coding-worker
Active · developer tool
I want agents to do more than write code. They also explore projects, verify changes and review code. coding-worker gives those tasks a shared MCP interface.
MCP · Pi / OpenCode · Bubblewrap · Git
PHP & other languages · Architecture
Hi, I’m Vojta Biberle.
Open to long-term opportunities
Software Architecture · Agentic Engineering · Backend
Remote preferred · Occasional on-site · B2B or employment
Let's talkI've spent most of my life writing software. I started with PHP, and I still enjoy it. As I got into agentic development, other languages and infrastructure quickly became part of the work, and my main focus shifted towards architecture. The language itself is secondary now.
Agentic development has shifted the work from writing code to setting clearly defined goals for software. But without an environment set up so agents can find their way around, all you get is incomplete pieces and software that doesn't work.
How I work with agents →
from brief to verification
That's why I've been focusing on tools and approaches that help agents understand a project and on building the right environment for developers, architects and agents alike.
I build environments and tools that help agents navigate a project and make it possible to validate their work thoroughly. I'm experimenting with advanced agent orchestration and pushing the limits of how much work I can safely hand over to agents, while keeping security boundaries in place.
And when I'm not at the computer, I spend time with my daughter, ride my scooter or bike, look after the house and garden, and sometimes fly kites or look at the stars through a telescope. :-)
- Agentic Engineering
- Processes and environments for autonomous work by AI agents.
- Software Architecture
- Backend systems, data platforms, APIs and microservices in PHP and other languages.
- Developer Infrastructure
- CI/CD, containers, isolated environments and developer tools.
Selected projects
These projects are where I explore what helps agents work and what I need to keep under control. They cover task delegation, development environments and backends.
AgentBox
Active · server preview
When an agent runs commands and edits files, I want to be clear about what it can access. AgentBox separates its workspace and state from the rest of the machine.
Python · Podman / Docker · Git worktrees
BareAPI
Experiment
In BareAPI, I'm experimenting with describing data models through JSON Schema and building a JSON API around them. Still using familiar PHP, Symfony and Doctrine.
PHP · Symfony · Doctrine · JSON Schema · PostgreSQL · FrankenPHP
git-wt
Developer tool
I don't want to keep setting up working directories and switching between them by hand. git-wt combines worktree creation, environment setup and directory navigation in one command.
Bash · Git worktrees
How I work
An agent gets a concrete brief and a bounded workspace. The result must pass tests, an independent review and finally human approval.
01 / 06 · SPEC
First, I need to know what I actually want
The agent needs a clear goal, project context and criteria that tell us when the work is actually done.
02 / 06 · TASKS
I break a large task into smaller ones
A milestone becomes specific tasks with clear dependencies. Agents can then work independently without me directing every move.
03 / 06 · AGENTS
Now they can get on with the work
Agents get their own environments, the tools they need and clear security boundaries. I want to give them enough freedom, without unrestricted access to everything.
04 / 06 · TESTS
“It works for the agent” isn’t enough
The result needs real verification. I rely mainly on integration and E2E tests, together with static analysis.
05 / 06 · REVIEWS
An agent shouldn’t approve its own work
An independent model reviews the changes. If it finds problems, the work goes back for fixes and another round of verification.
06 / 06 · HUMAN APPROVAL
I want to make a decision, not repeat every check
I want a change ready for review, with test results, an independent review and an explanation of any remaining risks. Then I can decide whether to approve it.
What I'm building
I'm exploring how far I can take autonomous development: how to set up agents' environments, divide the work and thoroughly check the result. I'm focusing on orchestration, automated checks and the decisions that still need a human.
- Open WebUI SIWC Linux — A Linux adapter for Sign in with ChatGPT in self-hosted Open WebUI. Sign-in works from Linux, no macOS app needed.
- coding-worker — I keep implementation, analysis and review separate, with traceable results from agents' work.
- AgentBox — I work on separate workspaces and security boundaries; the server runner is still a preview.
From the blog
Selected reading
- WordPress Theme from Lovable HTML in One Day
I took an HTML export from Lovable, opened Claude Code, and in one day had a complete WordPress theme -- 11 custom blocks, 4 custom post types, deployment…
- Moving 30 PHPUnit Suites from Azure Pipelines to GitHub Actions
Everything started on Azure Pipelines. A year ago we moved PR validation to GitHub Actions, but deployments and E2E testing stayed on Azure. I designed the…
- How This Blog Works in 2026
Back in 2019 I wrote about how this blog works. A lot has changed since then — automated deploys, dev.to sync, AI-generated tweets. Time for an update.
