Djam Deen · Backend Dev · based in Benin
← All articles
Why I built AgentMob
AgentMob · · 6 min
It’s 11 pm. My agent is running on the PC, in the other room. It writes code, runs tests, moves forward. And I’m lying down, hoping it won’t need anything — because if it does need something, I’ll have to get up.
That night, it needed something. An approval. One line to validate in a terminal, a few metres away from me, when the only thing I wanted was to answer “yes” from my phone.
I got up. I approved. I lay back down. And I told myself it was absurd.
AgentMob was born from that tiny frustration, repeated a dozen times.
The real problem is not “coding from your phone”
People often imagine a mobile IDE, a tiny editor, a painful touch keyboard. That’s not it.
The problem is staying inside the control loop. A working agent makes decisions: reading a file, launching a command, changing something. Some are harmless, others deserve a pause. The terminal forces you to be present at that exact moment — and that moment rarely arrives when it suits you.
You start a task, you go do something else, and you come back to discover that the agent has been waiting for twenty minutes in front of a question you would have answered in two seconds.
This is not a productivity problem. It’s an availability problem.
The promise: “Telegram for your agents”
Once the problem was properly stated, the solution became obvious. I didn’t want a dashboard, graphs or logs. I wanted the most ordinary thing in the world: a messaging app.
One conversation per session. Bubbles. Notifications when the agent needs me. And an approval card that lets me answer “allow”, “deny” or “allow for this session” with my thumb. Everything else — the steps, the tool calls, the file reads — is tucked into a collapsible block, because I don’t need to watch the agent breathe.
The idea fits in one sentence: your PC works, your phone controls.
What I did not want to build
There are projects where you know very early what you refuse. Here, the list mattered as much as the features.
No network tunnel to install: most of these tools ask for forty minutes of system administration before the first message is sent. That’s not an experience, it’s an entrance exam.
No account: for a tool that talks to agents running on your own machine, an account adds nothing and already announces a future paid plan.
No conversation storage: the agents, files and history stay on the PC. The service that connects the two devices only needs to see one thing: that they are looking for each other.
Honesty as a constraint
AgentMob supports Codex, Claude Code and Antigravity. They are not equal: different protocols, different capabilities, and one of them simply cannot ask for approval.
The temptation would be to announce “compatible with X, Y and Z” and let the user discover the limits by themselves. I chose the opposite: a support badge per agent — full, basic, experimental — clickable, explaining what it means in practice.
It sells less. But a user who discovers a limit in the field does not come back; a user who was told about it organises around it.
The project I wrote before the code
The hardest part was not the code. It was deciding.
With AI agents speeding up the writing, producing a lot and fast in the wrong direction becomes the real risk. So before the first line, I wrote a single document, a source of truth: vision, version 1 scope, architecture, security, experience, tests, risks.
Every decision is numbered and justified, including the alternatives that were discarded. That is what helped me the most: six weeks later, you no longer remember why you rejected an option, and you reintroduce it for comfort.
What I take away from it
The most useful tools often come from a tiny, repeated annoyance. I did not start with market research: I started by no longer wanting to get up.
Then, that constraints are a gift. One device, one folder per project, one source of truth. Each of those restrictions simplified the architecture and reduced the error surface.
Finally, that working with AI agents does not remove the need to think: it moves the effort. You write less code by hand, you spend much more time deciding, framing and testing in real conditions.
That is exactly why AgentMob exists. Because the field is where everything happens — and the field, most of the time, is not in front of the screen.
Djam Deen
Backend Dev · based in Benin
← All articles · Contact