Bart Knijnenberg. Marketing since 2008. Building since always.

Three products. Zero employees.

I replaced my marketing department with eighteen AI agents. They work the night shift. At seven in the morning the work is done and there is exactly one thing left for a human to do, which is decide what goes out.

system idle CET --:--:-- next run in -- nightly window 22:30 → 07:00 18 agents in sequence last cycle , 1/2 completed · 1 stopped at a gate 14 nights

The pipeline runs every night from 22:30 to 07:00 Amsterdam time. Fixed order, each agent reads what the previous one produced. Nothing leaves the building without my approval. The office above is that team drawn small: eighteen robots on the night shift, the green light marks whoever is on duty, the Chief of Staff walks the floor asking every specialist a different question, the coffee corner is there for morale rather than for coffee, and the desk on the right is where everything waits for my GO. The last cycle in the header is read from the run logs when this page is built, so it is the real number and not a claim.

Worked with, over 18 years

  • Amazon
  • Adyen
  • Heineken
  • Financial Times
  • Polestar
  • WE Fashion
  • Colourful Rebel
  • Nestlé

What is running

Three products, built and marketed by the same system. Each with an honest status, because two of them are not finished.

Live, with paid tiers

PlainConsent

GDPR-proof cookie consent at a flat price per site. No pageview limits, no surprises on the invoice.

The AI team took this from first line of code to live site, including security reviews that killed the work twice before it was allowed through.

plainconsent.com →
Opens September 2026

Ensemblo

A complete AI marketing team for Dutch companies that do not have one. You send a message, your team does the work, you approve with one click.

This is the productised version of what I run myself. I only started selling it after it had run in my own business for a year.

ensemblo.nl →
In production, real users

Sterka

An AI strength coach that knows your whole body, from training and nutrition to recovery and injuries, and turns that into one calm answer per day. No confetti, no streak shaming.

The system is not marketing-only. The same agent structure does the engineering here: plan, build, review, test, ship. Twelve coach personas, multilingual, in production.

no public link yet

Sterka has no public site yet. These are mockups from its own design system, not screenshots of the app.

Sterka product view: one calm daily answer with sleep and recovery context
One answer per day
Sterka product view: estimated one rep max progression over eight weeks
Progression, week by week
Sterka brand card: quiet strength, a little more every week
The brand in one line

How it works

A fixed crew and a fixed route, and the route ends at me.

Bart sets the priorities what counts this week 22:30 to 03:30 the work 18 specialists, fixed order 03:45 and 04:00 the gates visuals, then copy 07:00 the briefing one page, three actions and the decision feeds the next night
One night, end to end. The two gates are the only places where work can be sent back, and the only human decision is the last one. Times are Amsterdam time and match the schedule on this page.

18 agents, one job each

Strategy, SEO, content, design, outreach, advertising, data, branding, publishing. Plus a management layer that guards priorities and a development team that builds the products.

It runs at night

The pipeline starts at 22:30 and runs until 07:00. Each agent reads what the previous one produced and hands off to the next. I wake up to a one page briefing.

Nothing passes the gate without review

Copy goes through brand review, visuals through creative review, code through security and quality gates. Work that fails goes back, not forward.

I approve what goes out

Autonomous in execution, me in control. That is not a marketing line, that is how it is built. External actions are closed by default.

The crew · 18

What broke

Building in public only works if the failures are in there too. A few:

first release hard fail

Two security reviewers gave my first product release a hard fail. XSS through innerHTML, unvalidated input, prototype pollution. They were right. It only shipped after the fixes.

months wrong

One agent kept using the wrong job title in external copy for months. I had to build a separate brand gate for it.

3 weeks deleted

I built and shipped an integration, then deleted it three weeks later because it depended on an approval that never came.

factor 66 wrong

My own two websites published the same number with a factor 66 difference for months. Nobody flagged it. I found it myself.

the 18 years underneath

This is not an AI hobby

I have been doing this for 18 years. In that time I have helped 200+ companies and allocated €200M+ in media budget, across agencies, platforms and in-house teams. I took The Next Web to 70,000+ attendees and grew Labelnone from 2 to 27 people in twelve months.

So I know how this works with people. That is exactly why I can see what a system does and does not take over from them.

18years in marketing
200+companies helped
€200M+media budget allocated over 18 years
Bart Knijnenberg

Work with me

Next to building I take on a few Fractional CMO engagements a year. Senior marketing leadership for a company that needs it and is not ready to hire it full time. And I speak about AI teams, usually about the parts that did not work.

Fractional CMO

2 to 4 days per month, strategy and execution. Cancel monthly.

Review your AI setup

One session, I look at what you are building.

Speaking

On autonomous AI teams, with the numbers and the failures included.

Want to see whether this works for you?

No marketing team and you want one? Look at Ensemblo. You have a team and you are stuck? Book a call. Just curious how I built this? Read the blog, it is all there.