We’ve got all been there. You discover a piece of business software that appears like it should clear up every thing. The demo is polished. The touchdown web page guarantees the world.
You join, pay, get inside – and uncover that the function you truly wished is locked behind a “Premium” tier. Or an “Enterprise” plan. Or, my private favourite, a “Contact Gross sales for Pricing” button that results in a 45-minute discovery name the place somebody tries to upsell you on a bundle you by no means wanted.
Article continues beneath
Co-Founding father of Man of Many.
The reason being easy. If an AI agent can do the work of ten folks, why would you pay for ten software program seats?
A $350,000 Salesforce contract was reportedly terminated for a custom-built various. Retool’s 2026 Construct vs. Purchase Report discovered that 35% of enterprise groups have already changed a minimum of one SaaS device with {custom} software program, and 78% plan to construct extra this yr.
I run a media firm, not a software program firm. However once I checked out what we had been paying for – and what we had been truly getting – I made a decision to cease shopping for and begin constructing.
The issue we had been fixing
Man of Many is Australia’s largest independently owned males’s life-style writer. However we’re a small group working in an trade below existential strain.
In the meantime, each SaaS vendor was knocking on the door with level options. One device for SEO. One other for social media scheduling. One other for analytics. One other for CRM. The typical massive enterprise now runs over 2,000 purposes, with greater than 60% not formally accredited by IT. That isn’t a tech stack. That’s device sprawl.
So as a substitute of subscribing to extra software program, I constructed an AI operating system.
What we constructed
Otto OS is an AI Chief Working Officer that runs our back-office operations via a community of specialised AI brokers. It handles editorial workflows, gross sales intelligence, aggressive evaluation, monetary reporting, content material technique, and enterprise monitoring.
The structure follows what is named the WAT Framework – Workflows, Brokers, and Instruments (hat tip to Nate Herk) – which separates AI reasoning from deterministic execution. This separation issues as a result of when AI tries to deal with each step instantly, accuracy compounds downward. 5 steps at 90% accuracy every offer you 59% general success.
I constructed it utilizing Claude Code as each the builder and the mind. No conventional developers had been concerned. The preliminary construct took roughly every week of energetic improvement.
What it truly does
Each morning, Otto generates a briefing that pulls reside income knowledge from Google Advert Supervisor, checks site visitors traits from GA4, scans for SEO anomalies, evaluations excellent invoices in Xero, and flags something that wants human consideration. That used to require somebody logging into 5 totally different platforms and manually compiling the data.
The safety structure makes use of a traffic-light system. Inexperienced zone: read-only entry. Amber zone: draft-only entry requiring human evaluate. Pink zone: any motion involving cash, public posting, or knowledge deletion requires express human approval each time.
What I discovered (and what’s transferable)
1. Map your device sprawl. Listing each SaaS product you pay for. Subsequent to every one, write down which 20% of its options you truly use.
2. Begin with the boring stuff. Reporting. Information consolidation. Standing updates. Invoice chasing. These are the duties the place automation delivers speedy, measurable ROI.
3. Separate considering from doing. Let AI deal with reasoning and orchestration. Let deterministic scripts deal with execution.
4. Use plain textual content as your database. Easy markdown information are the simplest reminiscence system for AI.
5. Safety is non-negotiable from day one. Construct the guardrails earlier than you construct the options.
6. Doc every thing as you go. The system ought to write its personal SOPs.
The larger image
We aren’t the one writer considering this manner. Dow Jones, Enterprise Insider, and Forbes are all investing closely in inside AI programs. Reuters Institute’s 2026 predictions report discovered that 97% of publishers now think about back-end AI automation “essential.”
Bain and Firm frames this as a elementary restructuring of the software program trade. Dean Shahar, who manages a 3 billion euro fund at DTCP, put it bluntly: “The SaaS world is dying. Not software program itself, however SaaS as a enterprise class.”
For us at Man of Many, Otto OS is how a small impartial group competes in opposition to publishers with tons of of employees. It price a fraction of what we had been spending on SaaS subscriptions that underdelivered.
The instruments to do that can be found to anybody, proper now. The query isn’t whether or not your trade will likely be disrupted by AI. It’s whether or not you may be the one constructing the system, or the one nonetheless ready on a vendor to ship the function you wanted six months in the past.
Check out our list of best Large Language Models (LLMs) for coding.
Source link
