AI Agents and Assistants
What is an AI agent? A plain-English guide for business owners
What an AI agent is, how it works, where it helps a small business, the risks, and a 7-point check to see if you are ready for one.
Supervised Agents
Human-in-the-Loop Support Agent
Drafts every support reply; an agent approves or edits before it reaches the customer.
Support agents spend much of the day typing answers they have written many times before. This agent reads each incoming question and drafts the reply from your approved knowledge base. The agent you name approves it, edits it or writes their own before anything reaches a customer. When the answer is not in your material, the draft says so rather than guessing.
- Support replies drafted from your approved knowledge base
- A person approves every reply before a customer sees it
- Knowledge gaps flagged instead of filled with guesses
$5,400
20–30 working days
Optional maintenance $270/mo
What it is
Human-in-the-Loop Support Agent reads incoming support questions and drafts the reply from your approved knowledge base, but nothing sends to a customer until the agent you name approves it, edits it, or writes their own. It keeps your team in control of every reply while removing most of the typing.
Why businesses buy it
Customers expect quick, accurate answers, and support teams get stretched thin trying to provide them. Fully automatic replies feel risky when one wrong policy answer can cost a customer. This agent sits in between: the AI does the drafting, your team checks and sends, and replies go out faster with the same care. The written scope sets the channel, up to two connected systems and up to six workflows with your approval rules before you order, so you know exactly what is covered.
Problems this solves
Support staff type the same answers to the same questions all day
The agent reads each incoming question and drafts a reply from your approved knowledge base. Your team member approves, edits or writes their own, so the typing drops while the final word stays human.
You want quicker replies but cannot let AI answer customers alone
Nothing sends to a customer until the person you name approves it. Up to six workflows hold your approval rules, so sensitive topics can follow a stricter path than routine ones.
Answers vary depending on who picks up the ticket
Drafts come from one approved source of up to a hundred documents. Every team member starts from the same answer, which helps replies stay consistent across shifts and channels.
New support staff take a long time to learn every policy
Draft replies show newer team members how your policies apply to real questions. They review and send, learning as they go, while fifty test scenarios confirm the drafts follow policy.
A good fit if
You want AI drafting support replies, but your team is not ready to let answers go out to customers unreviewed.
Your support agents already answer from a knowledge base by hand and want a faster first draft, not full automation.
You can name the agent or team who will approve, edit or reject drafts within an agreed response time.
You want a record of what the AI proposed versus what an agent actually sent, useful for training and for catching drift in the knowledge base.
Not the right fit if
You want the agent resolving support questions and sending replies on its own, without a person reviewing each one first. See the already-published AI Customer Support Agent, which resolves directly from your knowledge base.
The approval step you need is really about sales outreach, not support replies. See Human-in-the-Loop Sales Agent, built around that workflow instead.
You need the queue and assignment logic for drafts waiting on an agent, not the support agent producing them. See Agent Review Queue.
Your support questions are specifically about travel bookings, fares or documents. See AI Travel Support Agent, built for that domain.
How it works
- A support question arrives through your chosen channel, and the agent checks it against your approved knowledge base before drafting anything.
- A reply is drafted using only the FAQs and documents you approved; if the answer is not there, the draft says so rather than guessing.
- The draft, with the customer's question and the source it was drawn from, lands with the agent you named, inside the helpdesk or channel they already use.
- The agent approves it as written, edits it, or rejects it and answers directly; nothing reaches the customer until one of those three happens.
- Approved replies send from your own helpdesk or channel, so the customer sees a reply from your team, on your system, not a third-party address.
- Every draft, edit and outcome is logged against the ticket, building a record you can use to spot gaps in the knowledge base or coach the team.
What you provide within 72 hours
Your portal lists these the moment you order. Delivery starts the working day after everything is in and your access is checked.
- Your business information (company basics, hours and brand voice), shared during intake.
- Admin or API access to your helpdesk and knowledge base, sandbox or test records if available.
- Your support playbook: refund policy, escalation limits, and the tone your team uses.
- Ten real sample conversations or emails so drafts are built against genuine cases, not guesses.
- The agent or agents who will approve drafts, and how quickly they can turn them around.
- Admin access and API keys for every connected system, shared by adding our team as a user or creating a key you control, never by email.
- One decision-maker who can approve the knowledge base and rules before launch.
- An API key for your own AI model account (for example OpenAI or Google Gemini), created for this build with a spending limit.
The 72-hour rule. Send everything on your checklist, including access, within 72 hours of ordering so your build starts on time.
Timeline
Start
Intake verified and your support playbook agreed with you.
Weeks 1–2
Helpdesk connected and drafting logic built from your knowledge base.
Week 3
Approval routing configured for the agent or agents you named.
Weeks 3–4
Fifty documented test scenarios run.
Week 4
Your review and two rounds of changes.
Go-live
Live, recorded handoff session, support starts.
| Option | Delivery time | Price |
|---|---|---|
| Standard | 20–30 working days | $5,400 |
| Express | 15–25 working days | $6,200 |
| Rush | 10–20 working days | $7,000 |
What you receive at handover
- The support agent working in one channel, connected to up to two systems with read and write access
- Your knowledge base loaded from up to a hundred approved documents
- Up to six workflows carrying your approval rules, prompts and guardrails
- A test report covering fifty documented scenarios
- An admin guide, the recorded handoff session and 30 days of post-launch support
Expected benefits and how to measure them
- Replies can go out sooner because most of the drafting is already done.
- Helps keep answers in line with your approved policies.
- Keeps your team in control of every word a customer sees.
- First response time
- Compare first-reply times in your helpdesk for the four weeks before launch with the four weeks after.
- Drafts sent unchanged
- Track how often staff send drafts as written versus edited or rewritten, from helpdesk or approval records.
- Tickets closed per person
- Compare weekly tickets closed per team member from helpdesk reports, before and after launch.
Before you start: Export first-response time and tickets per person from your helpdesk for the last month. List the questions you answer most often.
These are things to track, not promised results. Outcomes depend on your data, your team and your market.
Running costs
The build price is one-time. Running Human-in-the-Loop Support Agent afterwards costs roughly $50 to $250 a month in your own accounts for AI model usage and hosting, on top of whatever helpdesk plan you already pay for. AI usage during the build and testing is billed the same way, directly to your accounts, so every provider relationship stays yours.
Ask about running costsCustom requirements
If your support rules go past the Scope Card, add a note at checkout. You get a written Change Quote, and that work only starts once you accept it. Your original order runs on its normal timeline either way. Larger or bespoke projects go through the Custom AI request form instead.
Request a custom versionKeep it running
We recommend monthly maintenance: monitoring, fixes and monthly change hours, which cover new FAQs, an updated refund policy or a tone adjustment. Maintenance starts after the 30 days of support and can be cancelled any month.
$270/mo (5% of the build price a month, includes 2 change hours)
Optional. Starts after the post-launch support period. Cancel any month.
How maintenance worksHow this compares
| Do it yourself with a SaaS tool | Hire a custom agency | Upmark Labs | |
|---|---|---|---|
| Price | A helpdesk AI add-on charges a recurring per-agent fee | A custom support-agent build is quoted after discovery | Ours is one fixed price, shown on this page before you buy |
| Scope | A generic AI drafts from whatever it can find, not just your approved sources | An agency agrees scope during the project | Ours only drafts from the FAQs and documents you approved |
| Delivery | Helpdesk AI add-ons still need real knowledge-base setup before drafts are usable | An agency build usually takes several weeks, set per project | Ours has a fixed delivery time shown before you buy, with faster options |
| Control | Some helpdesk AI tools auto-send low-risk replies | An agency deliverable depends on what was agreed | Ours never sends until your agent approves the draft |
Questions about this product
How much does Human-in-the-Loop Support Agent cost?
The build is one fixed price, and the panel on this page lists it with each delivery speed. It covers everything on the Scope Card: one support channel, up to two systems such as your helpdesk and knowledge base with read and write access, and up to six workflows with your approval rules. AI usage and hosting afterwards are billed to your own accounts.
How is this different from a fully automatic AI Customer Support Agent?
The already-published AI Customer Support Agent resolves questions and sends replies directly from your knowledge base, without a person reviewing each one first. This build drafts the same kind of reply but never sends on its own; the agent you name approves, edits or rejects it first. Choose this when your team is not ready to hand sending over entirely.
What does Human-in-the-Loop Support Agent cost to run each month?
Running it afterwards costs roughly $50 to $250 a month in your own accounts for AI model usage and hosting, on top of whatever helpdesk plan you already pay for. There is no separate platform license on top of that figure. The exact cost depends on ticket volume and how many drafts the agent produces.
What access does Human-in-the-Loop Support Agent need to my systems?
We need admin or API access to your helpdesk and knowledge base, sandbox or test records if available, shared by adding our team as a user or creating a key you control, never by email. Your written support playbook and ten real sample conversations or emails let us build drafting rules that reflect how your team actually answers customers.
Can we update the knowledge base or approval rules later?
Yes. Adding a new FAQ, updating a policy, or adjusting who approves a given ticket type usually fits inside the monthly change hours of maintenance. Adding a genuinely new workflow beyond the six in scope, or a third connected system beyond the two included, goes through a Change Quote agreed before that work starts.
Can we cancel Human-in-the-Loop Support Agent maintenance later?
Yes. Maintenance has no lock-in, so you can cancel any month once the 30 days of included support has ended. Unused hours do not roll over. Bigger changes to your knowledge base go through a Change Quote, priced before work starts.
What happens when the answer is not in our knowledge base?
The draft says so rather than guessing, and flags the ticket for the agent to answer directly or escalate. It never invents a policy or a fact that was not in the FAQs and documents you approved. Gaps like this are useful signals for what to add to the knowledge base during monthly maintenance.
Guides on this topic
All guides Let AI draft the reply; keep your agent on the send button.
Order Human-in-the-Loop Support Agent and every reply reaches an agent for approval before a customer ever sees it, live on a fixed delivery date.
Works well with
Supervised Agents
Human-in-the-Loop Sales Agent
Drafts sales actions that a rep approves before anything is sent.