Build your own Coding Agent
Get Bash-Pilled
At the AI Engineering Summit, Nik Pash, creator of Cline, dropped a bomb: "Terminus still beats everything with one tool design." No clever tool calling. Just the terminal in a sandbox.
The next day, at Anthropic's Claude Agent SDK workshop, the refrain was the same: the team was "bash-pilled." Just give the model a shell and step away.

This isn't laziness—it's insight. Model Quality is what matters. Years of architectural workarounds were compensating for limitations that are rapidly disappearing.
So we decided to prove it.
Weekend Warrior Series
We built a coding agent in a weekend. Without writing a single line of code.
Starting from a 50-line bootstrap script, we used the agent to build itself—adding tools, prompts, subagents, session logging, cost tracking, web search, even image and video generation. Every feature was added by telling the agent what we wanted in plain English.
The tutorial walks you through every step:
- Bootstrap with just bash and an API key
- Add file navigation, smart diffs, and verification tools
- Implement cost tracking so you don't blow your budget
- Build subagents for focused tasks like code mapping
- Add skills that load on demand (images, video, web search)
- Watch it build itself, one prompt at a time
No frameworks. No magic. Just the primitives you need to understand how tools like Cursor and Claude Code actually work.
So grab yourself a cup of coffee, block out a few hours, and dig in!