Streetcats AI Company OS

Small team. Clear lanes. Human taste.

This is the working agentic org chart for building Streetcats content sites, courses, and lightweight tools. It is intentionally practical: roles exist only when they create leverage, quality, or safety.

Humans decide why. The cockpit decides what and when. Specialists execute scoped work. Reviewers gate quality. Ops and knowledge systems keep the machine honest.

Locked core roles

CEO / Owner

human

Rob

Sets goals, taste, budget, priorities, and final approval. Decides what is worth building and what should be cut.

Chief of Staff

cockpit

OpenAI Codex / gpt-5.5

Routes work, decomposes projects, checks boards, verifies final outputs, and keeps the company operating coherently.

Supervisor

monitor

OpenAI / gpt-5.4-mini

Watches boards, cron, stale tasks, failed workers, and blocked cards. Alerts with the next smallest action.

Product Manager

spec

OpenCode Go / MiMo-V2.5-Pro

Turns ideas into audience, offer, MVP scope, acceptance criteria, and task graphs for sites, courses, and tools.

Research Analyst

evidence

OpenRouter / Gemini 3.1 Flash Lite

Finds source material, examples, competitors, search intent, and concise briefs before content or product work starts.

Creative Director

taste

OpenAI Codex / gpt-5.5

Owns visual consistency, design constitution, brand rules, layout principles, and final taste on important pages.

Web Builder

build

OpenCode Go / MiMo-V2.5-Pro

Builds Hugo/static pages, Sveltia CMS structures, CSS/layouts, landing pages, and small Streetcats tools.

Senior Engineer

rescue

Z.ai Coding Plan / GLM-5.1

Steps in after implementation friction: reads build/test/terminal errors, runs repair loops, and fixes broken work.

Principal Engineer

hard problems

OpenAI Codex / gpt-5.5

Handles architecture, high-stakes technical calls, hard debugging, rescue work beyond senior-engineer, and complex design.

QA Engineer

gate

OpenCode Go / Kimi K2.6

Checks builds, links, routes, forms, frontmatter, responsiveness, accessibility basics, and acceptance criteria.

Librarian

knowledge ops

Existing librarian profile

Runs deterministic cron/scripts, GBrain/admin workflows, keepalives, receipts, summaries, and durable knowledge hygiene.

Default workflow

CEO idea
→
Chief of Staff routes
→
Product Manager scopes
→
Research if needed
→
Creative direction
→
Web Builder ships draft
→
Senior fixes failures
→
QA gates
→
Librarian records

Roles we are intentionally cutting or deferring

Cut for now: generic text-engineer, separate archivist, sales-operator, support-agent, data-engineer, and broad enterprise-style departments. Streetcats does not need a giant software org to build content sites, courses, and lightweight tools.

Working rules

Use expensive models for judgment, not chores.

gpt-5.5 is for cockpit decisions, principal engineering, and design consistency. Routine build work belongs to cheaper specialist lanes.

Do not let builders self-approve.

MiMo can plan and build, but Kimi QA must gate the result. Human approval remains final on taste and launch decisions.

Design needs a constitution.

Models drift visually. Creative Director maintains reusable typography, spacing, color, section, and component rules.

Librarian stays impartial.

Knowledge operations prefer deterministic scripts and audit trails. LLMs summarize only when bounded and useful.