Jorge
700+ hours stuck on his dream app. One month working together. Shipped.
A 12-week cohort for semi-technical operators and software engineers who want to stop plateauing inside Claude Code and start pushing it the way it was meant to be pushed. Live sessions, pre-session materials built for your project, and the MACHINE framework: the same system used to run a 150,000-line command center solo.
20 seats. Cohort one starts soon. Application only.
700+ hours stuck on his dream app. One month working together. Shipped.
Skeptic stuck in planning. Built end-to-end, launched to alpha testers, running multi-agent pipelines for testing and marketing.
Built a semi-autonomous dev pipeline that 10x’d his output. Reinvested the freed time into sales.
A 12-week program built around one question: how do you get a coding agent to do the work of a team?
Three concrete outcomes by graduation.
By the end of 12 weeks you’ll be in the top 1% of agentic coders. Mastery doesn’t mean there’s nothing left to learn. It means you’re operating at a level almost nobody reaches. The skills are transferable: they work inside any agentic coding harness, not just Claude Code.
A personal system that saves you at least 20 hours a week and automates every part of your work you want automated. Romij’s 150,000-line command center is the blueprint, not a template to copy. Yours will be built for your business, not his.
By week twelve, either a shipped product or app, or end-to-end automation of the work you wanted off your plate. You leave with something running, not something planned.
Seven pillars for operating any coding agent at a high level. The scaffolding underneath everything in the cohort, not a sequence to march through.
Planning tasks, navigating unfamiliar codebases, exploration patterns. How an agent figures out where it is before it does anything.
Subagent orchestration, parallel execution, delegation. Turning one agent into many, then getting them to work together without stepping on each other.
Making every token fight for its place in the context window. Active context control and LLM-friendly architecture: the single biggest lever on output quality.
CLAUDE.md, hooks, slash commands, MCP, tool configuration. Shaping the environment the agent runs inside so the right behavior is the default.
Model intuition, trust calibration, failure detection, monitoring. Knowing when to let the agent run and when to step in, before the damage is done.
Prompt craft, dictation workflows, XML tags, specs-as-code. How to talk to an agent so it does what you actually meant.
TDD, CI/CD, deployment. What engineering discipline looks like when most of the code is written by an agent.
Applications are reviewed by Romij. If it’s a fit, you’ll get a call. No bot in between.
Romij will review and reach out within 3 business days if it’s a fit. No reply = not this cohort.