Delivery Hero: Autonomous AI Agent Herogen Does the Work of 130 Engineers

Delivery Hero, the German food-delivery giant operating in dozens of countries, has announced the rollout of Herogen, an autonomous coding agent that independently writes, tests and fixes code based on natural-language instructions. The company says the tool now performs well enough to be credited with productivity comparable to the work of 130 senior engineers per year.

How Herogen Works

According to Delivery Hero, engineers assign tasks to Herogen simply by describing them in natural language. The agent writes the code itself, runs tests, fixes bugs, and only submits a finished proposal for review once it's ready. The company says this sets it apart from earlier generations of coding assistants, which required constant supervision and repeated corrections at every stage of the work.

Before a change reaches a human, it passes through an internal quality-control mechanism Delivery Hero calls the "agent council." This is a set of models from different LLM providers that evaluate the proposal from different angles before an engineer makes the final call. The company says most tasks require zero or one human intervention, and the system has built-in safeguards allowing the agent's work to be manually halted.

Scale of the Rollout

Herogen was first introduced to Delivery Hero's product and engineering teams in February, with the full announcement following a few months later. The rollout currently covers 18 percent of engineering teams, and the agent already accounts for 9 percent of all code changes across the organization. The company says it wants to raise that share to 20 percent by the end of the year.

The figure of 130 engineers that Delivery Hero cites as the annual productivity equivalent of Herogen refers to the work of senior developers and comes from the company's internal calculations, not an independent audit. Likewise, the estimate of 250,000 hours saved annually is a Delivery Hero claim, not an externally verified metric.

Voices from the Company

Behind the project is Delivery Hero's Chief Technology Officer Benjamin Mann, who argues the agent lets the team deliver more without growing headcount. Rodrigue Schäfer, VP of Platform, speaks of a shift in the very nature of developers' work, from writing code in complex languages to formulating tasks in natural language.

Herogen lets us deliver dramatically higher output with the same elite team - Benjamin Mann, CTO, Delivery Hero

It's a shift from complex programming languages to natural language, which lets engineers focus on creative exploration and complex problem-solving - Rodrigue Schäfer, VP Platform, Delivery Hero

What This Means for the Industry

Delivery Hero's case fits into a broader trend in which large tech companies are testing autonomous coding agents not as an add-on to a developer's work, but as a standalone source of team capacity. Unlike popular assistants such as Copilot or Cursor, which suggest code snippets while a human works, Herogen is meant to carry out entire tasks from start to finish with minimal human involvement.

For Polish software companies and IT departments, this case signals that pressure to measure engineering team efficiency in terms of "full-time-equivalent headcount" replaced by agents will keep growing, regardless of whether the specific numbers can be fully verified. At the same time, the stated 85 percent pull request acceptance rate shows that even with a high degree of system autonomy, a substantial share of code proposals still gets discarded or requires fixes.

Delivery Hero has not disclosed exactly which base models Herogen is built on, beyond a general statement that it uses "leading" large language models. The company also did not say whether the tool will be offered to other businesses or remain a purely internal solution.