1,341 documents covering every operational decision in our business. 167 executable procedures any AI can run. 44 integrated tools the brain reads and writes to.
One operating layer that turns legal policies, financial data, operational processes and company rules into something LLMs can actually use to execute.
Now every employee runs on AI. Seven agents run operations on the same brain.
€352/month. 18:1 ROI. Six months live in our 8-figure consumer brand.
For every €1 a brand spends on software, it spends €6 on services and headcount to operate that software. Shopify costs €2K/year. The people managing inventory, processing orders, answering tickets, and closing the books cost €200K+.
Most "AI agents" shipped in 2026 are chatbots that forget. Tools that break when models update. Demos that never make it to production. Static playbooks that are outdated the day they're published. Prompt libraries that break on the next model release.
After eighteen months of trying tools, hiring agencies, and stitching together automations, we stopped. We rebuilt the operation around one principle: every recurring decision should be made by an agent with memory, context, and accountability. This site is what came out of that.
Eighteen months ago, every recurring decision in our business was made by someone remembering something. The refund policy lived in three different Slack messages and a Notion page from 2024. Pricing exceptions were tribal knowledge. The incident playbook was the founder's memory. How customer service responded to a delayed shipment depended on which agent picked up the ticket.
We extracted all of it.
Pulled from email, Slack, support tickets, past reports, meeting notes, the founder's memory, and 18 months of operational decisions. Structured. Versioned. Searchable. Versioned again every week.
Procedures the brain can run on demand. Close the books. Triage a refund. Run weekly P&L. Reconcile last week's bank movements. Each one documented, parameterized, and callable from any agent or any employee with a Claude window open.
Every business system the brain can read or write — Shopify, Klaviyo, Notion, Slack, GA4, the accounting stack, the helpdesk, the warehouse, the bank. Exposed via MCP. Governed via ACL. Audited via logs.
Every operational decision adds to the brain. Every conversation that mattered. Every fix that worked. Every pattern that emerged. The brain at month six is qualitatively different from the brain at month one — and that gap keeps widening.
Y Combinator's Tom Blomfield recently called this "the missing layer between company data and reliable AI automation. Every company in the world is going to need one." We just call it the brain.
Once the brain existed, the way we work changed. Not because we asked the team to "use AI more." Because suddenly there was something worth using.
Opens Claude. "What changed last week and what should I look at first?" The brain returns the weekly synthesis — revenue, three escalated tickets, an inventory imbalance forming on the new collection, marketing fatigue on the welcome flow. Six minutes of review. No tabs opened.
A customer asks why their refund hasn't processed. "Status of order 8341, refund timeline, who promised what." The brain returns the full thread, the policy that applies, what's been said, what to say next. Reply drafted in 90 seconds. No checking five systems.
"Reconcile last week's bank movements against expected payments. Flag anomalies." The brain runs through accounting + bank + Shopify, generates the variance list, explains each anomaly. What used to be Saturday morning work is now Friday afternoon.
"Sell-through by size for the current collection, versus the same week last year, flag any size at risk of stockout or markdown." Brain pulls Shopify + warehouse + wholesale pipeline. Decision made before lunch. No spreadsheet, no analyst.
This is the difference between AI as a feature and AI as a foundation. When the brain holds the context, the model becomes useful for everyone — not just the technical team.
Once the brain existed, automation became cheap. The same context that powers human work also powers seven AI agents running 24/7 across the operation. They don't replace employees — they handle the recurring decisions nobody wants to make at 3 AM. Each agent reads from the brain, writes to the brain, and contributes patterns the brain keeps.
Morning briefings, cross-domain synthesis, competitive scans, knowledge mining, agent coordination.
Ticket triage, WISMO responses, drafts with brand voice, pattern detection across tickets, escalation routing.
Weekly P&L, AR follow-ups, invoice reconciliation, multi-currency treasury, variance alerts.
Daily store reports, foot-traffic-driven staffing, inventory transfer flags, store A vs store B comparisons.
Campaign analysis, segmentation, subject line patterns, SEO opportunity mining, attribution tracking.
Sell-through analysis, inventory distribution by variant audits, markdown candidates, pricing positioning, wholesale ops.
Absence reports, payroll prep, vacation balances, onboarding, expense categorization.
Claude Code with all 44 MCP tools. The founder's direct interface to every agent and every system in the swarm.
Most AI vendors quote numbers they can't defend. "10× productivity." "50× return." "Pays for itself in a week." Then you ask how they calculated it and the conversation gets vague. Here's exactly how we get to 18:1.
| CS Agent — triage, drafts, policy lookups | 20h |
| Founder time freed — briefings & decisions | 10h |
| Finance — P&L, AR, reconciliation | 8h |
| Merchandising — sell-through & sizes | 6h |
| Retail — daily reports & staffing | 5h |
| Marketing — campaigns & segments | 5h |
| Strategy — synthesis & knowledge | 5h |
| HR — admin & payroll prep | 3h |
| Total hours offloaded weekly | 62h |
€21/h loaded labor derived from personnel budget ÷ headcount ÷ ~2,000 working hours × role mix. Conservative. No revenue impact claimed — every number here is a cost avoided.
We run our brand on this system. Every Monday the agents push the week's learnings to a shared brain. Every Friday we review what changed and decide what gets promoted into the playbook. You're not reading AI theory. You're reading the operating manual of a brand that is, right now, processing tickets, closing books, and reordering stock.
This is not a course you buy and forget. It is operating documentation backed by a production system. Updates happen when there is something real to ship, not to simulate motion.
Production activity feed, honest ROI breakdown, real cost data, and a governed public snapshot of the swarm. No mockups, no vanity KPI theater, and no hidden spreadsheet math.
Most AI demos show one capability at a time. A system running for six months accumulates depth: corner cases, weird customer behaviors, recovery patterns, decisions that are too specific for a sales deck but too valuable to forget.
Self-evolving prompts. The CS agent measures its own response quality, mutates underperforming prompts, and auto-promotes improvements. 94.7% accuracy after 3 months of evolution.
Six domain-expert agents + blind peer review for high-stakes decisions. Strategy, finance, retail, CS, marketing, merch. ~€1 per deliberation. 2-4 minute response time.
Cross-company knowledge library with strict anonymization. New deployments start at 70%+ autonomy instead of zero. REST API, live. The real moat.
Inbox → OCR → classify → rename → file in Drive → reconcile against POs → log in master sheet → route for approval. 5 minutes per invoice drops to <30 seconds.
Real-time CM3 per product across representative commerce, inventory, ads, analytics, and finance sources. Shopify, Stockagile, and Holded are typical examples. Every product has a live CM3 number.
1,114 email campaigns analyzed for subject line, body, CTA, and performance patterns. Learned rules like "ALL CAPS = 2.7× revenue (if <15% of sends)" drive future drafts.
Contract-based cascading execution for multi-step operations. Each step has acceptance criteria. If any step fails, the entire operation rolls back atomically.
Tracks brand visibility in ChatGPT, Perplexity, Claude, and Google AI Overviews. Reverse-engineers cited content and optimizes. Mention rate improved from 35% to 60%.
The strategy hub has its own Visa corporate card with a monthly limit. Autonomous expense management on approved categories. Receipts auto-filed.
Most AI deployments start at zero. The first month goes into calibrating prompts, tuning confidence thresholds, debugging edge cases nobody warned you about. We're keeping the patterns we've built and sharing them. Anything that follows starts where we left off — not at zero.
3 months to reach 91% autonomy. Learns everything from scratch.
Starts at 70% autonomy day one. Pattern library pre-loaded.
Starts at 85% autonomy with industry-specific patterns built in.
Every deployment makes every other deployment smarter.
The category is noisy. Most "AI for business" products are one of these things. We're not any of them.
compAI is the operating manual for running a brand on AI agents. Six months in production across CS, finance, merchandising, retail, marketing, and people ops. 32 documented lessons. Honest 18:1 ROI. Every claim auditable. Read it, run it, or have us wire it in.
Not a marketing site about AI. The actual playbook, the actual skills, the actual prompts, the actual templates. Each one written and validated inside an 8-figure consumer brand operating on AI for over a year.
The brain primitive. The MCP server. The agents. The advanced capabilities. The production lessons. The honest open questions. Read it online or fork the markdown.
Executable procedures any AI client can run. Close the books. Triage a refund. Run weekly P&L. Reconcile bank. Each callable from a Claude window with the brain mounted.
The prompts behind every agent: customer service, finance, marketing, retail, merchandising, people ops, strategy hub. Categorized, anonymized, ready to adapt.
How to wire your stack — Shopify, Klaviyo, Notion, the helpdesk, the accounting tool, GA4 — into the brain via MCP. Three-step guides with sample code.
The folder structure of the brain. The agent SOUL.md template. The factory.yml. The skill.md template. The custom-instruction template. Drop-in starting points.
Anonymized operational patterns from a year of running this. New deployments inherit instead of starting from scratch. YAML schema + REST API.
Daily DTC contribution-margin calculator, validated to within 0.4% of the monthly accounting close. The spec, the bug fixes, the dashboard. All open.
What broke and what we did. Token leaks. Webhook downtime. Memory degradation. Provider rate limits. Each one with a one-rule lesson others can use.
License: Free to use, attribution requested. Updated as production learnings are verified.
Everything we built is in a public repo: the playbook, every executable skill, the prompts library, the integration guides, the templates. Free to use, attribution requested. The kind of resource we wish had existed when we started.
License: Free to use, attribution requested. Need hands-on help wiring this into your business? hello@usecompai.com — we work with up to three brands at a time.
Open playbook. 33+ chapters. 146 skills. 21 patterns. 32 production lessons. Live dashboard. Every artifact downloadable. Updated as production learnings are verified.
Free to use, attribution requested · Questions? hello@usecompai.com
