenue run-rate passed by Advantage+, the AI suite that writes, tests and reallocates ads largely on its own. Execution has already moved.
Lovable · 2026
1m / week
New projects started every week on a single vibe-coding platform. The market is flooding with decent first versions — yours included.
Ahrefs · 2026
−58%
Click-through drop for top-ranked pages where Google's AI answers appear. Buyers increasingly get their shortlist from an answer, not a results page.
What has become nearly free is the first working version. UX, testing and taste still need humans, and the more human-centred the product, the more of them it needs. But that makes the problem worse, not better: getting found and getting chosen is now the scarcest part of building a company. The question is no longer whether you can build your product — it is whether you can build the machine that reliably finds and keeps your customers.
The Role
From Growth Hacker to Growth Engineer
Marketing went analytical around 2010, when Sean Ellis coined “growth hacking”: tracking, attribution, A/B tests. The next step is happening now — AI agents do the execution itself. The person who thrives in this setup is the growth engineer: a builder measured on lift in one metric, orchestrating a stack of agents instead of running campaigns by hand. The role is going institutional — a16z launched a Growth Engineer Fellowship this year, and OpenAI runs a dedicated GTM growth engineering team.
- Measured on one metric, not on campaign output
- Orchestrates agents instead of briefing them
- Owns the integration layer: APIs, MCP servers, connectors
- Decides what to kill and what to scale — in writing, before the test
- Writes every learning back into the source repository
- Treats tracking as prerequisite, not reporting
Practice, Not Theory
How I Run My Own Growth Machine
I hold the growth engineer role at one of my own ventures. Everything on this page is what I run there before I recommend it to anyone else.
The account runs on scripts I wrote
At one of my ventures the entire ad account runs through scripts I vibe-coded myself — from building the campaigns to the daily checks. It runs better than it did by hand, with a fraction of the manual work.
I learned step 4 the honest way
At that same venture the true bottleneck was not creative and not budget. It was a tracking gap that kept purchases invisible to the platform. Everything the machine optimised before that fix was noise — which is why tracking now comes before loops, not after.
Mornings are thirty minutes of decisions
Campaigns used to be handwork spread across the week: briefings, drafts, uploads, reports. Today the machine does that overnight and I read what the loops did, make the kill and scale calls, and write the learnings back. Same channels, same budgets, a different job.
A note on scope, because it matters when you are choosing who to work with: growth engineering is the newest part of my practice, built on my own ventures and on client work in progress — not on the twenty-year track record behind my advisory and business-model work. What I bring to it is a PhD in multi-agent systems, a habit of building the thing myself, and a machine I run every morning.
How We Work Together
Three Ways In
Each one is designed to stand on its own. Most teams start at the first and stop when the machine runs without me.
GMDNA Sprint
We build your Single Point of Truth: what goes in it, how it is structured, how it stays in sync so dozens of agents work from it without drifting off brand. You leave with the repository and the rules for maintaining it.
- Brand, personas, tone of voice
- Value propositions & competition
- Structure, ownership, sync rules
- The review gate before anything ships
First Loop Live
One channel, wired end to end, until a loop is running and moving one metric. Tracking goes in before the loop does. We pick the channel together — the one that earns you the credibility to wire the next.
- Integration layer: APIs, MCP, connectors
- Conversion & AI-referral tracking
- Content generated from your GMDNA
- A scorecard your team actually trusts
Growth Engineering Advisory
Ongoing, as more loops come online: the review cadence, the kill and scale calls, the discipline that stops a machine from optimising itself into noise. For teams building the role in-house, this is where the parameters get handed over.
- Cadence & learning-phase discipline
- Test design and decision rules
- Scaling from one loop to several
- Coaching your first growth engineer
My Framework
Every growth machine has a DNA
Everyone can rent the same models. Two competitors prompting the same system get roughly the same ad. The difference is what you feed it — which is why the method starts with a repository, not a tool.
BMDNA decodes the business model. GMDNA builds the growth machine that feeds it. One is the genome of how you make money; the other is the genome of how you reach the people who pay.
The Method
Five Steps, In the Order That Works
-
1
Build Your Single Point of Truth
The repository everything else derives from: brand and personas, tone and voice, value propositions, competition, market trends — everything the machines cannot invent. Encode it once and every automated touchpoint still sounds like you. What belongs in it is easy to guess; how it is structured and kept in sync is the actual craft.
-
2
Wire Your Pipelines
Rebuild existing campaigns as loops: agents draft variants, launch small, measure lift against your one metric, kill the losers, scale the winners. Then open the channels you do not serve yet — visibility in AI answers is a build task with shipped artefacts, not a media buy. Start with one channel, not five.
-
3
Generate the Content From the Source
Only now does content enter: posts, ads, landing pages, video — drafted by agents from the Single Point of Truth, curated by a human before anything ships. Skip step one and you get faster generic output, which is worse than slower generic output. Adoption is not advantage. Your source is.
-
4
Make Tracking Non-Negotiable
Before a single loop goes live, the machine has to be able to see: conversions back to the ad platforms, every channel separately, AI-assistant referrals tracked on their own. For EU founders, GDPR and the AI Act belong in the definition of done. Without this, every later decision is guesswork dressed up as data.
-
5
Start the Loops, Stay in the Loop
Define what a winner looks like in writing, before the test starts. Review on a schedule, make the kill and scale calls, and write what you learned back into the source. That write-back is what makes the machine compound: every review makes the repository smarter, and every agent works from the smarter repository.
Two things separate the professionals here, and neither is a tool. Restraint: bidding algorithms need stable learning phases, so a system that changes everything daily performs worse than one with a disciplined cadence. And patience: some tests need longer runtimes before they mean anything, and killing them early means deciding on noise. When generating fifty variants costs nothing, the bottleneck moves from production to judgement.
Questions
Before You Ask
GMDNA is my method for building an AI-operated growth machine. It runs in five steps: build a Single Point of Truth, wire your pipelines, generate content from that source, make tracking non-negotiable, then start the loops and stay in the loop as the human orchestrator. The name is deliberate — the repository in step one really is the DNA of everything the machine produces afterwards.
BMDNA decodes the business model. GMDNA builds the growth machine that feeds it. BMDNA makes a business model measurable through 9+1 metric categories and target metrics like CAC, CLV and MRR. GMDNA is the go-to-market counterpart: the machine that actually moves those numbers. They work well together and neither requires the other.
A growth engineer is measured on lift in one metric — sign-ups, activation or revenue — and orchestrates a stack of agents rather than running campaigns by hand. The role is going institutional: a16z launched a Growth Engineer Fellowship this year and OpenAI runs a dedicated GTM growth engineering team.
No. The agents deliver volume and consistency; you deliver taste. Every piece of content passes a human review gate before it ships, and a human makes the kill and scale calls. What changes is the work itself: less production, more orchestration and judgement. When generating fifty variants costs nothing, the bottleneck moves from production to deciding which one is right.
Whichever one is actually your bottleneck — and that differs in every team. Some have no source repository at all. Some have a good one but no tracking, so every optimisation is guesswork. Some have both and no discipline in how tests are run. Working out which step is holding you back is usually the first conversation we have, and it is worth having before anything gets built.
Less time than you would expect for the plumbing and more than you would like for the judgement. With today's coding agents the integration work takes days rather than quarters. A realistic path is a short sprint to build the Single Point of Truth, then a few weeks to get one channel wired end to end with tracking in place. What takes longer is the discipline: stable learning phases and tests given enough runtime to mean something.
Both, online or in person. Most growth machines end up running in English regardless of where the team sits, but the working language is yours to pick.
Let's build it
Which Step Is Your Bottleneck?
If you want hands-on support building your growth machine — from your GMDNA to the first running loops — that is exactly what I do with startups and scale-ups. Tell me where you are stuck and we will start there.
Get in touch