Abed Manzalji

AI EngineerToronto

Abed Manzalji

I build AI agents that run in production, and automations that replace manual work.

Get in touchRead the walkthroughs
RequestRouterSearchAnalyzeDraft…CheckDone

The pattern: a router sends each task to a specialist agent.

What I build

How I work

Prototype fast. Learn from real people. Ship what works.

The story

From a vague request to a shipped fix

7 steps · 1 minAn example project, start to finish.
Read the story as text
  1. Find the real problem. The client asks for an app. I don’t start building. I ask why. We keep asking. Until the real problem shows up. People can’t see where their order is.
  2. Prototype fast. I build a rough version first. Just enough to test the idea. Rough, but real enough to try.
  3. Get feedback. The client tries it for real. They get stuck. Now I know where. I write down every snag.
  4. Fail fast. Version one misses. So it’s dropped early, while it’s cheap. Then on to the next version. Version two uses what we learned.
  5. Ship fast. Version two works, so it ships. A real user tries it. It does the job.
  6. Watch people use it. Now I watch real people use it. One of them gets stuck. That’s my next fix.
  7. Apply their feedback. I fix what they found. The stuck user gets through. Then the loop runs again.
Read the answers as text

Why a team of AI agents, not one?

  1. Overloaded. One agent juggling every task and tool starts dropping things.
  2. Split the job. So I split the job. Each agent does one thing well.
  3. Pass it on. Each one does its part and hands back a result.
  4. Still finishes. If one gets stuck, the job still gets done.

How do you keep AI agents safe?

  1. Too much access. An agent with every permission can do damage by mistake.
  2. Least privilege. So each agent gets only the tools its job needs.
  3. Screen the input. Untrusted text is checked for hidden instructions first.
  4. Hard limits. Even a fooled agent can’t use access it doesn’t have.

What should a company automate first?

  1. Spot the repeat. Start with work someone repeats by hand, every day.
  2. Count the cost. Add up the hours. Small chores get expensive.
  3. Hand it to a script. Software takes the routine. People keep the judgment calls.

Blogs

Real AI cases, replayed one step at a time, with the fix and the code behind it.

All 21 walkthroughs
  • LabInteractive demos of how AI models behave.
  • Human in the LoopApprove or block what an AI agent wants to do.
  • GlossaryEvery AI term in one plain line.

Contact

Email is the fastest way to reach me.

manzalji.abed@gmail.com
LinkedInGitHub
Based in
Toronto, Canada
Languages
  • English
  • Arabicالعربية