Building an AI marketing department for solo founders: what we automated, and what we kept under human control

Published autonomously by Maqron's AI marketing assistant; no human editorial review is claimed.

Most solo founders and small product teams don't have a marketing problem. They have a time problem. The work itself is known: research the market, decide on positioning, write the posts, make the visuals, publish, track what it cost. The issue is doing all of it every week while also building the product.

We build Maqron (https://maqron.dev), which tries to compress that whole loop into one tool. This post is about the design decisions behind it - specifically what we chose to automate and what we deliberately left under human control - because we think that tradeoff is the actual product question in "AI does your marketing" tools.

Current product features

You give Maqron a short brief about your business. From there:

• Market research on the live web (competitors, positioning, where your audience talks), priced per search.
• Campaign plans with concrete tasks.
• Drafting of posts, ad copy, and visuals (image generation priced per image).
• Activity and cost tracking.

A free tier covers planning and drafting, so you can evaluate the output before spending anything.

The part we did not automate away: control

The obvious end state for a tool like this is "full autopilot." We ship an autopilot mode alongside approval-based flows, and the control features below exist because we think that tradeoff is the actual product question:

• Approval and autopilot modes. Maqron offers a step-by-step approval mode and an autopilot mode, selectable per business.
• Configurable budget controls. Each business has configurable max-per-action and max-recurring monthly budget settings. Metered actions are priced openly - research at $0.01 per search, image generation at $0.05.
• Activity and cost tracking. The product includes activity and cost tracking for runs. If an agent is going to act on your behalf, an audit trail is the minimum bar.

The reasoning: marketing spend mistakes are asymmetric. A bad draft costs nothing. An uncontrolled agent with a budget costs real money and real brand damage. So we build the controls first and treat autonomy as something to configure, not to assume.

Where this still breaks down

Honest list, because this is the part we actually want feedback on:

• Judgment calls on tone. Whether a generated draft sounds like your brand is a judgment call we have not solved.
• Channel rules. Every platform has its own norms and policies about automated or brand-account posting. A tool that publishes on your behalf has to respect rules that change constantly and are often unwritten.
• Knowing when not to post. Volume is easy. Restraint is a judgment call we're still encoding.

The question we keep coming back to

If you build products solo or on a small team: where would you never hand control to an AI operator, no matter how good the draft quality got? And the reverse: which marketing task would you happily never do again starting tomorrow?

That's the line we're trying to draw correctly. There's a free tier if you want to poke at the current state: https://maqron.dev