Building AgencyIQ
AgencyIQ is the operating system running my own businesses — CriticalKPI, Best of Zion, and the catalog and agency work behind both. It plans the work, then hands it to Claude Code overnight:
What AgencyIQ does
Running several businesses at once means running several sets of tools, and the best time to build is when nobody's asking for anything — but nothing runs itself overnight unless something is built to do exactly that. AgencyIQ is what I built to close that gap: a companion to Claude Code that plans the work, dispatches it, ships it, and heals what breaks before I'm awake.
- The Conductor: reads every tracked project's milestones, checks real Claude usage headroom, and dispatches isolated, worktree-scoped Claude Code sessions to build the next piece — unattended, budget-capped, one branch per run.
- The Governator: consolidates task, project, milestone, and incident logs into one stream, then runs a dedicated overnight shift correcting what's safe to correct on its own — never restarting, merging, or deploying live code without a human gate.
- Morning Coffee: turns the night into one page — reminders, what needs my call, what shipped and is ready to push.
Why this matters
Handing real work to an unattended system only pays off if the leash is as considered as the reach. AgencyIQ is built to earn more autonomy over time, not assume it up front — every guardrail below exists because "it ran overnight" isn't the same as "it ran safely."
The Conductor
Prioritizes what's ready and dispatches the next milestone to a worktree-isolated Claude Code session.
The Governator
Takes the back half of the night — one incident stream, reviewed and self-healed where it's safe to.
Morning Coffee
The daily digest every module writes into — five reminders instead of a wall of notifications.
What it took off my plate
- Context scattered across every business's own tools, logins, and half-finished task lists
- Work that only made sense to run at night, with nobody around to run it
- CRM, email, LinkedIn, and social outreach each living in their own silo
- No single place to see what actually shipped versus what just got dispatched
AgencyIQ at a glance
AgencyIQ runs CriticalKPI, Best of Zion, and the catalog and CRM work behind both, every night, before it runs anything for anyone else. If a module isn't trustworthy enough to run unattended against my own work, it doesn't ship — that's the bar, and it's the reason the guardrails came before the growth.
Two shifts, one loop, before I'm awake.
Dispatch at night, review with coffee. Every module feeds the same overnight stream, and the same stream is what tells the Governator whether tonight's incidents are actually going down.
Everything the agency runs on.
Each module works standalone, and each one feeds the same overnight cycle — the same milestones, the same incident stream, the same morning digest.
CRM & Campaign Lists
Contacts, stages, and segmented lists that feed outreach without becoming a second source of truth.
liveEmail Marketing
Mailbox-throttled campaigns with real open, click, and bounce tracking — not just a send button.
liveLinkedIn Engagement & Campaigns
Budget-throttled outreach and an inbox with AI-drafted replies — every send held for my approval first.
liveSocial Media Marketing
Generate, review, schedule, and publish across brands from one install, with performance flowing back in.
liveSEO / GEO / AEO Audits
Judgment-heavy site audits run as a standing runbook — the same discipline as a daily health sweep.
buildingCatalog Enrichment & Scoring
Turns a raw product feed into an enriched, scored catalog — the discoverability layer behind CatalogIQ.
buildingSellers CoPilot
The client-facing side of the same engine — a companion app sellers use directly, not just an internal dashboard.
buildingProject Management
Reads every tracked business's real project files — milestones, active tasks, blockers — as one roadmap.
liveThree steps, every night.
What to build is decided continuously, straight from each project's own tracked milestones. What happens after dark is three steps.
Dispatch & build
23:00 — worktree-isolated Claude Code sessions, one per milestone, resource-checked before launch
Heal & optimize
Back half of the night — the Governator reviews the incident stream and fixes what's safe to fix
Report
07:00 — Morning Coffee turns the night into reminders, blockers, and what's ready to push
Then it repeats — tomorrow night.
Autonomous overnight. Accountable every morning.
Every rule below exists because handing real work to an unattended system means the leash matters more than the reach.
No unattended deploys
Overnight sessions build and commit to their own branch. Merging, restarting, or shipping live code always waits for me.
Worktree isolation
Every dispatched session runs in its own git worktree and never touches a file outside the project it was assigned.
Fails open, not silent
If usage or resource data goes stale, the engine assumes capacity is fine rather than quietly blocking real work.
Budget-capped, every night
A nightly ceiling on jobs and spend, the same discipline applied to LinkedIn sends and email throttling.
I didn't build this to sell it first. I built it to run my own agency.
AgencyIQ runs my own businesses every night, before it runs anything for anyone else. If a module isn't trustworthy enough to run unattended against my own work, it doesn't ship.
AgencyIQ isn't a dashboard. It's a system that runs itself, and answers to me every morning.