Zarif Choudhury
AI Automation Builder & Educator
Zarif builds AI agents and automation workflows. He writes about what keeps working once real people use it: who is worth reading, the new jobs AI is creating, and agent workflows you can check step by step.
Things you can check yourself
The advice here comes with something you can open: a list of sources, an example you can run, or the work itself.
What I do
I build AI agents and automation workflows. This site is where I write down what held up: which sources are worth following, what forward-deployed and AI-engineering jobs involve day to day, and how working agent workflows are put together.
The source lists show where each pick came from and the date it was last checked.
One workflow, run in the open
This is one of three workflows I publish in the open, built in n8n, a tool for chaining automation steps together. It runs here on a sample lead, so you can see how it scores the lead, what it sends back, and what it needs to set up.
It shows the workflow working on test data. It is not a result from a client.
INSPECTABLE EXAMPLE · LEAD INTAKE, ENRICHMENT, AND ROUTING · FREE PACK v1.0.1
What one free workflow actually does with a lead
This demonstration runs the published workflow against its included test fixture. The rules are the scoring expressions inside the export, and the output is the response the webhook returns. It shows the workflow's behavior, not measured customer results.
Input · fixtures.json → success
{
"leadId": "lead_demo_1001",
"name": "Jordan Lee",
"email": "jordan@example.com",
"company": "Example Co",
"companySize": 12,
"budget": 2500,
"source": "website",
"notes": "We need help automating lead response and qualification."
}Visible rules · score and route
- Budget 2,500 (1,000 or more)+25
- Company size 12 (5 or more)+20
- Notes 56 characters (40 or more)+20
- Score (capped at 100)65
- Route is priority only when budget is 5,000 or more or company size is 20 or more. This lead meets neither, so it routes as standard.
Output · HTTP 202
{
"accepted": true,
"leadId": "lead_demo_1001",
"route": "standard",
"acknowledgementQueued": true
}Invalid input "invalid" · HTTP 400
{
"accepted": false,
"error": "A valid name and email are required."
}What it does not do
- The export ships inactive and credential-free. You add the header-auth and SMTP credentials before it can run.
- Nothing is persisted. The caller is responsible for storing the accepted lead and its route.
- It replies 202 before sending mail, so a lead can be acknowledged even if the acknowledgement email later fails.
- There is no duplicate detection, delivery log, or error-trigger workflow. Those are production concerns the starter leaves to you.
- Open-source sales skills — skills for AI agents. You can read every one before you install it.
- Start here — where to begin, whether you want sources, a new job, or agent workflows.
- The stack — the tools I build with right now, and what each one is for.
Where to find me
- Still Building — The newsletter: one story from the AI world, told properly, and what I make of it
- X (@zarif_cho) — Notes on agents, AI careers, and who is worth reading
- GitHub — Open-source agent skills you can read before you install
- YouTube — Builds, recorded start to finish
Why this site exists
AI tools change every week. The hard part isn't finding new ones. It's knowing which sources, jobs and ways of building are worth your time.
So this site keeps up-to-date lists of who to follow, a clear path into AI-engineering jobs, and agent workflows you can check from start to finish. That leaves you more time to pick a tool and build something with it.