Project Centre: the control room of our AI agent team
Project Centre is our own dashboard for every project we run. A team of AI agents — developer, tester, designer, marketer, security guard, SEO specialist and more — works on each project in parallel, while one person checks, approves and stays accountable. Every project on this site, including the site itself, is built and maintained from it.
- Problem
- One person running several live products can't write, test, design and market everything alone. And AI must not act on live systems unchecked.
- Solution
- A self-built panel where 12+ AI specialists work in parallel on each project. A lead checks the results and sends weak work back; the business lead writes reports with suggestions; the owner only approves. Risky actions wait for a human “yes” with a plain-language explanation. A pause switch stops everything at once to keep costs under control.
- Result
- Every project on this site is built and maintained from it — including this website. That is how we work faster and keep prices lower.
Key facts
- 12+ specialist AI agents per project, each project with its own rules, memory and brand book
- Risky actions — live server, e-mails, deleting, publishing — wait for the owner's “yes”, explained in plain words
- A pause switch stops all agents at once
- Morning overview: every project as a card (website, server, git, open tasks) plus a business lead report with suggestions
- Access from a phone only through a private Tailscale network with a PIN; 5 wrong PINs lock it for 10 minutes
- The engine runs on Node.js without extra libraries and uses a Claude subscription, not per-request API billing
- Node.js
- Vanilla JS
- CSS animations
- Claude Code
- MCP
- Playwright
- Tailscale
How an AI agent team builds and checks software
Each project has its own specialists: developer, tester, bug hunter, deployer, security guard, data master, designer, marketer, SEO specialist and project-specific roles. They work in parallel, not one after another, and a lead checks every result and sends weak work back.
- Specialists work in parallel
- Every change is checked in the browser and on a 390 px phone screen
- Security review of payments, logins, personal data and keys before launch
- Each project keeps its own rules, memory and brand book
Human approval for every risky action
When an agent wants to touch a live server, send an e-mail, delete or publish something, it asks first. The owner sees a plain-language explanation and answers yes or no. A pause switch stops all agents at once.
- Permission requests with a sound and a flashing tab
- A plain-language explanation before every risky step
- One switch to pause the whole team
- Private access from a phone through Tailscale with a PIN
What this means for your project
The routine work — writing, testing, checking — is done in parallel, so projects move faster and prices stay lower. Responsibility stays with a person: you talk to the developer, not a bot.
- Faster delivery without cutting checks
- Prices published before the first call
- One accountable person for your project
- The same approval principle can run in your own systems
Questions
If AI agents write the code, who is responsible?
A person. The agents do the routine work in parallel — writing, testing, checking on a phone screen — and every result is reviewed. Anything risky on a live system waits for a human “yes” with a plain-language explanation. You talk to the developer, not a bot, and the developer answers for the result.
Does working with AI agents lower the quality?
Every result is checked, and weak work goes back. A tester agent proves each change works in the browser and on a 390 px phone screen, a security agent reviews payments, logins and personal data before launch, and the owner approves anything risky on a live system before it happens.
Can you build something like this for my business?
Yes, on the scale your business needs. The same principle — AI prepares, a person approves — runs in Inventory, where AI office assistants draft invoices and replies through the system's API. We usually start with one repeating job as an AI integration and add more once it proves itself.
See what the team could build for you
Tell us what slows your business down. We reply within 1 working day with a concrete suggestion and a price range.
Free consultationRelated services and work
- ServicesAI integrationAdd AI to your website, shop or system for one repeating job: answers, descriptions, social posts. Human approval, usage limits, EU servers. From €1,500.Learn more
- ServicesBusiness systems and automationStock, orders, registrations and invoices in one system instead of Excel and copy-paste. Built around how you already work, hosted in the EU. From €5,000.Learn more
- Case studiesInventoryA warehouse and production system for a group of two companies: orders, invoices, FedEx labels, WooCommerce orders and AI assistants with owner approval.Read the case study
- Case studiesBlusa.euHow we built a second-hand marketplace with Stripe Connect payments, AI listings from photos, a flea market organiser tool and automated social posts.Read the case study
Let's talk about your business
Tell us what slows you down. We'll reply with a concrete suggestion and a price range — free of charge.
hello@neirolabs.comRiga, UTC+2/+3 · Prices in EUR · card or bank transfer