The company in this story is Hulken, where Northstone built and runs the operating system described here.
People hear the numbers and ask the same question: a consumer brand doing more than $70 million, run by fewer than ten people. How?
They expect a list of AI tools. It isn't one. Hulken didn't get here by buying more software. It got here by changing how the company runs, and then putting AI on top.
The problem was never effort. It was coordination.
As the company grew, the work itself didn't get harder. Keeping everyone aligned did. Launches, inventory, retail accounts, campaigns and vendors all moved at once, and the context lived in Slack threads, spreadsheets, inboxes and people's heads. Every decision meant someone going to find the context, and every handoff was a chance for something to fall through.
Adding people relieves that for a while, then brings it back at a bigger scale. Leverage comes from a system, not a headcount.
Three layers, built in order
The mistake almost everyone makes is to start with the AI and skip the two layers underneath. Those two layers are the reason the AI works.
- 01
One source of truth
We started with one connected place for Hulken's information. Products, projects, samples, timelines, production, freight and retail accounts all reference each other, and at least twenty spreadsheets are gone. The system won't create a record it can't connect to something real.
- 02
Workflows that match how work moves
We rebuilt the routes work travels, around how it actually moves. Set one launch date and the system works backwards to build the full production calendar.
- 03
AI on top
Only then did the assistants go in. Each one owns one function, with its own instructions and its own permissions, and each lives in Slack and Notion, where the team already works.
Every leader gets a chief of staff
One function each. Loops closed.
A request written in plain language in Slack is matched to the right records, assigned to an owner, tracked to completion and confirmed back in the same thread.
A retailer emails about a delayed shipment. The request goes to the logistics chief of staff and lands with a draft reply and the tracking context attached. A person approves it, the loop closes, and the record stays in the system instead of someone's inbox.
What it runs every day
- Production and freight report in on their own.The factory's order system, the warehouse and the parcel carriers feed the same records, so purchase orders, freight and delivery dates update without anyone re-keying them.
- Revenue and partnerships in one view.Wholesale, retail and ecommerce together: partner commitments, sell-through from the retailer's vendor portal, and daily sales from three storefronts.
- Marketing work stops falling between people.Campaigns, calendars, asset tracking, product-page specs and retailer asset packages in one place.
- No number without a source.A shared assistant answers questions about status, deadlines and risk from Hulken's own records, and shows what it read. A short true answer beats a complete one that guesses.
Built so people trust it
- Nothing leaves the company without a person.No assistant sends email, moves money or commits Hulken to anything. Messages to factories, retailers and partners are drafted and wait. Anything that can't be undone waits for a yes.
- An assistant can't read what nobody gave it.Each one gets only the records and channels its function needs. Access is a permission, not a request in an instruction.
- Commercial data stays with the people who own it.Costs, margins and partner terms are visible only to those approved to see them.
- The system checks its own work.Nine of the 33 automations do no business work at all. They watch the other 24 and raise a flag in Slack the moment a feed stops reporting.
- Every action carries a name and a source.Each task records who asked and links back to the original message. 12,671 logged changes so far.
- Correct it once. It remembers.Say "from now on" in Slack and the assistant writes the rule to its own memory and reads it every time it runs.
Six tools, on Hulken's own accounts
Notion: the workspace and the assistants. Slack: where the team talks to them. n8n: the automation layer. Claude: drafting and document work. Gemini: reading inbound email. Vercel: live views of the business.
"Or changed how our business runs. It feels like he doubled our team. We are far more efficient and we get significantly more done."
What we learned
The most valuable thing we built isn't an agent. It's the operating system around the agents: the source of truth, the workflows, the ownership and the guardrails.
Decisions that used to take a week of chasing context now happen in a day, often in hours. Nobody was replaced. Everybody got leverage.
Humans for judgment. Systems for memory. AI for speed.
What it means for your company
If you're trying to do more with a lean team, the instinct is to reach for another tool. We'd suggest the opposite. Fix the process, build the source of truth, redesign the workflows, and then put AI on top. The tool was never the constraint. The operating system underneath it was.
For the framework behind this, read The 90% Problem.