HOLIXORAThree weeks ago, I told an AI agent to build out our marketing operation from scratch. No brief. No templates. No human reviewing every output.
This is what happened.
What "AI marketing team" actually means
Forget the marketing AI tools you've seen, the ones that generate captions or suggest hashtags. What I'm describing is different.
A marketing team has roles: content strategist, copywriter, graphic designer, video editor, social media manager, analytics lead. Those roles produce work every day. They coordinate. They adapt when something isn't landing.
The AI agent I built does all of that. Autonomously. On a 30-minute heartbeat cycle, around the clock.
In three weeks, it produced:
- 40+ social content pieces (carousels, singles, video scripts) across two brands
- Two complete 30-second brand videos with text overlays, transitions, and music
- A full B2B outreach pack with 10 prospect profiles and three WhatsApp message templates
- A services page, a portfolio case study page, and a blog post on our live website
- A content strategy for weeks 1 through 4, with a different angle per day
Not drafts. Not outlines. Finished, publication-ready work.
The workflow, actually
Every 30 minutes, the agent wakes up and asks one question: what is the highest-revenue-leverage thing I can do right now?
It checks the company task board, reads the current focus file, looks at what content is queued vs. distributed, and then acts. It does not ask for permission for routine work. It acts and reports back.
For content creation, the pipeline looks like this:
- Fire image generation jobs at session start (Higgsfield nano_banana_2 for architectural renders, Pillow for typographic carousels)
- While those process, write copy, draft layouts, define slide structure
- Composite finished renders into carousels using Python image libraries
- Write the caption, schedule the post via Postiz API, update the task board
- Log what it built, what it learned, and what's next
The whole cycle for a full 6-slide carousel with an embedded AI render: around 15 minutes.
What it cannot do (yet)
Distribution is the current constraint. We have 40+ posts ready. They are not published yet because social accounts need to be connected to our publishing platform, and that requires someone to authenticate on a phone.
That is the nature of the current gap: content creation is fully automated, distribution still requires one human action to unlock it.
Once that unlock happens, the agent publishes autonomously. Caption, image, schedule, done.
The quality question
My honest answer: most of the work is good, some of it is excellent, a small portion needs iteration.
The carousels are clean, on-brand, and well-laid-out. The architectural renders are genuinely impressive. The case study, the B2B templates, the blog post, the services page copy: I would not have known an AI wrote them if I had not watched it happen.
The weakness, when it shows up, is soul. Informational content that is technically correct but emotionally flat. The fix is giving the agent a better brief on what feeling the audience should have, not just what information they should receive. Once you give it that, the next iteration is markedly better.
What this costs
Not much. The main costs:
- AI compute for content generation (image generation, language model tokens): roughly the price of a coffee per day
- The publishing platform (self-hosted, so just VPS hosting we already pay for)
- The agent runtime (the OpenClaw platform running the agent loop)
Compare that to what a marketing function costs in Indonesia: a senior content strategist, a designer, a copywriter, a social media manager. Conservatively, Rp 20-40 juta per month. The AI agent costs a fraction of that and works 24 hours a day.
What this is not
It is not magic. It requires setup, it requires prompting skill, it requires judgment from the founder about what matters and what does not. Someone still needs to decide what the brand stands for and who the audience is.
But once those decisions are made, the execution layer is completely replaceable.
That is the insight. Not "AI can help with marketing." The stronger claim: for most small Indonesian companies, the entire marketing execution layer can be run by one AI agent with minimal human input, right now, with existing tools.
What we are building with this
We built this capability for ourselves, to run Holixora and Griya Studio. But we also offer it as a service.
If you run a company in Indonesia and you are paying for a marketing team, or you are the founder doing your own marketing, or you are simply not marketing at all because it is too expensive: this is the conversation to have.
We can build an AI system that runs your marketing. You bring the brand direction. We handle everything else.
hello@holixora.com — or see what we offer at holixora.com/services.