Field notes on work that works.
How we build AI employees that read the documents, learn the process, and work a visible queue, with a named person approving every output.
Why we build AI employees, not copilots.
The difference is not intelligence. It is the contract. A copilot suggests and leaves you holding the queue. An employee owns the queue and submits its work for approval.
Read the post →Story · Jul 6, 2026 · 6 minWe built the wrong thing first.
Our first version was a beautiful place to do your work. Nobody needed a place. The lesson took two products to learn, and it is the reason Winsen One looks the way it does.
Read the post →The unreasonable effectiveness of HTML
Every artifact a Winsen One employee produces is an HTML page. Not markdown, not a binary blob. A format sketched at a physics lab in 1991 turns out to be the most load-bearing engineering decision in the product.
Jul 22, 2026Industry · 10 minThe scope of AI nativity in financial services.
Underwriting, fraud, and collections went AI-native years ago. The functions that check that work still run on people and spreadsheets, and every model upstream manufactures more work for them. Full nativity means staffing both sides.
Jul 21, 2026Industry · 10 minThe scope of AI nativity in manufacturing.
Lighthouse factories, predictive maintenance, vision QC: the shop floor got the budget. The batch records, deviations, and supplier documents behind it are still read by hand. The last mile of manufacturing AI is paperwork.
Jul 21, 2026Industry · 10 minThe scope of AI nativity in logistics.
Routing, visibility, and customer contact got world-class AI at the giants. Freight bills, claims, demurrage, customs entries, and certificates did not, and that is where the money leaks. Office-side nativity for carriers, 3PLs, ship managers, and ports.
Jul 21, 2026Engineering · 5 minHow an AI employee reads a 300-page document.
Not by stuffing it into a context window. Index first, then deliberate section pulls, with a page citation on every finding, so the review can actually be audited.
Jul 16, 2026Product · 5 minApproval-first is not a limitation.
Every output an AI employee produces is signed off by a named person. That is not training wheels. It is maker-checker, the control serious operators already run their own people on.
Jul 13, 2026Engineering · 5 minOne deployment per enterprise customer.
Winsen One's enterprise arrangements run as dedicated single-tenant units: a database with your name on it, on our infrastructure or yours. The managed tier is shared, and says so. Here is why we think anything else is dishonest for regulated buyers.
Jul 9, 2026Team · 4 minWhat "hire an AI employee" actually means.
Not a chatbot with a job title. A specialist with a role, a queue, a named manager, and a first day.
Jul 2, 2026Engineering · 5 minWhy we deleted our own dashboards.
A dashboard is a question you have to remember to ask. We replaced ours with the surface our AI employees already work on: a queue of items that arrive with the evidence attached.
Jun 30, 2026Product · 4 minYour SOP is a photograph of a process that moved.
Every company documents its processes with good intentions, and every wiki becomes a graveyard. Here is what an AI employee does with your SOPs instead: a living process map with a page citation on every step.
Jun 24, 2026Engineering · 9 minResponsible token-maxing.
Throwing tokens at a model gives you a demo that falls over on the second click. Here is how we let AI write production code in our own repo, with a plan, a memory, and a paper trail.
Jun 8, 2026Craft · 9 minWhat we learned building AI workflows.
Durable, multi-step work that runs without a human triggering each step is mostly an exercise in runtime, restraint, and refusing to trust your own demo. Here is what held up.
Jun 8, 2026Craft · 5 minApproval-first isn't a feature. It's the whole philosophy.
Why we'd rather ship an AI that asks too often than one that acts too freely.
May 5, 2026Team · 7 minStop hiring for the work software should do.
The roles you're about to post are the roles Winsen was built to retire. A hard look at the org chart.
Mar 26, 2026Story · 6 minWhat we will never let AI do.
An AI company drawing a line. The human moments that stay human, on purpose.
Mar 14, 2026
