I've been my own boss since 2009 — for over fifteen years I've owned and run an architecture and engineering group — a real business, with real payroll, real deadlines and real margin pressure, in a market where bureaucracy weighs more than the wage bill. I do not write code. Everything I'm about to describe was built by directing AI the way I'd direct a sharp junior who works for pennies and never sleeps.
Eighteen months ago I made a quiet decision: when someone left the back office, I would not replace them. Not a layoff — an experiment, with one question attached. Could AI close the gap before I had to post the next job ad? The honest answer surprised me. Today my company runs on two engines I built myself — and a third that you are reading right now.
What follows isn't theory. It's an inventory of what's actually running, described by what it does for the business, not by how it's wired underneath.
01Engine one — the back office that runs itself
The unglamorous core: the work that used to eat salaries and swallow partners' evenings, now handled by a set of always-on assistants. And here's the key point: this back office looks the same everywhere. Whether you make salads for a supermarket chain or run a law firm, your invoices, correspondence, reports and documents look almost exactly like mine. My example is from construction — but the problem is yours.
Invoices. Through the day, the system checks the accounting inbox, reads each invoice, and quietly handles the validation a careful bookkeeper would do — the tax logic, the cost allocation, the duplicate checks, the dozen local quirks I'd never trust a generic tool to get right. The clean ones it files; only the genuine exceptions reach a human. It now does about ninety percent of what a junior accountant used to do for me — without a raise, a holiday or a resignation.
Correspondence. Every piece of incoming mail gets read, classified, dated, given its deadline and logged — the kind of statutory register most firms keep badly until a missed date bites them. Nothing slips through now. Not because we got more disciplined, but because the assistant doesn't sleep.
Opportunities. The public tenders in our market arrive in a daily firehose nobody has time to read. The commercial tools that watch it cost a fortune; specialists cost more. So I built my own watcher — it reads every notice, scores it against what we actually do, and surfaces only the live ones. That isn't replacing a person. It's a capability I'd long written off as unaffordable, now running for the price of a few coffees a month.
Filings, confirmations, contracts, classification. A cluster of smaller assistants generates the regulatory documents we're forever filing, reconciles supplier confirmations, drafts commercial contracts against our own playbook, and routes incoming leads to the right place. None dramatic alone. Together they hand a senior partner his evenings back.
The thing nobody tells you: the AI is the easy part. The hard part is encoding an operator's judgement about which cases a human still has to see. That judgement is the product.
02Engine two — the machine that markets the new company
Here's the part that should matter most to you, because it isn't about my old company. It's about this one. The article you're reading was published by a system I built.
An idea becomes a topic. A topic becomes a draft. The draft is shaped into the company's voice. And then — the step every "fully automated" pitch quietly skips — it stops and waits for a human to say yes. Only after approval does it schedule itself and go out across every channel, in two languages. AKINT.AI's entire content presence runs on it.
The point isn't the automation. The point is the gate. AI does the heavy lifting; a person keeps the judgement. That is the exact philosophy I bring to anything I build for another owner: speed, without surrendering control.
03What every one of them shares
Each of these is built to survive a bad Tuesday. When something breaks — and things break — a separate fault path catches it, tells me precisely what failed, and waits. That's the difference between a demo and a system. A demo dazzles in a meeting. A system holds at 7am when an outside service changes without warning.
The clever model is never the system — it's one small part of it. The system is the error handling, the monitoring, the human checkpoint, the spending cap, the alerts. If someone is selling you an AI "agent" without those, they're selling you a demo.
04The honest math
Around twenty of these run in production today. Between them: two back-office hires I never had to make, several capabilities I'd always told myself we couldn't afford, and a real slice of a working week handed back to me and my partners — at the running cost of a modest monthly bill.
You'll notice I'm not quoting you a tidy savings figure. That's deliberate. The math is obviously positive — but any precise number I invented here would be exactly that: invented, the way most AI case studies invent theirs. If you want a real number for your business, that's precisely what an audit produces. I won't pretend I already have yours.
05Two places I deliberately keep AI out
Knowing where not to use AI matters as much as knowing where to. Two parts of the business I could automate — and choose not to.
Contracts
A model can read a contract and flag risky clauses — and it's right about ninety percent of the time. On a contract, ninety percent is worse than seventy: at seventy a lawyer reads every line, at ninety they trust it and skim. So contracts stay with a human who owns the outcome. AI drafts; a person decides.
The judgment calls on my own desk
Most of what reaches a CEO is a one-off judgment about people and context. The higher the work sits, the less of it should be automated. AI is brilliant at the repetitive layer below — and that's exactly where the return is.
The takeaway: AI is a better gift to the layer below you than to you. Don't start with yourself. Start with the part of your team you wish you had more of.
06Three things this means if you own the company
1. Your biggest cost isn't the salary line. It's the work no salary covers.
The "where's that file." The deadline missed because someone was out. The deal that died because a follow-up was sloppy. Your accountant has never had a way to put a number on that — and it's exactly where these systems hit hardest.
2. Don't start with the org chart. Start with the calendar.
Look at where a week of your team's time actually goes. Find the recurring blocks of dull, repeating work. That's the map. Every system I built began with someone — often me — saying "I am so tired of doing this."
3. The technology is the easy part. Your real edge is judgement.
Anyone can buy the tools now. The hard part is knowing which thirty percent of your business should be automated and which seventy percent absolutely should not. I haven't run your business — but I've run one, and that judgement carries across industries. The shared back office I've already proven on myself; for whatever is specific to you, I bring the method, not guesswork.
07If you want help
I built AKINT.AI because other owners started asking how I'd done this. I'm not selling you something you couldn't, in theory, build yourself. I'm selling you the eighteen months of mistakes you get to skip — and the instinct for which thing to build first.
If any of this landed, the next step is small. Take the readiness scorecard — ninety seconds, no email required — and you'll leave with three concrete places to start. Or book a thirty-minute call, and I'll come with an opinion or two about your operation already formed.
Either way: don't post the next back-office job before you've had this conversation. It might be the most expensive hire you never needed to make.