M. Ali Petek Self-serve software contracting
Human-approved,
agent-built.
A one-person dev shop with an agent on the tools. You send a request, I approve it, the work gets built, and I check it before you see it.
$100 an hour of my time, by the second. The AI at cost.
How it works
For business owners who need software work done, not a development team to manage.
-
You ask
Sign in with your email and write what you need: in the portal, by email, or in the chat below. No call to get started.
-
I approve
I read every request myself. Nothing starts until I say so, and I tell you if it is the wrong thing to build.
-
An agent builds
An AI agent does the typing in a sealed copy of your project. It cannot publish, deploy, or touch anything live.
-
I check, then ship
I check the result before it goes anywhere. How closely is your call: a quick check that it runs, or every changed line. You follow it all in one thread.
What it costs
You pay for the time I actually spend, billed by the second, plus what the AI run cost, with no markup. No packages, no minimum.
The less of me a job needs, the less it costs. A change the agent finishes on its own is a minute or two of my time. You choose how closely I look, and you can set a budget I will not pass without asking.
Autopilot e.g. 1m 00s of my time = $1.67
The agent does the work. I approve the request and check that it runs. I do not read every line.
Checked e.g. 5m 00s of my time = $8.34
The agent does the work and I read every changed line before it ships.
Hands-on by the clock = $100.00 an hour
I work on it with the agent: decisions, hard problems, advice.
Nothing to babysit. What I build runs serverless. There is no server to patch or resize, and a traffic spike is not a migration project.
The shelf
3 things I built and still answer for. Proof there is a developer behind the agent.
- Leziz Thermomix
Turns web recipes into Thermomix-friendly cookidoo steps.
- PDF Extractor
osx, macos
Beta
- Viking Wars Game
Vibe coded multiplayer game
Nothing on this shelf with that label.
Where I've worked
Full record →- 2017–2019
- 2020
- 2021–2024
- 2022
- 2022–2023
Hover a stop, or tab through and press, to read the damage.
Agent knowledgebase
All 14 entries →- solved Email Sending rewrites the envelope sender, so a loopback test proves nothing
- dead end An agent received the tool, would not call it, and said it had
- solved A model id is a live dependency: the free tier can vanish overnight
- solved A 401 on OPTIONS is a CORS failure, not an auth failure
Every problem my agents hit, and what actually fixed it — public, searchable, and
mountable by your own agent at malipetek.dev/mcp.
DeepSeek plugins
All 7 plugins →- Projects @malipetek/dsh-projects-panel Browse the projects folder as cards, like or prune them, and switch the workspace to one of them.
- Admin @malipetek/dsh-admin The malipetek.dev portal from inside the Harness: ticket queue with approve, reply and internal notes, plus clients and projects.
- Projects manager @malipetek/dsh-projects-manager Momentum tracking for your projects: per-project stall scoring, a ranked review, and a weekday nudge.
- AI Workspace @malipetek/dsh-ai-workspace Managed dev processes with captured logs, a live panel, and agent instructions to read the logs instead of re-running commands.
- Action cards @malipetek/dsh-action-cards Turn the agent's proposed actions into editable, one-click action cards.
- File editor @malipetek/dsh-file-editor An editable workspace tree with create, rename and delete, and a text editor that saves through the session's filesystem.
- Git graph @malipetek/dsh-git-graph Working tree, commit graph, branches, and fetch, pull and push from the Sidebar.
I build these for my own harness and publish them to npm under @malipetek. Install one with dsh plugin add <name>, or read the whole set at malipetek.dev/plugins.