How I built an AI operating system to run my publishing company

We have all been there. You find a piece of business software that looks like it will solve everything. The demo is polished. The landing page promises the world.

You sign up, pay, get inside – and discover that the feature you actually wanted is locked behind a “Premium” tier. Or an “Enterprise” plan. Or, my personal favorite, a “Contact Sales for Pricing” button that leads to a 45-minute discovery call where someone tries to upsell you on a package you never needed.

Article continues below

Scott Purcell

Co-Founder of Man of Many.

The reason is simple. If an AI agent can do the work of ten people, why would you pay for ten software seats?

A $350,000 Salesforce contract was reportedly terminated for a custom-built alternative. Retool’s 2026 Build vs. Buy Report found that 35% of enterprise teams have already replaced at least one SaaS tool with custom software, and 78% plan to build more this year.

I run a media company, not a software company. But when I looked at what we were paying for – and what we were actually getting – I decided to stop buying and start building.

SEO. Another for social media scheduling. Another for analytics. Another for CRM. The average large enterprise now runs over 2,000 applications, with more than 60% not formally approved by IT. That is not a tech stack. That is tool sprawl.

So instead of subscribing to more software, I built an AI operating system.

What we built

Otto OS is an AI Chief Operating Officer that runs our back-office operations through a network of specialized AI agents. It handles editorial workflows, sales intelligence, competitive analysis, financial reporting, content strategy, and business monitoring.

The architecture follows what is called the WAT Framework – Workflows, Agents, and Tools (hat tip to Nate Herk) – which separates AI reasoning from deterministic execution. This separation matters because when AI tries to handle every step directly, accuracy compounds downward. Five steps at 90% accuracy each give you 59% overall success.

developers were involved. The initial build took roughly a week of active development.

Google Ad Manager, checks traffic trends from GA4, scans for SEO anomalies, reviews outstanding invoices in Xero, and flags anything that needs human attention. That used to require someone logging into five different platforms and manually compiling the information.

The security architecture uses a traffic-light system. Green zone: read-only access. Amber zone: draft-only access requiring human review. Red zone: any action involving money, public posting, or data deletion requires explicit human approval every time.

Invoice chasing. These are the tasks where automation delivers immediate, measurable ROI.

3. Separate thinking from doing. Let AI handle reasoning and orchestration. Let deterministic scripts handle execution.

4. Use plain text as your database. Simple markdown files are the most effective memory system for AI.

5. Security is non-negotiable from day one. Build the guardrails before you build the features.

6. Document everything as you go. The system should write its own SOPs.

automation “important.”

Bain and Company frames this as a fundamental restructuring of the software industry. Dean Shahar, who manages a three billion euro fund at DTCP, put it bluntly: “The SaaS world is dying. Not software itself, but SaaS as a business category.”

For us at Man of Many, Otto OS is how a small independent team competes against publishers with hundreds of staff. It cost a fraction of what we were spending on SaaS subscriptions that underdelivered.

The tools to do this are available to anyone, right now. The question is not whether your industry will be disrupted by AI. It is whether you will be the one building the system, or the one still waiting on a vendor to ship the feature you needed six months ago.