An AI agent here is software with one agreed job, not a widget that will talk about anything. It reads systems you already have, follows the steps we agree, and stops when a person should take over. Mani writes the instructions, the tool calls, the limits, and the path for when the model is wrong. It can plug into WordPress, WooCommerce, Shopify, or a custom app, and it keeps a log of what it did.
Mani Khan · software engineer & freelancer since 2009
AI agents that do one job, then hand it back
An AI agent is software with one defined job, not an open chatbot. It is for small and medium businesses that want a task taken off the team, and for WooCommerce and Shopify store owners, WordPress site owners, and agencies that need that job wired into a system they already use.
01 The job
What you get
The agent takes a specific job off your plate. It is not a widget that will talk about anything. It reads the systems you already have, follows the steps we agree on, and stops when a person should take over. The jobs I take on are: answering from your own pages, docs, or product data and saying when it does not know; drafting a reply, summarizing a thread, or looking up an order before anyone hits send; and plugging into WordPress, WooCommerce, Shopify, or a custom app with a log of what it did.
I write the integration myself: the instructions, the tool calls, the limits, and the path for when the model is wrong. You are not working with a broker.
-
Sources
Your pages, docs, or product data. It says when it does not know.
-
The steps
The instructions, tool calls, and limits we agree before it runs.
-
The handoff
A log of what ran, and a path for when the model is wrong.
02 Who it’s for
Who it is for
Buyers are small and medium businesses that want automation, WooCommerce and Shopify store owners, WordPress site owners, and digital agencies that need the same build as white-label development. Freelance work has run since 2009.
An agent uses the tools the job needs: WordPress, WooCommerce, Shopify, GitHub, n8n, Zapier, HubSpot, and GoHighLevel. Webhooks, PHP, Laravel, and JavaScript are available when the integration calls for them.
03 What you leave with
Deliverables
- A written definition of the one job, the sources it may read, and when a person takes over.
- The instructions, tool calls, and limits, written and wired by me.
- A log of what the agent did.
- A path for when the model is wrong.
- Notes your team can use to see what ran and how to stop it.
04 The recipe
Process
Every project goes through the same six stages. The steps below are how a website is built. Stores and automations go through those same stages.
Step 01 — Discussion
We talk about the goals of the website. Once we know why you want a new site, we work together to identify the content it needs and the features it should have.
Step 02 — Planning
Once I understand your needs, I create a plan to design and develop the website to meet your specifications.
Step 03 — Design
Functionality matters, but good design makes a visitor want to stay. I work on the layout with responsiveness, a user-friendly experience and an aesthetically pleasing look in mind.
Step 04 — Coding
The most specialized stage. I pride myself on writing clean, organized code, so you have a highly functioning website for years to come.
Step 05 — Testing
I test the site internally first, then you test it on your end. Everything should work, and it should launch with no loading errors or bugs.
Step 06 — Launch
The site goes live on your domain for the world to see. I hope you'll come back to me for future updates and digital marketing.
On an agent, discussion names the one job and the sources it may read. Planning is the sequence of steps and the stop rule. Design is the flow a person can follow. Coding is where I write the integration. Testing checks a normal run and a failed step. Launch is the agent running on your site or store, with later updates available if you want them.
05 Before you write
Questions
The agent stops at the handoff we agree in discussion, and when it does not know the answer. It says when it does not know, and a person takes over rather than the agent inventing a next step. The log shows what it did before that handoff. The exact stop rule for your job is set before coding, not after a surprise in production.
Yes. The build includes a log of what it did, so you can see what ran. Agreed per project and written into the scope. The integration also includes a path for when the model is wrong, so a bad answer is not a dead end. The log shows the steps the agent took before a person takes over.
A project starts when you email mani@manikhan.com or use the form on this site, and we discuss the goal before anything is built. Discussion identifies the content, the features, and the job to be done. Planning comes next, then design, coding, testing, and launch. You do not need a finished specification. The goal, the deadline, and what already exists are enough to begin.
The starting rates are listed on the pricing page. Send your project details for a written quote. Payment handling is part of that written quote, and it is set before coding starts. The engagement is a one-off project, monthly support, or white-label work. Email mani@manikhan.com with the job.
06 Also in the lab
Related
07 Summon the lab
Start a project
The starting rates are listed on the pricing page.
Or send the form. Say the goal, the deadline, and what already exists. I reply to the email you enter.