Lunatechs Event · May 30, 2026
A cafe session for testing an open-model coding workflow.
This was not a talk dressed up as a workshop. It was a compact build lab: DeepSeek as the model, OpenCode as the coding-agent interface, laptops open from the start.
I helped host as one of the Lunatechs organizers. The goal was simple: make the room approachable for first-time builders without boring the people who already live in a terminal.
Drop By Dough helped. A cafe lowers the pressure. People can ask setup questions, move tables, grab something sweet, and get back to debugging.
Hosting the room
The goal was to make the first step feel less awkward.
Many AI coding events assume everyone is already comfortable with terminals, installs, model keys, and broken agent output. Real rooms are not like that.
My job was logistics plus momentum: help first-timers ask the first question, and give stronger builders room to explore without turning the session into a lecture.
The workflow question
The real question was how much control the human keeps.
Choose the task
Pick something small enough to build in the room, but real enough that the agent has to work inside a project.
Let the agent propose
Use OpenCode to generate a first pass, then inspect the commands, files, and assumptions before trusting the result.
Run the thing
Local feedback matters. A working screen, a failing command, or a broken import teaches faster than another abstract tool comparison.
Steer, do not drift
The human job is still taste, constraint, and review. The agent can move quickly, but the builder has to keep the shape of the project in view.
The session
Free credits helped people get from interest to action.
Alibaba Cloud joined as tech partner, gave a short database-tech session, and helped people get started with DeepSeek credits. That removed a boring blocker: getting the tools working.
Once people could run things, the conversation got better. Less “is AI coding impressive?” More “what can I build, what should I inspect, and where do I keep control?”
What changed the conversation
The tooling was unfinished, which made it worth testing.
AI coding is not one interface winning everything. Cost, openness, model choice, and control start to matter as soon as you build with the tools.
That made the session honest. We tested where the workflow felt smooth, where it broke, and which builders might prefer it.
Cafe texture
Donuts, laptops, and enough noise to make the session feel alive.
The venue mattered. A cafe makes basic questions easier. People could drift between tables, grab something sweet, and come back to the screen.
For vibe coding, that informality is part of the learning environment. You want people trying things before they feel fully ready.
Closing note
The stack is still unsettled enough to test in public.
That is why I like these Lunatechs sessions. Run the tool, watch it break, compare notes, and leave with a sharper instinct for where it belongs.
With thanks to
Partners who made the room work.
Venue partner
Drop By Dough
Opened the Central space and kept the afternoon feeling like a cafe build session, not a classroom.
Tech partner
Alibaba Cloud
Shared a database-tech session and helped people start with DeepSeek credits, removing friction between curiosity and a working local project.
