The roster · 101 agents, 11 teams available, 1 still planned

101 agents. 11 teams available. One conductor who never touches the work.

This is a real, running system, not a diagram. A growing bench of specialized agents, organized into teams, each team built for exactly one job. Above all of them sits one additional agent whose only role is deciding what runs next. The roster has already grown twice since this page was first written; it will again.

Every one of the 101 has exactly one job. That's not a limitation. That's the entire design.

101
Agents in the system
11
Teams built and running
20+
Distinct disciplines
1
Conductor above them all
CONDUCTOR RAILS ENG · 10 GO ENG · 10 TAURI ENG · 10 DESIGN · 12 RAILS QA · 9 TAURI QA · 9 IDEATION · 6

A sample of the roster, not all ten teams. See the full list below.

The teams

Eleven built and running today, one still only planned. This roster has already grown twice since it was first counted: a security family didn't exist yet, market research was still a promise, and copywriting was one lone scaffolding file. Each team is independent, built for one job, and never calls another team directly.

10 agents

Rails Engineering

Builds and reviews a Rails web feature end to end, from a plain-language description to a reviewed implementation.

engineering family · production-proven

10 agents

Go Engineering

Builds and reviews a Go command-line tool end to end, the same job as the Rails team, different stack.

engineering family · production-proven

10 agents

Tauri Engineering

Builds and reviews a Rust and Tauri desktop feature end to end, the same job again, a third stack.

engineering family

9 agents

Rails QA

Verifies a finished web feature against the actual running application across six distinct disciplines.

QA family

9 agents

Tauri QA

Verifies a finished desktop feature against the actual running application, purpose-built for a native window.

QA family

6 agents

Rails Security

Verifies a finished web feature against its actual live security posture: real headers and TLS config, real dependency CVEs, and, where explicitly authorized, real penetration testing in an ephemeral Kali Linux container.

security family

5 agents

Tauri Security

The same live-verification job for a desktop build: capability boundaries and network surface, checked against the shipped binary's real behavior, not its declared config.

security family · one fewer specialist than Rails on purpose

15 agents

Market Research

Assesses whether an idea is viable before any engineering effort is committed to it: demand, competitors, pricing, regulatory exposure, and more, across the largest team on the roster.

6 agents

Ideation & Discovery

Turns a raw, brand-new product idea into a structured, human-approved product brief.

12 agents

Design

Turns an approved idea into a reviewed static design concept, checked for accessibility, performance, and AI discoverability.

8 agents

Copywriting

Writes marketing copy for a completed feature: a content plan, a narrative angle, landing page copy, an email sequence, and a humor pass, staged rather than run as one parallel panel.

planned

Marketing

Produces launch assets and messaging once a feature is built and verified.

The disciplines

Not one generalist doing everything passably. Specialists, each covering one discipline, spread across the roster.

Orchestration Discovery Architecture Design Engineering Code Review Security Review Performance Review Interface Review Accessibility QA Functional QA Cross-Platform QA Load QA Visual Design QA Copywriting QA SEO AI Discoverability Research Brainstorming Log Analysis Skill-Building Live Security Testing Dependency & CVE Auditing Market Sizing Team Auditing

The conductor

One additional agent sits above every team on the roster and does none of their work. It doesn't build, doesn't design, doesn't review a single line. Its only job is deciding which team runs next, and it confirms that decision with a human before acting on it, every time.

A second, separate agent sits beside it, not above the teams' work, but above the teams themselves: it audits whether a newly built team is actually constructed correctly before anyone trusts it. It was run for real against the newest addition to this roster and found two genuine bugs a first read had missed.

It's the only agent in the system whose job is knowing about all of the others, and touching none of their work directly.

Questions worth answering directly

How many agents are in this system?

101 in total: 100 spread across 11 available teams doing real engineering, design, QA, security, market-research, and copywriting work, plus one conductor sitting above all of them.

What disciplines does the system cover?

Roughly 20, spanning discovery, architecture, design, engineering, several kinds of code review, six kinds of QA, SEO and AI discoverability, research, log analysis, and skill-building.

Is every team the same level of maturity?

No. Some teams are production-proven across many real runs; others are newly built and still being hardened. The roster is a living one, not a fixed set.

What does the conductor do?

It decides which team runs next and confirms with a human before every handoff. It does not build, design, or review anything itself.

This roster is one artifact of a larger practice.