Build and Sell AI Agents to Local Businesses (in PHP)

Before you start

• PHP 8.1 or newer and MySQL on your computer; you can read PHP and write a SQL query
• No API key is needed to follow the lessons: every example runs against free stand-ins on your own computer. To go live you need an Anthropic API key with a little credit, a Meta developer account and hosting with HTTPS
• Every code example was tested against simulated versions of the Claude and WhatsApp APIs, not the live services; real answers from Claude will differ, and lesson 10 shows how to rehearse them before a client sees the agent
• The Agentic AI Course is a useful companion but not required

Curriculum

Lesson 1 is free to read, no account needed. The rest unlock the moment you buy.

1. 1 What a local business will pay an AI agent to do Preview free The one job a reply agent does for a guest house, clinic, coaching class, salon or shop; why that job sells when "an AI chatbot" doesn't; what Meta allows on WhatsApp since January 2026; and what you will build, test and sell in this course. 5 min
2. 2 Choosing a niche and finding your first client Why one niche beats five, the signs of a good first client, how to find one without spamming anybody, the ten questions to ask in the first conversation, and the one fact about WhatsApp numbers every owner must hear early. 4 min
3. 3 Scoping honestly — what the agent will and won't do Write a one-page scope before writing any code. Decide what the agent answers, what it collects, when it hands over, what it never does, and how it tells customers it is an AI. Then learn to say no to the jobs that don't fit. 4 min
4. 4 Anatomy of a reply agent The whole path of one customer message, from Meta's webhook to the reply on their phone. The files of the agent you will build, the settings kept outside the web root, the webhook, and the order of decisions in run_batch(), the heart of the agent. 7 min
5. 5 Practise for free with a stand-in API Two small PHP files that pretend to be the Claude API and the WhatsApp Cloud API on your own computer, so you can build and test the whole agent with no key and no cost. Also the Claude client itself, every way a call can fail, and how to switch to a real key safely. 10 min
6. 6 The client's knowledge as a cached system prompt Turn the business's facts into a system prompt that is identical on every call, so the API can cache it. Write the knowledge so the model can't misread it, ask for a structured answer your code can act on, and keep the changing details out of the cached part. 6 min
7. 7 Guardrails — what the prompt asks, what the code makes sure of Two layers of safety. The prompt asks the model to behave; your PHP makes sure the worst mistakes can't reach a customer. Covers the price guard, masking OTPs and Aadhaar numbers, off-topic requests and prompt injection, clinics and emergencies, and what guardrails can't do. 7 min
8. 8 Hand-over and the 24-hour window How a chat passes from the agent to a person and back, what the customer is told, how the owner hears about it, and WhatsApp's 24-hour customer-service window, which decides when anyone, person or agent, may reply freely. 5 min
9. 9 Reliability on shared hosting The six things that go wrong when real messages arrive — duplicates, bursts, slow AI, crashes, outages and a full budget — and the database tricks that handle each one on plain shared hosting, with no queue server and no background workers. 7 min
10. 10 Rehearsal before the client sees it A rehearsal file of real questions, a script that runs them through the agent's own prompt, model and guards, and the routine that turns "it seems to work" into an answer sheet the owner has read and signed. First against the stand-in, then with your own key. 5 min
11. 11 Connecting a client's number to the WhatsApp Cloud API The steps to take an agent live on the WhatsApp Cloud API, in plain words. Start on Meta's free test number, then the client's own Business portfolio and app, the webhook and its verification, a permanent token, registering the real number, and the checks that tell you it works. Plus the code that checks every webhook is really from Meta. 7 min
12. 12 What it costs to run, in rupees The three running costs of a reply agent, WhatsApp, AI and hosting, worked out for a small business with the prices dated. Includes the cost calculator, why caching changes the bill, why thinking tokens count, and how to swap the estimate for real numbers after a week. 5 min
13. 13 Privacy — your client's customers' data What personal data the agent handles, where it goes, and the habits that keep a small business on the right side of India's Digital Personal Data Protection Act: a clear purpose, collecting the minimum, deleting on schedule, erasing on request, and a privacy notice customers can read. Not legal advice. 5 min
14. 14 Pricing your service Price your time, not the AI. List the work, time yourself, set a one-time setup fee and a monthly care fee from your own hourly rate, keep the running costs on the client's own accounts, and explain the price in the owner's terms. No income promises here, just a method. 5 min
15. 15 Proposal, agreement and handover Turn the scope and the price sheet into a one-page proposal, agree the terms that prevent the usual disputes (who owns the accounts, what care includes, how it ends), hand the agent over so the owner can run it day to day, and pick the right licence if you build on the AI Reply Agent. 4 min
16. 16 Monthly care, and the capstone The routine that keeps an agent right after go-live: what to read every week, what to check every month, the outside changes to watch for (prices, models, Meta's API), when to end an engagement, and a capstone that takes one business from first conversation to signed-off rehearsal. 4 min

How it works

1. 1 · Buy once

   One payment in your own currency, no subscription. Your account is created from the email you pay with.

2. 2 · Log in and read

   Every lesson opens in your account on any device. Mark each one done and pick up where you left off.

3. 3 · Keep the Word file

   The whole course as a .docx — annotate it, print it, read it offline. Re-download it any time from your orders.

Where this course fits

1. AI for PHP Developers
2. →
3. Build and Sell AI Agents · you are here
4. →
5. Agentic AI Course

Questions about this course