The last 10% is the product
Planning, retries, permissions and logging are not polish. They are the difference between something you can leave running and something you have to babysit.
I build the part after the demo.
I'm Ali, an independent automation engineer building agentic AI systems, which is part engineering and part research. Most of my work is the part after the demo.
That means tool contracts that don't fall over when an API returns something unexpected, retrieval that finds the right passage instead of a plausible one, evaluations that catch a regression before your customers do, and spend limits so a looping agent doesn't turn into an invoice.
I spent years building ordinary software products first, which is why the agents I ship come with a real interface, proper error handling and documentation someone can follow. The interesting problem is rarely the model. It's the workflow around it.
A lot of this work starts the same way: something half works, nobody is sure why, and the person who built it has moved on. That is a good starting point rather than a bad one, and it is often the quickest route to something your team can actually use.
I work directly with the people who own the process, in short feedback loops, and I would rather tell you a project is a bad fit than sell you something that quietly rots.
At a glance
Focus
Autonomous agents, multi-agent systems, retrieval and automation
Based
Remote, working with teams worldwide
Reply time
Within one business day
Status
Open to new work
How I work
Four things I don't compromise on
Measure before you scale
If we can't tell whether the agent got better or worse, we don't put it in front of customers.
Humans stay in charge
Approval before anything irreversible. Automation should remove typing, not accountability.
Plain language, no mystery
You should be able to explain how your own system works to a colleague. If you can't, I've explained it badly.
Background
How I got here
Now
AgentCrafts
Working independently under the AgentCrafts name, building autonomous agents and automation for small teams that move quickly.
Recent
Agent and automation engineering
Multi-agent pipelines, retrieval systems and tool integrations across operations, support and content workflows.
Before
Product engineering
Full-stack TypeScript and Python work delivered end to end. It is why these projects have real interfaces and not just a notebook.
Who you would be hiring
Questions about the person
Is this one person or a team?
One person. There is no account manager and no one to be passed to, which is the point: whoever answers your message is whoever writes the code. For work that needs several people at once I will say so and point you elsewhere.
Where are you based, and what hours do you keep?
Remote, working with teams worldwide. I keep a deliberate overlap with your working hours for calls and for anything that needs a decision, and the rest of the work happens on my own schedule.
How many of these have you built?
64 agents and automations in production, across 33 systems. Plenty of them are unglamorous: an inbox triage agent, a report that used to take someone's Friday, a voice agent that books appointments.
Do you work with small teams?
Mostly, yes. Small teams are usually where the return is obvious fastest, because the person doing the task by hand is also in the room when we scope it. Company size matters far less than whether the workflow is clear enough to describe.
Will you sign an NDA?
Yes, happily. Send yours, or I can provide a simple mutual one. It costs nothing to sort out before we talk about anything sensitive.
What happens if you are unavailable?
You are never left holding something only I understand. Every project ends with documentation, a walkthrough recording and credentials in your own accounts, so any competent developer can pick it up. That is deliberate, not an upsell for support.
Let's build
Worth a thirty minute call?
Tell me what your team does by hand every week. I'll tell you honestly whether an agent is the right answer, what it would take, and roughly what it costs to run.