Level 2 credential · 8 clock hours
AI Agent BuilderDesign it. Ground it. Launch it.
Learners learn how voice agents work, then build one for their own business: grounded in the business's facts, connected to a real tool, broken twenty ways on purpose, and put live on a phone number.
Hands-on training using Vapi, with the agent's tools built in Claude Code
The credential at a glance
- Credential
- AI Agent Builder Certificate
- Type
- Certificate
- Clock hours
- 8
- Format
- Four 2-hour build sessions, one a week, in person
- Price
- $1,000 a seat, invoiced after the last session
- Class size
- 8 to 15 people
- Assessment
- Rubric on the work itself, a live demo and a 20-question quiz
- Issued by
- Paglow Automations (Ohio AI Solutions LLC), Lima, Ohio
- Before the first session
- The AI Automation Builder Certificate, or the same skills shown in a short build check.
- A Vapi account opened by the employer, with a payment method on file.
- A phone line the business can forward to the agent, or approval to add a new number.
- A Google account the business owns, for the agent's sheet and its small backend.
- The business's own facts: services, hours, service area, prices it quotes, jobs it turns down.
What a holder can do
TechCred pays for applied skills in developing, building, configuring or operating technology. Every competency below ends in something the learner built or ran, and an assessor checks the evidence.
Build
Make the thing that does the work.
Design an agent's scope: the calls it takes, its call flows, what it refuses, when it hands off, and how it discloses that it is AI.
Evidence: Agent design document
Build a knowledge base from the business's own facts and keep the agent's answers inside it.
Evidence: Knowledge base and FAQ call log
Connect the agent to real tools — a transfer to a person, a message to the owner by text or email, and an API tool that does a real task — with the tool's small backend built in Claude Code.
Evidence: A real task completed end to end on a test call
Configure
Set the rules it runs under.
Configure a voice assistant's model, voice, greeting and a prompt written for speech, not for reading.
Evidence: Assistant configuration
Configure AI and recording disclosure and data handling, and put the agent live on a phone number.
Evidence: Live line and disclosure script
Write the business's service profile — trades, service range, jobs it won't take — that the agent and referral networks use.
Evidence: Service profile
Operate
Run it, test it, keep it running.
Test the agent against 20 scripted scenarios, including emergencies, prompt injection, abuse and out-of-area jobs, and fix every failure.
Evidence: Scenario test report
Run the agent: a weekly call review, cost per call, and a change log.
Evidence: Runbook and first call review
Analyze
Measure the work before and after.
Explain how a voice agent works end to end — phone line, speech-to-text, language model, tools, text-to-speech — and where delay, cost and mistakes come from.
Evidence: Design document and quiz
Two hours each. Something built every time.
One session a week, in person. Each one ends with an artifact the learner keeps and the assessor checks.
Session 1 · 2 hours
How voice agents work, then design one
A designed agent and a first assistant answering test calls.
- The theory: phone line, speech-to-text, model, tools, speech, and why delay matters
- The agent's scope, call flows and what it will never do
- When it hands off to a person, and how it says it is AI
- A first assistant in Vapi answering test calls
Proof: Design document and a working first assistant
Session 2 · 2 hours
Ground it
An agent that answers from the business's own facts.
- A knowledge base of the business's services, hours, area and policies
- A prompt written for the ear: one question at a time, names and numbers read back
- Ten FAQ test calls, scored and fixed
Proof: Knowledge base and FAQ call log
Session 3 · 2 hours
Connect it
The agent completes a real task on a call.
- A transfer to a person, with the reason passed along
- A take-a-message note to the owner, by text or email
- One tool that does real work — log the lead or offer a time — whose small backend is built in Claude Code
Proof: A real task completed end to end
Session 4 · 2 hours
Break it, launch it
Tested hard, disclosed properly and live on a number.
- Twenty scripted scenarios: angry caller, emergency, prompt injection, out-of-area job
- AI and recording disclosure, and what is kept and for how long
- Live on a number: an after-hours forward or a new line
- The service profile, a weekly call-review routine and cost per call
- A five-minute demo and the quiz
Proof: Test report, live line, runbook, profile and quiz
How it's earned
The credential is earned on the work, not on attendance alone. An assessor scores the learner's own artifacts against six criteria.
Proficient or better on all six rubric criteria, 80% or more on the 20-question quiz, and all 8 hours attended.
Criterion 1
Design and scope
Proficient means
The design names the calls it takes, the ones it refuses, every handoff point and the disclosure line.
Criterion 2
Grounding
Proficient means
Answers come from the knowledge base; unknowns get an honest handoff, never an invented answer; 10 FAQ calls pass.
Criterion 3
Tools and handoff
Proficient means
Transfer, a message to the owner and one working API tool complete on a test call, with confirmation before any action.
Criterion 4
Testing
Proficient means
Twenty scenarios are logged, including emergency and prompt injection, and every failure has a recorded fix.
Criterion 5
Safety and disclosure
Proficient means
The agent says it is AI, gives the recording notice, sends emergencies to 911 and keeps to the data rules.
Criterion 6
Operation and results
Proficient means
Live on a number, with a runbook, a first weekly call review and a cost-per-call figure.
- Not yet
- Developing
- Proficient
- Exemplary
Every criterion has to reach Proficient.
A missed session is made up from the recording plus a 30-minute build check before the next session. Each certificate carries a verification code anyone can check at paglow.com/training/verify.
Your business owns the accounts.
They're opened by your business, billed by the software companies, and never part of the training invoice. Owning them is what keeps the work running after class.
Vapi
The voice-agent platform the learner builds in, plus the agent's phone number.
Pay as you go: Vapi's platform fee plus speech, model and phone minutes, billed by Vapi.
Usage and phone numbers are software costs, not training, so they are billed to you directly and never appear on the training invoice. A healthcare office also needs Vapi's paid HIPAA mode and a signed agreement, because callers mention their health.
Google account
Holds the agent's Google Sheet and the small Apps Script backend its tool calls.
Free, or part of a Google Workspace plan you already have.
Owned by the business, so the agent's records stay yours.
Claude (Pro or Team plan)
The same seat from Level 1, used to build the agent's tool.
Pro is $20 a month; Team is $25 a seat a month ($20 billed yearly). Billed by Anthropic.
Already in place from Level 1.
Paglow isn't affiliated with Anthropic or Vapi. Claude and Claude Code are Anthropic's products; Vapi is Vapi's.
What never goes into an AI tool
In session 1 every learner sorts their data into green, yellow and red. Red never goes into an AI tool. These are the red lines each industry starts from.
No patient information goes into any AI tool in class, ever.
Manufacturing
Red
- Export-controlled drawings or data (ITAR, EAR)
- Employee personal and medical records
- Customer data a contract keeps confidential
Healthcare administration
Red
- Anything that identifies a patient or their care (all PHI), including a name tied to a visit request
- Insurance member IDs and claims
- Staff licenses and personnel files
Trades and home services
Red
- Payment card numbers
- Gate codes and alarm codes
- Employee personal records
Professional offices
Red
- Social Security, bank and account numbers
- Privileged or confidential client material, unless office policy and a business-plan account allow it
- Tax returns and financial statements outside an approved account
What phone agents do, by industry
The same skills in every class. These are the kinds of calls an agent built in class takes.
Manufacturing
Plants, machine shops and fabricators: estimating, purchasing, maintenance and the front office.
- After-hours order-status line that reads from a daily export
- Driver and delivery check-in line with dock directions
- Supplier callback line that takes the message and texts purchasing
Healthcare administration
Clinics, dental, therapy and home-health offices — the front desk and back office, never clinical work. Anything that touches a patient needs a HIPAA-covered setup.
- Hours, directions, parking and new-patient forms, on Vapi's HIPAA mode with a signed agreement
- Sending the new-patient packet link by text or email on request
- After-hours line that gives the on-call instructions and 911 guidance word for word
Trades and home services
HVAC, plumbing, electrical, roofing, cleaning and contractors — the office behind the trucks.
- After-hours triage: emergency to the on-call tech, routine to a morning callback
- New-job intake that checks the address against the service area
- Estimate scheduling into open slots the office controls
Professional offices
Law, accounting, insurance and real estate offices — intake, documents and follow-up.
- New-client intake line that collects the basics and books a consultation
- Hours, drop-off and document questions answered from the office's own facts
- After-hours message line that texts the right person
Plan a class with this credential
Classes start in 2027. Tell me about your team and I'll reply myself with dates, a plan and the TechCred steps.