Startup Profile
AthenaRun

Introduce yourselves

Explain in three sentences what AthenaRun is. How long has AthenaRun been in existence?

AthenaRun is a platform that covers the entire software lifecycle using AI agents, from requirements right through to the running, operational application. Business departments describe what a system needs to be able to do; the platform checks, researches, builds and operates it, with human approval at every critical stage. Its memory is the Vault: it stores tried-and-tested patterns and known errors so that every new project learns from the previous one.

How did the idea come about and what challenges did you have to overcome?

The idea stems from a phrase we heard time and again in companies: ‘The process has no memory.’ Requirements are lost when colleagues leave, approvals end up in shared inboxes, auditors cannot trace decision-making chains, and IT cannot keep up with development. The biggest challenge was not generating code, but making it reliable: every application must be tested, traceable and operational; otherwise, it cannot be used in small and medium-sized enterprises or in regulated sectors.

How does AthenaRun work, and how does AI support this project?

It works in two stages. First stage: An idea or requirement is turned into a vetted package comprising three documents – what the platform covers, a research report with sources, and a compliance report. This takes place before any development budget is allocated. Second half: The package is turned into running software, including code reviews, security checks for each release, operations, ticket handling and documentation. AI agents handle the work in both halves; humans give final approval.

What happens if your product cannot solve a task on its own?

Then it stops and does not offer any further advice. The pipeline is deliberately designed so that a human makes the decision at defined points: an agent that gets stuck submits the status along with its reasoning as a ticket, and a developer or the relevant department makes the decision. Nothing goes into production without approval. We consider this to be the most important difference compared to the ‘prompt in, app out’ approach.

How does the autonomous generation of applications work technically behind the scenes?

The platform does not operate in a single step, but in many small ones: each building block is automatically checked before the next one is created, and the Vault provides templates that have proven their worth in previous projects. Exactly how these checks interlock is at the heart of what we do – and we’d rather demonstrate this live than explain it in writing. Operations are based in Germany, using local models where necessary and without training on customer data.

Let’s take a look into the future: where might the project be in five years’ time?

Software is no longer built, but ordered. A business unit describes what a system needs to be able to do; by the evening, it’s up and running, tested, documented and operational. The IT department no longer writes code; it sets guidelines and approves the work. For small and medium-sized enterprises, this means they have the same software capabilities as a corporation with a thousand developers – only without the thousand developers. Legacy systems, which nobody wants to touch today, will be rebuilt piece by piece alongside the core by the system itself. And AthenaRun is where this knowledge resides: the Vault will then contain not hundreds, but millions of tried-and-tested patterns, from every project that has ever run on it. Our ambition: to be the platform on which European SMEs run their software – in Europe, under their own control.

What is your biggest point of debate within the team?

How much autonomy we give the agents. Our technical instinct says: more. The customer’s perspective says: every approval we take away costs trust. We argue about this with almost every new feature and usually end up with a ‘human-gated’ approach.

How do you deal with your biggest critics?

We take them seriously and demonstrate our approach. We address the harshest criticism – that AI-generated code is error-prone and unmanageable – ourselves before anyone raises it, and respond by highlighting the verification chain and human approval. And we don’t show slides; instead, we demonstrate the platform live: 30 minutes, a real-world requirement, and real-time results.