Cyberport’s journey to becoming an AI-first company

01 / 03

Starting point

In a highly competitive market, standard SaaS solutions no longer provided Cyberport with the flexibility required for true differentiation and development speed.

To radically boost innovation, Etribes designed a measurable AI-first Software Factory. The target path is to cut time to live per feature from eight weeks to three days. The result is a framework of nine highly specialized, autonomous agents that produce code behind deterministic quality gates. As a proof of concept, the factory ran on the price comparison feature while the live shop kept running.

The focus was on establishing an organizational learning system. AI agents handle software development, while humans steer and approve at defined checkpoints.

02 / 03

Approach

The five-phase project approach

1. 1

   We started in internal software development as the most controllable training ground for the entire organization. In collaborative workshops, we designed a framework with nine focused, autonomous agents that take on specialized task areas such as security or coding.

2. 2

   To ensure full transparency, every agent leaves behind a verifiable work log in the form of standardized documents after completing its task. No agent finishes its work unchecked, which ensures clean documentation throughout the entire development process.

3. 3

   Process logic and approvals were decoupled from unpredictable AI models and moved directly into the code. These deterministic quality gates make iterations reproducible and reliably auditable for enterprise requirements.

4. 4

   Final responsibility always remains human, which is why clear human-in-the-loop processes were established for critical decisions like code reviews. Through scheduled retrospectives, the system continuously learns from human feedback.

5. 5

   To guarantee scalability and make business impact measurable, we implemented specific factory metrics. We track KPIs such as lead time and token costs to systematically manage long-term business value.

03 / 03

Result

Cyberport faced the strategic challenge of regaining its competitive edge through tailored technological solutions rather than being limited by off-the-shelf software. To achieve this, an intensive AI engineering sprint laid the foundation for future-proof, AI-powered software development.

Internal software development served as the ideal starting point for gradually aligning the entire organization with an AI-first approach. Instead of relying on isolated AI support, we designed a full-scale, agent-based AI infrastructure in collaborative workshops. This resulted in a framework of nine specialized, autonomous agents, including agents for architecture review, solution design, security, and coding, that solve complex tasks collaboratively.

The focus was on three core areas. These were specialization to tame unpredictable AI models, building trust through verifiable work logs, and control via deterministic quality gates in the code. To ensure consistently high code quality, every agent generates standardized documents for complete documentation. The agents work in the background on existing systems like Jira or GitHub, while developers provide critical approvals through human-in-the-loop processes and continue to train the system through ongoing feedback.

To objectively evaluate developer productivity, tailored factory metrics (such as lead time per ticket, token costs, and gate results) were implemented. This blueprint enables Cyberport to sustainably increase development speed and position itself as a long-term technological leader in the market.

For detailed project insights (K5 Masterclass)