Skip to content

Mission & Vision

The Mission

Revolutionize software development. Make enterprise-grade custom SaaS accessible to SME business owners at the pricepoint of an intern.

What We Are

Growing Europe is a fully native AI software development agency. Not a tool. Not a chatbot. A self-learning, self-healing multi-agent framework that simulates an entire software company — from account manager to senior developer, from designer to tester — all orchestrated by a single human CEO.

Structure: - 1 human: Dirk-Jan Huizingh — Founder, CEO, Product Owner - 56 registered AI agents (54 active, 2 onboarding) across Claude, OpenAI, and Gemini - Source of truth: ge-ops/master/AGENT-REGISTRY.json

Tagline: Let's grow Europe together

Two-Layer Architecture

GE is built on two layers that work together. Layer 1 is the brain. Layer 2 is the operating model. Combined, they form a self-learning, self-healing multi-agent framework.

Layer 1: The GE Brain (Self-Learning Pipeline)

The brain is how GE gets smarter with every task. It's a MkDocs-based wiki fed by three learning sources:

1. PTY Learning Extraction Every agent session runs through a PTY wrapper that captures full CLI output. After execution, the pipeline extracts lessons: what worked, what failed, what patterns emerged. These learnings are written to the wiki automatically.

  • Struggle detector scores sessions across 5 dimensions (cost, turns, failures, tokens, outcome)
  • Wiki writer generates daily digests and per-agent learning pages
  • See Learning Pipeline

2. JIT Learnings (Just-In-Time) On boot, before an agent starts a task, it queries the wiki brain for learnings specific to that task's domain. "Am I about to work on Redis streams? Let me check what other agents learned about Redis streams." This injects relevant institutional knowledge into the agent's context, preventing repeated mistakes and accelerating progress.

3. Discussion Model (Multi-Agent Consensus) When an agent encounters something with architectural, security, or other deeper impact, they don't decide alone. They open a discussion with other relevant agents. The discussion model supports:

  • Initiation: Any agent can start a discussion on a topic
  • Participation: Relevant agents are invited based on domain expertise
  • Consensus: Voting system to reach majority
  • Escalation: If voting doesn't reach majority → human in the loop (Dirk-Jan)
  • Learning capture: When consensus is reached, the decision becomes a learning — future projects and processes benefit from it automatically

The brain is the wiki. The wiki is the brain. Every PTY capture, every discussion outcome, every extracted pattern flows into MkDocs pages that agents consume on their next boot.

Layer 2: Agency Simulation (Agent Chaining)

This is the operating model — how work actually gets done.

54 agents simulate human roles across a complete software agency:

Function Agents What They Do
Client Relations Margot, Dima, Aimée, Faye Intake, creative direction, scoping, project management
Development (Alpha) Urszula, Floris, Martijn, Boris, Marta, Alex Backend, frontend, iOS, DBA, GitHub, CI/CD
Development (Beta) Maxim, Floor, Valentin, Yoanna, Iwona, Tjitte Mirror team for parallel project capacity
Quality Koen, Eric, Marije, Judith, Antje, Ashley Code review, testing, adversarial testing
Infrastructure Arjan, Thijmen, Leon, Gerco, Otto K8s, deployment, sysadmin, backup
Security Ron, Victoria, Piotr, Pol, Julian, Hugo Guardian, scanning, secrets, pentesting, compliance, identity
Knowledge Annegreet, Eltjo, Mira, Nessa Curation, log analysis, incident command, performance
Orchestration Dolly Routes all work, never calls LLMs

Agents are chained together via Redis Streams with an inbox/outbox system: - TaskService creates work → Redis XADD to triggers.{agent} - Agent picks up work via consumer group → executes via PTY → writes completion - Completions can trigger downstream agents (hook chains, max depth 3) - Work packages with dependencies are enforced via DAG (swimming lanes)

This is not prompt chaining. It's a full company simulation where agents have persistent identities, accumulated learnings, and structured collaboration channels.

What We Deliver

Custom SaaS applications for European businesses. Every project:

  • Hyperscalable: What's built today for 1 user auto-scales to 100,000 users without migration. No prototype-then-migrate. Enterprise architecture from line 1.
  • Enterprise-grade: ISO 27001 and SOC 2 Type II proof. No shortcuts.
  • EU sovereign: Frankfurt primary, Amsterdam DR, Bunny.net CDN. All data stays in the EU. GDPR by design.
  • 12-month guarantee: Same team maintains it. No build-and-disappear.

Client Lifecycle

Client Contact → Intake → Scoping → Contract → Development → Testing → Deployment → Support
    (Margot)     (Dima)   (Aimée)   (DJ+Dolly) (Team Alpha  (Marije)   (Leon)     (12 months)
                                                 or Beta)

See Client Lifecycle for the full workflow.

Value Proposition

For SMEs

  • Production-ready software without the enterprise price tag
  • Same quality guarantees as Fortune 500 projects
  • Custom SaaS at the pricepoint of an intern

For Enterprises

  • Full EU data sovereignty (Frankfurt primary, Amsterdam DR)
  • GDPR compliance by design, not by retrofit
  • ISO 27001, SOC 2 Type II audit readiness
  • WCAG accessibility built in by default

Target Market

European SMEs and enterprises who need production-grade software without building an internal engineering team. Industries: SaaS, e-commerce, healthcare, fintech, logistics.

Differentiation

  • Self-learning: Every project makes the next one faster. PTY extraction + discussions → wiki brain → JIT on next boot
  • Self-healing: Monitoring agents detect failures, discussion model reaches consensus on fixes
  • Hyperscalable: Built for 100k users from line 1, zero migration tax
  • Native AI: Not AI-assisted. Not copilot. The agents ARE the team.
  • European identity: EU-hosted, EU-regulated, EU-governed