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.
Oversight
Agent Exception Handler
When an AI agent hits something it cannot resolve, this catches it and routes it to the right person immediately.
Most AI runs finish fine, but the few that fail tend to vanish into a log nobody reads until a customer complains. Agent Exception Handler watches your agent's runs for failures, timeouts and cases outside its rules, and sends only those to the right person with the context needed to fix them. Clean runs never touch your team's queue.
- Failed and stuck AI runs caught as they happen
- Each exception routed to the right person with full context
- Unresolved exceptions stay visible until someone closes them
$2,900
20–30 working days
Optional maintenance $249/mo
What it is
Agent Exception Handler watches your AI agent's runs for technical failures: errors, timeouts and unusual runs outside its rules. It routes only that exception to a person, with the context needed to fix it. Uncertain or sensitive content belongs with AI Escalation System instead. Routine runs that complete cleanly never touch a human.
Why businesses buy it
Silent failures are the expensive kind. An order that never synced, a reply that never sent or a record left half updated can sit unnoticed for a long time, and by then the fix is harder and the customer is already upset. Routing only the exceptions keeps your team focused on real problems instead of reading every run. The written scope sets out the workflow, the failure conditions and up to three connected systems before you order, so you know exactly what is watched and who hears about it.
Problems this solves
Failed AI runs go unnoticed until someone spots missing work
The handler watches each run for the moment it fails, times out or hits a case outside its rules. Only that exception goes to a person, so failures stop disappearing quietly.
Staff check every run by hand just to find the rare failure
Runs that complete cleanly never reach a human at all. Your team can stop watching routine work and deal only with the exceptions the workflow brings to them.
Fixing a failure drags on because the context is missing
Each exception arrives with the context needed to fix it, pulled from up to three connected systems. The person can see what the agent was doing and where it stopped.
A good fit if
Your AI agent already handles most cases fine, but the failures currently vanish into logs nobody checks until a customer complains.
You can name the specific failure modes that matter: a timeout, a low-confidence answer, a missing record, an API that returned an error.
You want only the genuine exceptions reaching a person, not every run copied into a review queue regardless of outcome.
Someone on your team can take ownership of exceptions within an agreed response time once they are routed.
Not the right fit if
You want every output reviewed, not just the ones that fail. See Agent Review Queue, which holds the full output stream, not only exceptions.
You want a rubric that scores completed work for quality, not a router for runs that broke. See Agent Quality Control.
The exceptions you are routing are specifically sales conversations awaiting a rep's sign-off. See Human-in-the-Loop Sales Agent, which is built around that one workflow.
You need this for supervising agents live, pausing or correcting them mid-action rather than routing what already failed. See the already-published AI Supervisor Agent.
Your concern is uncertain or sensitive content, such as a low-confidence answer or a complaint, not a broken run. See AI Escalation System.
You need support tickets escalated to the next tier with full context. See AI Escalation Agent, built for helpdesk teams.
How it works
- The AI agent runs its workflow as usual, for example pulling an order status, drafting a reply, or updating a record, and most runs complete without any exception firing.
- Failure conditions you define trigger the handler: an API call that errors, a confidence score below your threshold, a missing field, or a case that matches none of the agent's known patterns.
- The exact run that failed is captured, not summarized, so the person picking it up sees the same inputs, the same attempted output, and the point where it broke.
- The exception routes to the person or team you name for that failure type, through Slack, Microsoft Teams, email, your helpdesk or your CRM, whichever you already use.
- A response-time target runs on every open exception, so a failure that sits too long shows up rather than getting forgotten.
- Resolved exceptions are logged with what fixed them, building a record you can use to reduce that failure type in the agent itself over time.
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.
- The systems your agent's workflow touches, with admin or API access to each.
- A written description or screen recording of what should happen when the agent cannot complete a run today.
- Five to ten real examples of failed or unusual runs, so the handler is built against genuine cases.
- Who should receive each type of exception, and how quickly they need to respond.
- 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 failure definitions 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 the failure conditions agreed with you.
Weeks 1–2
Systems connected and the exception triggers built.
Week 3
Routing rules configured for each failure type.
Weeks 3–4
Twenty test runs on your sample data.
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 | $2,900 |
| Express | 15–25 working days | $3,400 |
| Rush | 10–20 working days | $3,900 |
What you receive at handover
- Exception monitoring running in the background on your agent's runs, across up to three systems
- Routing rules, prompts and guardrails set up in one workflow of up to twelve steps
- A test report from twenty runs on your sample data, covering the agreed scenarios
- An admin guide and the recorded handoff session
- 30 days of post-launch support
Expected benefits and how to measure them
- Failures can be fixed sooner, because a person hears about them straight away.
- Helps free staff from checking runs that went fine.
- Makes repeat failure types easier to spot and fix at the source.
- Time from failure to fix
- Compare how long failed runs took to resolve before launch with exception timestamps and fix dates after.
- Exceptions by type
- Count routed exceptions each week, grouped by cause, to see which failures keep coming back.
- Manual monitoring hours
- Ask staff to log time spent checking runs for two weeks before and two weeks after launch.
Before you start: Note how failures are found today and how long fixes take. List the failure types you have seen in recent months.
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 Agent Exception Handler afterwards costs roughly $20 to $100 a month in your own accounts, covering the automation platform and the AI model usage spent checking for failure conditions. 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 failure conditions 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 a monthly allowance of change hours, which covers adding a new failure condition or adjusting where an exception routes. It starts after the 30 days of support and can be cancelled any month.
$249/mo (the minimum fee, 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 workflow tool's error handling is a generic catch-all, not rules you define | An agency quotes exception-handling work after discovery | Ours is one fixed price, shown on this page before you buy |
| Scope | A generic error log leaves you to define what counts as a failure | An agency agrees scope during the project | Ours ships with your failure conditions already built in |
| Delivery | Off-the-shelf error handling still needs real rules before it is useful | An agency build usually takes several weeks, set per project | Ours has a fixed delivery time shown before you buy, with faster options |
| Noise | A shared error log mixes real failures with routine warnings | An agency deliverable depends on what was agreed | Ours routes only the exceptions you defined as worth a person's time |
Questions about this product
How much does Agent Exception Handler cost?
The build is one fixed price, listed with each delivery speed in the price and delivery panel on this page. It covers everything on the Scope Card: one workflow with the failure conditions you agree, routing rules for who handles each exception type, and up to three connected systems. The automation platform and AI usage that check each run are billed to your own accounts.
How is Agent Exception Handler different from Agent Review Queue?
Agent Exception Handler only routes runs that actually fail or hit a case the agent cannot resolve; clean runs never reach a person. Agent Review Queue holds every output for review regardless of outcome. Choose this build when most runs are fine and only the failures need attention, not the full output stream.
What does Agent Exception Handler cost to run each month?
Running it afterwards costs roughly $20 to $100 a month in your own accounts, covering the automation platform and the AI usage spent checking each run against your failure conditions. There is no separate license fee on top of that figure. The exact cost depends on how many workflows and runs you connect.
What access does Agent Exception Handler need to my systems?
We need admin or API access to each system your agent's workflow touches, shared by adding our team as a user or creating a key you control, never by email. A written description of what should happen today when a run fails, plus five to ten real examples of failed or unusual runs, lets us build triggers that reflect genuine cases.
Can we add new failure conditions later?
Yes. Adding a new failure condition, or changing who a given exception type routes to, usually fits inside the monthly change hours that come with maintenance. Adding a whole new workflow beyond the one in scope, or a fourth connected system beyond the three included, goes through a Change Quote agreed before that work starts.
How is Agent Exception Handler different from AI Escalation System?
They catch different problems. Agent Exception Handler deals with technical failures, such as errors, timeouts and runs that stop halfway. AI Escalation System deals with content that is uncertain or sensitive, such as a low-confidence answer or an angry customer. For support tickets that need the next tier, see AI Escalation Agent in Customer Support.
What happens to an exception nobody resolves in time?
It stays open and visible against its response-time target rather than disappearing, and you can set a second-level routing rule, such as escalating to a manager, for anything left unresolved too long. Nothing is auto-closed or silently dropped. That visibility is the entire point of routing exceptions instead of leaving them in a log.
Guides on this topic
All guides Stop letting failed AI runs disappear into a log nobody checks.
Order Agent Exception Handler today and it goes live on a fixed delivery date. The scope above is exactly what you get, in writing, before you pay.
Works well with
Oversight
Agent Quality Control
Samples and scores output from any of your AI agents against one quality standard.