The Operator's Stack: Running Six Companies Without Back-Office Headcount

Marine veteran. 32 years in enterprise IT — communications, collaboration, and infrastructure. No practical development or coding experience before this build. Two W-2 consulting engagements. Six companies under one holding entity. No back-office staff. This is the writeup of what running that experiment actually looks like — the tools, the costs, what it replaced, and where the ceiling still is.

The question I was trying to answer

I want to be upfront about what this actually was, because it matters for understanding why I built what I built. I wasn't trying to prove a point about AI. I was trying to answer a specific operational question: how far can one person scale a portfolio of real, operating businesses before the back-office overhead becomes the constraint?

Every business eventually hits a point where the operational overhead — the email, the task management, the CRM hygiene, the financial monitoring, the document handling — starts consuming the time that should be going into the business itself. The usual answer is to hire. I wanted to find out how far I could push the ceiling before that answer became unavoidable.

The six companies I was running the experiment across: Safire Business Services (B2B tech consulting), Safire Home Solutions (smart home automation), Invictus Systems (no-monthly-fee home security), ProfPrep (AI-powered exam prep), Strategic Series (AI professional publishing), and Noevant (AI operations commercialization). On top of those six: two W-2 consulting engagements with enterprise clients, both operating under NDA.

Each company has its own primary mailbox, a support address, and a set of aliases. Twenty-five mailboxes across six brands. Not six email accounts — twenty-five mailboxes.

Why the tools I had weren't the problem — and why they weren't the solution either

I wasn't disorganized before I built this. I had Zapier, GoHighLevel, a CRM, accounting software, and a reasonable system. The problem wasn't that I didn't have tools. The problem was that the tools I had were all opinionated about how I should work, and their opinions didn't match mine.

Zapier is a good product. But every automation I needed had a conditional, an edge case, or a data shape that Zapier's UI couldn't express cleanly without stacking five steps and hoping the parsing held. I replaced it with Cloudflare Workers. Writing the Worker took longer upfront. It hasn't needed touching since.

The deeper issue was tool conformity. Most software wants you to conform to it. It has a workflow baked in, and your job is to fit yourself into that workflow. That's fine when the tool's opinion matches yours. It's expensive — in time and mental overhead — when it doesn't. What I wanted was a system that conformed to how I actually work, not the other way around.

JARVIS — purpose-built for exactly one operator

JARVIS is the system I built to solve that problem. It's not a product I bought. It's a custom AI chief-of-staff running on Cloudflare Workers and Supabase, built specifically for how I run this portfolio.

What it actually handles across those twenty-five mailboxes and six companies:

• Email triage — classifies, summarizes, flags actionable items, quarantines spam across all mailboxes
• CRM sync — new contacts from GoHighLevel flow to Microsoft 365, deduplication runs automatically
• Financial monitoring — Plaid pulls transactions, Wave reconciles, anomalies surface as tasks
• Task management — creates tasks from email, calendar, and manual input without app switching
• Knowledge retrieval — 300+ entries covering every credential, policy number, serial number, and process I've documented
• DNS and infrastructure management — Cloudflare zone edits, worker deploys, R2 operations on command
• Callback and SMS routing — voice agents fire webhooks, JARVIS sends branded confirmations and alerts per company
• Document processing — scan a receipt or SOW, JARVIS extracts and files the key facts
• Content publishing — Ghost blog posts across 8 publications, social variants per platform (in progress)
• Pre-meeting briefs — before any client call, JARVIS pulls everything it knows about who I'm meeting

What it runs on and what it actually costs

The full infrastructure runs at approximately $150 per month:

• Cloudflare Workers, KV, R2 — ~$10/mo
• Supabase (database + auth) — ~$25/mo
• Hostinger VPS (Ghost, 8 blog publications) — ~$20/mo
• Anthropic API (Claude Haiku for classification, Sonnet for agent responses) — ~$40/mo
• GoHighLevel (shared across all sub-accounts) — ~$30/mo
• Resend, EZTexting, and miscellaneous — ~$25/mo

Wave and Plaid are correctly characterized as good tools doing one thing well. I'm not replacing them — I'm connecting them. The tools that remain are the ones that earned their place by not requiring me to change how I work to use them.

Before and after — the honest ledger

Before JARVIS, I was either paying for or seriously considering: a fractional executive assistant (~$800/mo), a part-time marketing coordinator (~$400/mo), separate automation tools to replace the Zapier workflows that weren't scaling (~$150/mo), and a document management system (~$100/mo). That's roughly $1,300–$1,500 per month in recurring costs or headcount I hadn't pulled the trigger on yet.

JARVIS didn't eliminate the need for those functions. It moved the ceiling on what one person can manage before those functions become necessary. That's a meaningful distinction. The ceiling moved. It didn't disappear.

What it actually took

I want to be direct about the build cost, because this gets glossed over in most AI operator case studies. I spent real hours building this. Not hundreds of hours — Claude did the heavy lifting on implementation — but real hours of architectural decision-making, debugging, testing, and iterating. The system I have now is on its second major architecture. The first version worked. The second version works better and costs less to run.

The honest version: if you're a non-technical operator and you're thinking about building something like this, the build itself is not the hard part. The hard part is knowing your own operations well enough to know what you actually need automated versus what you think you need automated. Those are different lists.

What works, what doesn't, and what I haven't solved yet

What works well: email triage, task creation, knowledge retrieval, financial monitoring, SMS routing, and document processing. These are stable and I rely on them daily.

What's still being built: the content and social publishing workflow. The infrastructure is in place — eight Ghost publications, Cloudflare proxy workers, social accounts established. The automation that goes from a topic brief to published posts across all eight properties is still being built.

What I haven't solved: anything that requires genuine judgment about client relationships, priority calls between the companies, or reading a room. Those stay with me. No amount of automation changes that, and I'm skeptical of anyone who tells you otherwise.

What this actually does to how you operate

The practical effect is different from what I expected. I expected time savings. What I got was cognitive overhead reduction. The difference matters. Time savings means you do the same things faster. Cognitive overhead reduction means you stop carrying the weight of the things you haven't done yet, because the system is handling them or surfacing them at the right time.

Running six companies without a back-office team isn't about working longer hours. It's about eliminating the mental overhead of tracking what needs to happen across twenty-five mailboxes, six CRMs, six Stripe accounts, and six brands simultaneously. JARVIS doesn't run the companies. It runs the operating layer so I can run the companies.

Where this goes — and the two paths for someone reading this

Noevant is the commercialization vehicle for this system. After roughly 60 days of internal validation across all six companies, the plan is to offer two things: a custom-built version for operators who want exactly what I have, built for their specific operation — and a productized version for operators who want the core leverage without the custom build timeline.

Neither product exists yet in a form I'm selling. What exists is a working system, a clear thesis, and a track record of it actually running a real portfolio. If you're an operator scaling without headcount and want to talk through what this could look like for your business, reach out at jesse@2057hldgs.com. If you want to follow along as Noevant builds toward launch, you're already in the right place.

Read the personal account of how this was built on Jesse Myers' blog, or see how it fits into the broader portfolio strategy on the 2057 Holdings blog.