

Case study · AI-native operations on NYO
One operator, an agentic back office
Valley O'Ville is a private resort in Calatagan, Batangas, hosting one group of up to thirty at a time. This is not about the resort. It is about how its operator runs it: the back office - front desk, bookkeeping, social media - is agentic. One agent, VOV Agent, does the work and the operator directs it, with every model call going through NYO - in Telegram for the operator, on Messenger and the site for guests, on Facebook for the world, with the request log to debug it all.
- Operator
- Telegram → Hermes agent → preview → yes
- Guests
- Messenger and the site chatbot, answered from the resort's records
- Output
- Facebook posts and captions, sent on CONFIRM
- Model
- glm-5.3-flash through NYO, one key, one request log
In short
An agentic back office is one where an AI agent does the inquiries, bookings, balances and social media, working from the business's own records, and a person approves what it proposes. Valley O'Ville runs its operations this way - AI-native, with one operator and VOV Agent, whose model calls go through NYO for a few dollars a month.
AI-native operations
The back office is still there. It is agentic.
A place that hosts thirty guests a stay used to need a small staff behind the scenes: someone answering inquiries at midnight, someone reconciling deposits and payouts, someone chasing photos and posting them. At Valley O'Ville that work still happens, every day - an agent does it, the operator directs it, and NYO sits between the agent and the models.
That is an agentic back office. The operator talks to VOV Agent in Telegram, in plain language, about bookings, balances, rates and guests. The agent reads the resort's own records and drafts the answer, the reply or the ledger change, and the operator approves it with a word. One person carries the desk work of three, and spends the time saved on the guests.
And that is what AI-native operations means: a business whose operations were designed around an agent from the start, not an agent bolted onto old ones - small, hands-on and future-forward, with the model behind it as replaceable as a supplier. So when a guest-facing chatbot made sense, it was not bought off the shelf: it is the same agent, turned to face guests.

The agentic back office, job by job
Three back-office roles the agent carries, and what it deliberately leaves to a person.
The front desk
Inquiries arrive on Facebook and Airbnb at all hours. The agent quotes the right rate for the dates and group size from the resort's saved rate card, checks the calendar, and drafts the reply. The operator reads it and sends it.
The bookkeeper
Deposits, balances and Airbnb payouts live in one ledger the agent reads and updates. "Who still owes for the 14th?" is a question, not a spreadsheet session. Every change is previewed and saved only after the operator says yes.
The social media hand
After checkout the agent reminds the team to collect photos and a testimonial, turns a guest's handwritten page into a graphic, and drafts the Facebook caption. Nothing posts without an explicit CONFIRM.
What stays human
Greeting the group at the gate, the grill, the pool, the morning coffee. The agent takes the desk work so the operator can be the host - which is what a 4.95 rating is made of.
How VOV Agent works
Three layers: the operator asks, the resort's own records supply the facts, and NYO supplies the model that reasons and writes. The screens above are these three steps, seen from each side.
- 01
Ask
The operator types a request in Telegram: record an Airbnb booking, quote a stay for a Facebook inquiry, list who still has a balance.
- 02
Look it up
The agent reads the live booking ledger and calendar and applies the resort's saved rates, house rules and reply guidelines, instead of answering from memory.
- 03
Act, with a yes
A model called through NYO does the reasoning and the tool calls. Replies and captions come back as drafts; ledger changes are previewed; posts wait for CONFIRM. The operator decides, the agent does the typing.
New · now live
The chatbot, turned to face guests
VOV Agent used to work only for the operator, behind the desk. Now a chatbot on valleyoville.com answers guests directly, and it is live today. It runs on the same agent, the same model and the same NYO key, so the knowledge guests read is the knowledge the operator works from.
Answers, from the resort's own records
Rates, availability, house rules and how a stay works. The chatbot answers from the resort's saved facts, the same source the operator's drafts come from, not from memory.
Bookings stay with a person
It answers questions and hands the rest to the operator: a real quote, a real calendar, a real yes. Guests get a quick answer; the operator keeps the decision.
Same agent, one key
The chatbot calls glm-5.3-flash through NYO on the same key as the rest of VOV Agent, so its usage sits in the same observability view and the same prepaid balance.
Why NYO
Easy to plug in
VOV Agent uses NYO's OpenAI-compatible chat endpoint with one API key and standard tool calling. No provider accounts, no new SDK, no infrastructure to run.
Sized for a small business
The main model is glm-5.3-flash, listed on NYO at $0.15 per million input tokens and $0.50 per million output. Prepaid credits mean the operator knows what the assistant can spend each month.
Open models that call tools
Reading records, working out rates and moving bookings through approval steps are tool calls, and NYO's open models handle them. When a better model appears, the resort switches routes without changing the agent.
Stays on the resort
NYO's guardrails hold the assistant to its purpose, and the observability page shows what it did and what it cost, which is how a lean operation trusts an agent with its books.
Is this you?
The same week, in a different business
VOV Agent is a resort's, but the shape of the problem belongs to every solopreneur and small team: customers on chat apps, records in a spreadsheet, and evenings spent on both.
The resort, rental or homestay host
Messenger pings at 11pm asking for rates you have posted three times; a calendar in your head; deposits to chase before the weekend.
The clinic, salon or studio owner
Appointment requests across three apps, reminders you send by hand, reviews you mean to ask for and never do.
The online seller or small shop
The same ten questions on Facebook and Shopee, orders in a notebook, a weekly reckoning of what came in and what went out.
The one-person agency or service
Quotes, follow-ups, invoices and a social feed that goes quiet whenever the real work is busy.
Questions operators ask
- What is an agentic back office?
- A back office whose routine work - answering inquiries, quoting, keeping bookings and the ledger, drafting posts - is done by an AI agent that reads the business's own records, while a person approves anything that commits money or speaks publicly. Valley O'Ville's operator runs the resort's front desk, bookkeeping and social media this way through VOV Agent, with every model call going through NYO.
- What are AI-native operations?
- Operations designed around an agent from the start rather than an agent added to an existing process: the records are kept where the agent can read them, approvals are a word in a chat, and the model behind the agent is a replaceable supplier. It is how a one-person business carries work that used to need a small staff.
- Can a small business run its customer service and bookings with an AI agent?
- Yes, when the agent works from the business's own records rather than from memory. Valley O'Ville, a resort hosting up to thirty guests a stay, has one operator who handles inquiries, quotes, bookings, deposits, balances and social media through VOV Agent; the agent drafts and previews, and the operator approves with a word. The pattern suits any small business whose day is Messenger inquiries, a calendar and a ledger.
- What does an AI agent for a small business cost per month?
- The model is the main variable cost. VOV Agent runs on glm-5.3-flash through NYO at $0.15 per million input tokens and $0.50 per million output, so a month of inquiries, drafts and ledger questions for a resort costs a few dollars, paid from prepaid credits with no subscription. The work of building the agent is the real investment.
- Does the AI agent reply to customers on Facebook Messenger by itself?
- It can draft every reply and send the routine ones. VOV Agent quotes rates for the dates and group size from the resort's saved rate card and drafts the Messenger reply; the operator reads it and sends it, or lets the agent answer directly the questions it already answers well. Bookings, deposits and anything with money stay with the operator.
- How does a solopreneur keep an AI agent from making mistakes with money or posts?
- By making the agent propose and a person approve. Ledger changes are previewed before they are saved, Facebook posts wait for an explicit CONFIRM, and NYO's request log records every model call with its cost, latency and any error, so a wrong answer can be traced to the exact request.
- What do I need to start an AI agent for my own operations?
- Your records in a form the agent can read (a spreadsheet or a simple database), the channels your customers use (Messenger, a website chat, Telegram for you), a model behind an API, and rules for what needs your yes. NYO supplies the model side: one OpenAI-compatible endpoint, one key, guardrails and the request log. The form at the end of this page is where to start.
Want an agent like this for your operations?
Tell us what kind of business you run, where customers reach you and which workflows eat your day. We reply with a first plan: what an agent would take over first, what stays with you, and what it would cost a month.
VOV Agent, including the chatbot on valleyoville.com, is built and run by Valley O'Ville. The photos, logo and figures are the resort's, from valleyoville.com, used here to describe the project. VOV Agent can make mistakes, so the team checks bookings and payments before relying on a detail. NYO provides the model access described here and nothing else.