One coordinated brain that runs social, reviews, SEO, website, reporting and content.
Six departments · 22 agents · one human — you. Built in six weeks, run from AED 550 / month.
Automate the busywork — never the judgmentOne coordinated brain that runs social media, Google reviews, SEO, website fixes, content production and reporting — 6 departments, 22 AI sub-agents, one human approver. This is the whole plan: mind map, infographics, the tools and what they cost, an honest feasibility verdict, and a step-by-step you can follow even if you've never written a line of code.
EatCookJoy is a trust-heavy, two-sided UAE chef marketplace. The smart automation model is “AI-first with human approval” — the agents do the repetitive work (posting, requesting reviews, auditing SEO, fixing the site, pulling KPIs) and the owner only approves content, signs off negative-review replies, and makes the strategic calls. Every department writes to one shared database (Supabase), so the system behaves like one brain, not six disconnected tools.
Creates and publishes Reels, posts and stories to Instagram, TikTok, LinkedIn and Facebook from one content calendar.
4 sub-agentsRequests a review after every session, drafts AI replies, auto-posts positive ones, escalates negatives to you on WhatsApp.
3 sub-agentsWeekly technical audits, auto-fixes meta & schema, and tracks whether you appear in ChatGPT, Perplexity & Google AI answers.
4 sub-agentsDaily checks for broken links, slow pages and indexing errors. Fixes the safe ones, raises GitHub issues for code-level ones.
3 sub-agentsAll social, review, SEO and booking numbers flow into one Looker Studio dashboard. A WhatsApp GM report lands every Sunday.
4 sub-agentsIdea → script → AI voiceover → video render → auto-captions → ready to post. Full reel production with zero manual editing.
4 sub-agentsRead it top-down: the AI General Manager sits at the root; six departments branch off it; each department owns a small set of single-job sub-agents. Twenty-two specialists are far more reliable than one giant “do-everything” bot, because each one has clear inputs, outputs, an approval rule and a KPI.
Shared memory layer: every agent logs to Supabase → so each one can see what the others have already done.
Built & maintained with Claude Code skills
Not everything automates equally well. The bars below show how much of each department can run hands-off today, the build difficulty, and where a human still has to stay in the loop. Green = ship it now, amber = automate with review, terracotta = experimental / keep a human close.
Three departments — social, reviews and reporting — are production-ready and should go live first; they deliver most of the value. Content, SEO and tech are “automate the boring 70%, keep a human on the judgment 30%.” AI-visibility tracking is a nice-to-have signal, not a guarantee — treat it as a thermometer, not a thermostat.
A human equivalent (social manager + SEO + content + review/ops VA) costs ~AED 18,000–35,000/month. The agent stack runs from ~AED 550–2,400/month. Even with a paid build, payback is under one month versus a single hire, and the system works 24/7 without leave or handover risk.
Everything is no-code / low-code. Below: the exact tools, what each does, and a realistic UAE monthly cost. Then three budget tiers — pick the one that matches where EatCookJoy is right now. (USD ≈ AED ÷ 3.67.)
| Tool | What it does | Monthly (AED) | Tier |
|---|---|---|---|
| n8n (self-hosted VPS) | Runs every automation / agent workflow | ~AED 75 | Lean |
| OpenAI GPT-4o API | Copy, captions, replies, schema, summaries | ~AED 110–300 | Lean |
| Supabase | Shared database, auth, storage (the “brain”) | Free → ~AED 95 | Lean |
| Airtable / Google Sheets | Content calendar + approval queue | Free → ~AED 75 | Lean |
| Blotato / UploadPost | One-click multi-platform video posting | ~AED 110 | Lean |
| WhatsApp Business API | Review requests, reminders, GM report to you | ~AED 20–185 + msgs | Lean |
| Google Business Profile API | Read & reply to Google reviews | Free | Lean |
| Looker Studio + GA4 + GSC | Dashboard + analytics + search data | Free | Lean |
| ElevenLabs | AI voiceover for reels (EN + AR) | ~AED 80 | Standard |
| JSON2Video / Wavespeed | Renders the actual reel video | ~AED 90–185 (per render) | Standard |
| SubMagic | Auto captions / subtitles | ~AED 66 | Standard |
| Canva Pro / Predis.ai | Branded static posts | ~AED 48 | Standard |
| Porter Metrics | Social connectors → BigQuery | ~AED 75 | Standard |
| BigQuery | Central data warehouse for KPIs | Free → ~AED 75 | Standard |
| Vercel | Hosts the website (auto-deploy) | Free → ~AED 75 | Standard |
| Semrush / Ahrefs (optional) | Deep SEO audits & rank tracking | ~AED 460–520 | Full |
| Screaming Frog | Crawl audits, broken links, redirects | ~AED 75 (£199/yr) | Full |
| Stripe Connect | Chef split payouts | % per booking | Full |
Follow the 23 sessions over ~6 weeks. Cost = a Claude subscription for the build window, roughly AED 350–750 (~$95–200) total, plus your time. This is the recommended path — you keep full control and the know-how.
A freelancer or agency building the same system in the UAE typically charges AED 30,000–80,000 (~$8,000–22,000) one-time. Still pays back fast versus salaries — but you depend on them for changes.
All data flows nightly into BigQuery, then into a single Looker Studio dashboard, refreshed every 24 hours. Four sections — social, reviews, SEO and bookings — each with a target so you instantly know green from red. Here's the live shape of it:
| Section | Key KPIs | Targets | Source tools |
|---|---|---|---|
| 📣 Social | Reach, impressions, engagement rate, reel views, net-new followers, top post, click-through | Reach +10% MoM · Eng >3% · +50 followers/wk · CTR >1% | Meta / TikTok / LinkedIn APIs → GA4 |
| ⭐ Reviews | Avg rating, review count, response rate, response time, negative rate, request conversion | ≥4.5★ · 100% replied · <4h · <5% negative · >25% convert | Google Business Profile API + Supabase |
| 🔍 SEO/AEO/GEO | Organic clicks, avg position, impressions, LCP, AI visibility, schema & index coverage | Clicks +5% MoM · pos <20 · LCP <2.5s · AI mention >40% · schema 100% | Search Console + PageSpeed + custom log |
| 🍽️ Bookings | Total bookings, conversion, AOV, chef acceptance, onboarding pipeline, retention | Conv >15% · acceptance >85% · retention >40% | Supabase bookings + GA4 |
Live agentic operations · sample shift · 22 agents managing the business
Your AI team works 24/7, never asks for a raise, never calls in sick, and never ghosts you mid-campaign. It also has zero common sense and unshakeable confidence — which is precisely why there is exactly one human approver. The 22 agents handle the ten-thousand boring things; you make the three decisions that actually matter. You're not managing a team of 22 — you're running a kitchen where the prep never sleeps, and you still taste every dish before it leaves the pass.
Automate the busywork — never the judgmentThis is the engineering layer of the document: the core platform, the exact tools and APIs each department runs on, and the Claude Code skill used to build and maintain it. Everything is orchestrated by n8n and shares one Supabase data layer; the website is built and shipped by Claude Code.
| Layer | Tool | Role |
|---|---|---|
| Platform build | Claude Code (Anthropic) | Agentic full-stack builder + skills runner |
| Frontend | Next.js 15 + Tailwind CSS | SSR pages, App Router, SEO-ready |
| Database | Supabase — Postgres + RLS + Realtime + Storage | Shared “brain”, auth, file storage |
| Automation | n8n (self-hosted) | Runs all 22 agent workflows |
| AI / LLM | OpenAI GPT-4o | Copy, replies, schema, summaries |
| Social posting | Blotato · UploadPost · Meta Graph · LinkedIn API | Multi-platform publishing |
| Video | ElevenLabs · Wavespeed · JSON2Video · SubMagic | Voiceover → render → captions |
| Reviews | Google Business Profile API | Read & reply to reviews |
| SEO audit | Screaming Frog · Semrush/Ahrefs · GSC · PageSpeed | Crawl, rank, vitals, indexation |
| Warehouse + BI | BigQuery · Looker Studio · Porter Metrics · GA4 | Unified KPI dashboard |
| Payments | Stripe Connect | Chef split payouts |
| Messaging | WhatsApp Business API | Primary UAE channel |
| Hosting / CI | Vercel · Railway · GitHub Actions | Deploy + test pipeline |
| Agent | Function | Tech stack | Trigger | Build skill |
|---|---|---|---|---|
| SM-1 | Instagram Reels auto-post | n8n · Meta Graph / Blotato · SubMagic · Supabase | Airtable approve | market-social |
| SM-2 | TikTok auto-post | n8n · UploadPost API · Supabase | Airtable approve | market-social |
| SM-3 | LinkedIn brand posts | n8n · OpenAI · Canva API · LinkedIn API | Mon 9am cron | market-social |
| SM-4 | Facebook page updates | n8n · Meta Graph API · Supabase | IG trigger | market-social |
| GR-1 | Auto review request | n8n · WhatsApp Business API · Supabase | booking = completed | market-brand |
| GR-2 | AI reply drafter | n8n · Google Business Profile API · OpenAI | poll every 4h | market-brand |
| GR-3 | Sentiment monitor | n8n · Supabase · OpenAI · WhatsApp | daily 8am | market-brand |
| SEO-1 | Technical SEO auto-fix | Screaming Frog CLI · OpenAI · Next.js / Vercel · Supabase | Sun 6am | seo-geo-aeo |
| SEO-2 | AEO schema updater | OpenAI · Rich Results API · Next.js | bi-weekly | seo-geo-aeo |
| SEO-3 | GEO local citations | GBP / Bing / Apple Maps APIs · n8n | monthly | market-technical |
| SEO-4 | AI visibility tracker | OpenAI / Perplexity APIs · Supabase | daily 7am | seo-geo-aeo |
| TECH-1 | Broken-link checker | Screaming Frog / Ahrefs API · next.config.js · GitHub | daily 6am | github-workflow-automation |
| TECH-2 | Core Web Vitals monitor | PageSpeed API · GitHub Issues | daily 7am | verify |
| TECH-3 | Sitemap & index refresh | Supabase · GSC Indexing API | on new chef live | github-workflow-automation |
| KPI-1 | Social KPI puller | n8n · IG / FB / LinkedIn / TikTok APIs · BigQuery | daily 1am | market-report |
| KPI-2 | SEO KPI puller | n8n · GSC API · PageSpeed API · BigQuery | daily 2am | market-report |
| KPI-3 | Reviews aggregator | n8n · Supabase · BigQuery | daily 3am | market-report |
| KPI-4 | Weekly GM report | n8n · BigQuery · OpenAI · WhatsApp · Supabase | Sun 8pm | market-report |
| CF-1 | Idea generator | n8n · OpenAI · Airtable | Mon 7am | market-social |
| CF-2 | Script & caption agent | OpenAI · Airtable | on idea approval | market-social |
| CF-3 | Reel video renderer | ElevenLabs · Wavespeed · JSON2Video · SubMagic · Supabase Storage | on script approval | remotion-best-practices |
| CF-4 | Visual asset creator | Canva API / Predis.ai · Google Drive · Supabase Storage | calendar schedule | remotion-best-practices |
No jargon. Follow these stages in order. If you only do Stage 1–4, you already have an AI social media manager running. Stages 5–7 add the reviews, dashboard and full GM brain.
Create logins for the core tools: n8n (or n8n Cloud), Supabase, an OpenAI API key, Airtable, and a Blotato/UploadPost account. Connect your Instagram, TikTok, LinkedIn and Facebook business profiles. Apply for the WhatsApp Business API through a provider.
No code~3 hoursOpen Claude Code and run the sessions in order (Week 1: foundation & database; Week 2: booking engine). Each session: type /plan, read the plan, approve, let it build. You're the approver, not the typist.
In Airtable, you get a simple table: topic, status, asset, caption. This is your whole control panel. Each Monday the system drops in 14 reel ideas; you tick “approve” on the ones you like.
Click only10 min/weekRun the Week 3 sessions to switch on SM-1…SM-4 and the reel factory. From now on, approved ideas become finished reels and post themselves on schedule. You now have an AI social media manager.
MilestoneWeek 3Run the review sessions. After every completed booking the system asks for feedback on WhatsApp, sends happy customers to Google, and quietly routes unhappy ones to you for a personal save — before anything goes public.
Week 3–4Run the Week 4–5 sessions: weekly SEO audits, broken-link and speed monitoring, then BigQuery + Looker Studio so every number lands on one screen.
Week 4–5Week 6 hardens security and ships it. After that your week is: read the Sunday WhatsApp report, approve next week's ideas, sign off any negative-review replies. That's it.
Week 6~4 h/week forever| Week | Sessions | You get |
|---|---|---|
| 1 | 1–4 · Foundation | Database, auth, core pages scaffold |
| 2 | 5–7 · Booking engine | Chef search, booking flow, payments, WhatsApp trigger |
| 3 | 8–11 · Social + reviews | All 4 social agents + review request & AI reply |
| 4 | 12–15 · SEO + tech + content | SEO audit, schema, broken links, reel factory |
| 5 | 16–19 · Dashboard | BigQuery, Looker Studio, daily KPI pulls, GM report |
| 6 | 20–23 · Go-live | CI/CD, security hardening, performance, launch |
Six departments. Twenty-two agents. One human — you, approving and steering. Build it in six weeks, run it from AED 550 a month.
Talk to Aziz on WhatsApp Open the full playbookThis playbook is live for EatCookJoy UAE: one branded food reel + one email every day, batch-made and dripped; a cloud agent drafts each morning (07:00 Dubai); SEO/AEO/GEO maintained; and every social post is written, organised by platform, and one tap from publishing. Launching across the UAE · June 2026, inshaAllah.
📋 Post details: every day's caption — tailored per platform — lives in the Social Posts Library, copy-ready with the reel.
Your AI social media manager, agent by agent
This is the department that replaces a full-time social media manager. Four publishing agents share one content factory and one approval queue. You approve ideas once a week; they handle scripting, voiceover, video, captions, scheduling, posting and engagement logging — across four platforms, in English and Arabic.
The four publishing agents
The content factory pipeline (zero manual editing)
Posting schedule (UAE time)
The guardrails that keep it on-brand
Build prompt (paste into Claude Code)
Build the social media automation module:
1. Create /api/social/publish endpoint accepting:
- platform: 'instagram' | 'tiktok' | 'linkedin' | 'facebook'
- content_type: 'reel' | 'static' | 'story'
- asset_url, caption, scheduled_at
2. Connect UploadPost API (IG, TikTok, FB) + LinkedIn API
3. On success → insert into social_posts (post_id, platform, published_at, caption)
4. n8n cron reads the Airtable calendar, calls this endpoint for each
due post, sets the Airtable row status to 'posted'
5. 24h later → fetch likes/views/saves → social_analytics