We run what we recommend.
Every system on this page was designed and built by Steelworth, in our own business.
Most consultancies sell a pattern they have never had to live with. We take the opposite position: before we recommend an approach, we run it ourselves, in our own operations, where we absorb the cost of every design mistake.
The two systems below are first-party. Our own operation, our own platform. No composite case studies, no borrowed logos. What you see is how we actually work.
Our rental operation runs day to day on AI agents.
Claude-based agents inside Slack handle guest operations, bookkeeping, and month-end reporting for our own multi-market short-term-rental business, with an audit trail behind every action.
A short-term-rental operation across multiple markets generates a constant stream of small, time-sensitive work: guest messages at odd hours, maintenance incidents, a ledger that grows with every stay. Handled manually, that work either consumes the operator or gets done late and inconsistently.
We standardized the work first. Guest-operations playbooks, an incident escalation path, a chart of accounts with explicit categorization rules. Then we put Claude-based agents on top, working inside Slack, where the operation already lives.
Agents triage guest messages and incidents, handling the routine cases and escalating the ones that need a human decision. They categorize transactions against the chart of accounts as they land and assemble month-end reporting from the same ledger. People review the exceptions; agents work the volume.
Every agent action is logged: what it read, what it decided, and why. When a categorization is questioned or an incident is reviewed, the trail is already there. That audit discipline is not an add-on. It is what makes it safe to let agents work.
Codify the rules, put agents on the routine volume inside the tools your team already uses, keep people on the judgment calls, and log everything.
What runs here
- Guest-ops triage in Slack
- Incident escalation handling
- Automated ledger categorization
- Month-end reporting
- Audit trails on every action
#operations
Agents and people, one channel
Guest Ops2:41 PM
The guest in unit 12 says the door code isn't working, and check-in is at 3:00.
Steelworth Ops AgentApp2:41 PM
Reissued a fresh code and sent it to the guest. The lock battery is running low, so I opened a maintenance task for tomorrow morning and logged the incident.
Code reissued · guest notified · maintenance scheduled · trail logged
Finance9:04 AM
How did the month close?
Steelworth Ops AgentApp9:05 AM
Closed. Every transaction is categorized against the chart of accounts and the month-end pack is posted in #finance. Two refunds are flagged for a human decision in this thread.
An enterprise platform with agents designed into the architecture.
A small senior team built a full enterprise property-management platform with agentic orchestration and secure-by-design AI surfaces, our working proof of AI-native delivery.
Property management is dense with coordinated workflows: reservations, owner accounting, maintenance, compliance. Most software in the category treats AI as a chat box on the side. We wanted to know what a platform looks like when agents are part of the architecture from the first design review.
So we built one ourselves: a full enterprise property-management platform with an agentic orchestration layer that coordinates work across modules rather than answering questions beside them. A small senior team carried it from architecture to working software, with AI in the delivery process as well as the product.
Every AI surface was secured by design. Agents operate with scoped permissions, untrusted input is isolated from privileged actions, and nothing an agent does escapes the event log. Security review was a build gate, not a launch-week scramble.
The build is the credential. It is how we know what AI-native delivery actually requires, where agentic architectures break, and which controls have to exist before an agent touches production data. That knowledge is what we bring into an engagement.
When agents are first-class components of the architecture and every AI surface is secured by design, a small senior team can deliver enterprise-grade software without an enterprise-sized headcount.
What runs here
- Agentic AI orchestration
- Secure-by-design AI surfaces
- Enterprise platform architecture
- AI-native delivery
- Small senior team model
Let's find out what your operation is actually running on.
Bring us the process you're trying to fix. We'll tell you honestly whether it's ready for automation or still needs to be standardized first.