We built the duty manager
we couldn’t hire.

Three years ago, Dennis's phone rang at 2am. A guest checking in late was reporting human waste on the wall. Not a lost code, not a broken heater, but human waste! The guests had arrived in the afternoon and missed it, so the report came in the middle of the night. He drove over, took one look, and there was nothing to be done except organize them somewhere else to sleep, and then he lay in the dark running forensics on how a room ends up like that. What it turned out to be is a story Dennis tells better in person.

What it cost him is the part we remember: another night gone, and a day spent grumpy on the phone, asking everyone why. Between us we run short-term rentals, a couple of gyms and a few spas, and each of us has a version of that call. We always knew this was the reality of a small operation. We never made peace with it. The pain builds slowly, and it takes apart your time first, then your happiness.

We have spent our careers on the other side of that call. Viet spent 14 years at Microsoft carrying AI models out of research and into industrial adoption. Huy built AI for small businesses, the platform behind 200,000 support chatbots, and watched software get very good at answering without ever getting good at fixing. Dennis leads customer experience at Microsoft scale, inside the world where the middle of the night has a staffed desk: awake, trained on the standard, allowed to act, writing everything down. But the operator with ten listings has only a phone on the pillow.

Here is the belief this company is built on: the agentic era changes who can afford judgment, and nobody should benefit from it sooner than small business. We also know exactly why nobody has built this yet, because the reasons are the job:

• A 2am decision has no test suite. No build server says the outcome was right, so the system has to prove itself another way.
• There is no standard workflow to encode. Enterprises standardize; every operator runs on their own runbook, and the software has to learn yours rather than impose one.
• The stakes are real. The work touches money, guests and promises, where a wrong action cannot be quietly rolled back.

We write down what we learn as we go, if you want the technical version. The short version is that an AI harness for small business, operating in the open world, takes two designs done excellently:

• The AI system itself: smart enough to adapt to a messy, real operation instead of a standardized one.
• The product: a UX a busy owner can actually run their business through, without becoming a systems administrator.

Dewly is our answer, and its design follows straight from those lists. Your runbook becomes playbooks it works inside, written in daylight, in your words. Anything touching money or a promise waits for you, so a wrong draft is a correction rather than damage. And instead of asking for trust, it starts in shadow mode: a first week of drafting everything and deciding nothing while you mark its homework, which is the closest thing to a test suite a real operation can offer. We run it on our own listings first.

We are early, and we would rather say so than pretend otherwise. The demo call is with one of us, the roadmap still has room for what you tell us, and we will not put anything into your operation that we would not run in our own. Book a demo and bring us your worst night, or grab The Five-Star Playbook and stay in touch. We would like to meet you.