Agent Workforce Org Chart

Agent Workforce Org Chart
Room	Systems
Field	Agent architecture, autonomous operations, organization design
Known for	14 scheduled jobs audited into a workforce org chart under a target reasoning CEO
Key figures	Maestro (principal), the Nordisk worker jobs themselves

Agent Workforce Org Chart — Nordisk Renhet’s scheduled workforce, audited and mapped into a single organization chart, then field-scouted to show the actual current state: which jobs truly ship, which are dead, which are only schedule descriptions that never run. The chart is the target architecture; the field audit is the reality. The reasoning CEO that should sit on top is still the gap.

The vision

You asked a sharp question: do I need actual agents that get to do their own work for real, rather than a graph of scripts?

A graph of scripts — even ten strong ones — is a recipe. It follows steps in order. It cannot wonder why revenue dropped, cannot decide the real problem is the audience and not the campaign, cannot repurpose because it has no understanding, only branching logic.

A graph where each node is a reasoning agent is a team. On a trigger it reads context, forms a judgment, and acts on it. When the world changes it sees it and adapts. It hands off based on what it learned, not a fixed pipe. It resumes memory between runs.

A field scout on 2026-08-22 corrected the earlier optimism. The org chart below is the designed architecture; a per-job audit — reading every cron line, the syslog lineage, each job’s logs, and the live Shopify publish log — found the actual state is far less than “14 agents all running.” The status badges below now carry the verified truth, not the intended schedule.

──────────────────────────────────────────── ─── CEO GRAPH (bird’s-eye reasoning) ─── ─── revenue-router.py → graph-daily-run.sh ─── ─── reads all work products + nordisk.db ─── ───decides what must improve & why─── ────────────────────────────────────── ───────├ delegates & observes └─────── ───├────ü────├───├───├───┋ ┘───┖──┘─ ┘───┖──┘──┘──┘──┘─ ┘───┖──┘──┘─ CONTENT REVENUE GROWTH SHARED OPS OPS OPS RESEARCH MEMORY SYSADMIN

👤 Layer 0 — CEO Graph (the reasoning layer)

Agent	What it does	Status
Revenue Router `revenue-router.py`	Checks nightly metrics; decides `normal / monitor / diagnose`. If revenue=0 or orders=0, routes to diagnosis.	EXISTING
Diagnostic Agent (spawned by router)	Prime Agent that queries nordisk.db, finds why revenue dropped, writes a diagnosis report.	EXISTING

This is your CEO, but it currently does NOT fire and is only a RULES ENGINE when run manually. graph-daily-run.sh is scheduled at 7 7 * * * in /etc/crontab, but the only entry in its log is a manual run on 2026-08-18 — there is no 07:07 cron execution in syslog. A manual revenue-router.py 2026-08-21 run correctly flagged a CRITICAL “No revenue / No paid orders,” but because cron never invokes the graph, that alert produced no diagnosis and no report. Fix the 07:07 firing, then upgrade the rules engine into a reasoning CEO.

📜 Layer 1 — Content Ops (turns harvested signals into publishable work)

Agent	Schedule	Produces	Status
Nightly Topic Generator	03:07 daily	SEO topic candidates	SILENT SINCE 05-30
Organic Pipeline	03:07 daily	turns topics into organic content work	SILENT SINCE 05-30
GSC Opportunity Scanner	system cron	high-impressions / low-CTR query report	COMMENT BLOCK — DOES NOT RUN
Comparison Pipeline	09:00 daily	Swedish competitor-comparison articles	LIVE — RUNNING DRY (21/21 PUBLISHED)

💰 Layer 1 — Revenue Ops (turns money data into actions)

Agent	Schedule	Produces	Status
Daily Revenue Brief	system cron	daily revenue / orders brief	COMMENT BLOCK — DOES NOT RUN
Weekly Ads Audit	system cron	7d ads campaign analysis	COMMENT BLOCK — DOES NOT RUN
Weekly Ads Checkpoint	Mon 07:07	ads checkpoint	DEAD — FAILED TOKEN REFRESH
Ads Monitor	06:37 daily	ads health scan	DEAD — FAILED TOKEN REFRESH

🧠 Layer 1 — Growth / Research (feeding strategy)

Agent	Schedule	Produces	Status
Weekly Business Review (Prime)	Mon 07:37	weekly review + forward recommendation	SCHEDULED — NO OUTPUT FOUND
Validation URL Agent ×3	on-demand	site URL validation reports	ON-DEMAND
Research Hub / Hermes	human-triggered	web research via Tavily, etc.	ON-DEMAND

🗃 Layer 1 — Shared Memory (the workspace all agents read and write)

Agent	Schedule	Produces	Status
MemPalace Sync	08:07 daily	syncs knowledge graph across agents	LIVE
Harvester Suite `run-all.sh`	07:06 daily	pulls Ads / GSC / Klaviyo / Shopify / GA4 → nordisk.db	LIVE

⚙ Layer 1 — Ops / Sysadmin (keeps the box alive)

Agent	Schedule	Produces	Status
Enc1 Watchdog (Caddy)	every minute	keeps enc1 wiki live	LIVE
File-tree watchdog	every minute	keeps file-tree service alive	LIVE
Cleanup-Disk	03:30 daily	disk hygiene	LIVE
dr-field-manager watchdog	every minute	service health	LIVE

The honest diagnosis

Honest count after the field scout: 1 of 14 jobs is genuinely shipping organic output today — the comparison pipeline — and it is a deterministic script with an eval gate, not a reasoning agent. The harvesters and GMC feed work and keep the DB fed. The rest of the chart is either dead (ads tokens), never wired (comment-block Prime Agents), silent (organic pipeline since May), or the target but not yet real (the reasoning CEO).

What the field scout found to be true vs not there:

True: comparison pipeline published 16 Swedish articles Aug 17→22; harvester + GMC feed run daily; watchdogs keep services alive.
Not firing: the CEO graph at 07:07 (manual 8/18 run only) — so the 8/21 “No revenue” critical alert produced nothing.
Dead: ads-monitor and weekly ads checkpoint, both “FAILED to refresh token” for consecutive weeks.
Never wired: the daily-brief / ads-audit / gsc-scan “Prime Agent” entries in /etc/crontab are comment blocks, not commands; reports/ has zero of their outputs.
Silent: the nightly organic topic pipeline hasn’t written a candidate since 2026-05-30.

So the real work is not “just add a CEO over a running army.” It is: keep the one live shipper fed, repair or delete the dead jobs, then add the reasoning CEO on top of sources that actually run.

The 2026-08-22 field scout (verified evidence)

Every claim below was verified by reading the cron entries, the syslog lineage, each job’s own log, the SQLite state, and the live Shopify publish log. The full write-up lives at nordisk/prime-agents/workforce-field-audit.md on the box.

Job	Schedule	Field-scouted status
Comparison pipeline `daily-pipeline.sh`	09:00	ONLY live shipper — 16 articles 8/17→8/22; backlog 21/21 published, running dry
Nightly topic `nightly-topic.py`	03:07	Silent since 2026-05-30 (topic-candidate.json stale)
Organic pipeline `organic-pipeline.py`	03:07	Silent since 2026-05-30
Harvester suite `run-all.sh`	06:07	WORKS — harvests daily
GMC feed	06:07	WORKS — 7/7 feeds uploaded
Ads monitor `ads-monitor.sh`	06:37	DEAD — “FAILED to refresh token” every run
Weekly ads checkpoint	Mon 07:07	DEAD — “FAILED token refresh” 7/27, 8/3, 8/10, 8/17
Graph daily run `graph-daily-run.sh`	07:07	Scheduled but NOT firing — only manual 8/18 run in log
Revenue router `revenue-router.py`	(inside graph)	WORKS when run manually (8/21 critical alert) but not invoked
Daily brief / ads audit / gsc scan “Prime Agents”	—	COMMENT BLOCKS in /etc/crontab — never run, no reports
Weekly business review (Prime)	Mon 07:37	Scheduled; carries inline token; no weekly report found
MemPalace sync / cleanup / watchdogs	varies	Infra — working (watchdogs healthy)

The path forward (re-scoped to reality)

[Feed the one live shipper] Add a discovery node so the comparison pipeline isn’t stuck at 21/21 published. This is the highest-leverage +1 because it is the ONLY channel currently shipping.
[Repair or delete dead jobs] Fix the two failed Google-ads OAuth tokens (ads-monitor + weekly checkpoint), or remove them; convert the comment-block “Prime Agents” into real cron lines.
[Make the CEO actually fire] Find why graph-daily-run.sh never runs at 07:07, so the 8/21-style critical alerts trigger a real diagnosis.
[Resurrect or retire the organic pipeline] It has been silent since May; either rewire its topic source or confirm the comparison pipeline replaces it.
[Add IndexNow + GSC URL inspection] No pipeline currently confirms indexing to GSC.
[Reasoning CEO] Only once the above sources actually run, replace the rules-only router with a reasoning CEO that reads real work products, and observe its own outputs as referee.