The 130-Engineer Machine
Delivery Hero published the number. Now every board is asking for theirs.
The number that mattered in Delivery Hero’s April 24 press release was not 130. It was not the 85 percent success rate, not the 250,000 hours, not the claim about autonomous operation. The number that mattered was 9. Nine percent of all code change requests across a publicly traded company operating in more than 70 countries are now being written, tested, iterated, and submitted by a system that has been adopted by fewer than one in five of the company’s developers. The denominator is the story.
Benjamin Mann, Delivery Hero’s Chief Technology Officer, oversaw the deployment from the company’s Berlin headquarters. Mann had been promoted to CTO in July 2024 after years of building out the engineering organisation. By February 2026, his team had introduced a system called Herogen to its first internal users. By April, the system was autonomously merging more than 100 code changes per day with an 85 percent success rate, measured as the ratio of merged to rejected pull requests. The company did not bury this in an earnings footnote. It published a press release with specific numbers, specific metrics, and a specific claim: annual coding output equivalent to 130 senior engineers.
The claim is worth unpacking at the level of arithmetic. Delivery Hero’s engineering organisation includes roughly 1,500 developers, according to employment tracking data from early 2026. Eighteen percent of those developers have adopted Herogen. That is approximately 270 people working alongside the system. Those 270 developers, augmented by Herogen, are producing 9 percent of all code change requests company-wide. If the relationship between adoption and output holds at even half its current ratio as deployment spreads, full adoption would mean a single AI system generating a quarter or more of all production code at a company with a market capitalisation of roughly six billion euros, listed on the Frankfurt Stock Exchange, delivering food and commerce across four continents.
This is not a research demo. It is not a benchmark on a synthetic coding task. Herogen operates autonomously: product and engineering teams assign tasks in natural language, the system writes code, runs tests, iterates on failures, and submits the result as a pull request proposal. A “council of agents,” multiple large language models from different providers, reviews the code from different perspectives before a human developer performs a final check. The person who once wrote the code now approves the code that a machine wrote and that other machines reviewed.
The probability distribution for what happens next divides into four paths, and the path the market is pricing most confidently is not the most likely one. Forty percent belongs to the world where Herogen becomes the template and every major technology company builds or licenses an equivalent system within 18 months: engineering headcount growth across the sector goes to zero or negative, AI-written code exceeds 30 percent of production output industry-wide by the end of 2027, and tech company margins expand by 10 to 15 percentage points as their largest single cost centre compresses. Twenty-five percent belongs to the world where adoption proceeds but quality concerns slow the pace, after high-profile incidents of AI-generated code causing production outages or security vulnerabilities force companies to layer in expensive human oversight that blunts the headcount savings. Twenty-five percent belongs to the world where autonomous agents expand beyond code writing into product design, project management, quality assurance, and technical documentation, turning the 130-engineer equivalent into something closer to 300, and a new category of AI-native companies emerges with ten-person teams delivering products that previously required a hundred. Ten percent belongs to the world where regulatory intervention imposes restrictions on autonomous AI code generation in critical systems, splitting adoption into a two-speed market that favours incumbents in regulated industries.
If you are a CTO at a Fortune 500 company and you read Delivery Hero’s press release on April 24, the question you are now answering for your board is not whether to build your own Herogen. The question is why you have not already.
The structural insight is in the denominator. Eighteen percent adoption producing 9 percent of output is a ratio that implies the marginal cost of an additional unit of software has fallen by an order of magnitude at one company. The question for the industry is whether this is replicable. The evidence from the first four months of 2026 suggests it is.
Anthropic’s 2026 Agentic Coding Trends Report, published in March, found that 90 percent of professional developers now use AI coding tools and that 41 percent of all code produced in enterprise environments is AI-generated. The AI coding tools market reached $8.5 billion in 2026, according to the same report. But the report also surfaced something the headline numbers obscure: while engineers use AI in roughly 60 percent of their work, they report being able to fully delegate only zero to 20 percent of tasks to the machine. Herogen’s 85 percent merge rate suggests Delivery Hero has crossed a threshold that most of the industry is still approaching.
The crossing matters because of what sits on the other side. The Q1 2026 tech layoff data provides the early signal. According to Challenger, Gray and Christmas, tech employers announced roughly 52,000 job cuts in the first quarter, the highest Q1 total the firm has logged since 2023. Nearly half were attributed to AI automation and restructuring. Oracle’s reduction of 20,000 to 30,000 roles and Amazon’s 16,000-position restructuring are the headline events. But the leading indicator is not who got fired. It is who did not get hired. Companies that previously onboarded cohorts of five to ten junior engineers each quarter are now doing the same work with two or three senior engineers and AI tools, according to industry reporting citing Goldman Sachs compensation data.
Mann’s engineering organisation in Berlin sits at the inflection point of this shift. He did not build Herogen to replace his developers. He built it to redeploy them. The 250,000 hours freed annually are hours his engineers can now spend on architecture, system design, and the kind of strategic technical work that AI cannot yet do autonomously. But the second-order effect is the one that reprices the industry: if Delivery Hero can produce the output of 130 engineers with a software system, the next company entering food delivery or logistics or commerce does not hire those 130 engineers in the first place. The 28.7 million professional software developers worldwide, according to Statista’s 2025 estimate, are not going to be fired. But the next 28.7 million will never be hired.
The implications ripple outward in every direction the reader cares about. Engineering typically represents the single largest cost centre at technology companies, consuming up to 40 percent of operating expenses at mature firms and more at startups. Compressing that cost by even a third would add 10 to 15 percentage points to operating margins across the sector. The venture capital model, which has spent two decades assuming that building software requires raising capital to hire engineers, faces a structural repricing: if three people with AI agents can build what thirty people with keyboards once required, the amount of capital needed to launch a startup drops by half or more, and the mega-round becomes the artefact of a labour-intensive era.
The turn in this story is not that Herogen might fail. It is that Herogen might succeed and the consequences might be stranger than the market assumes. If every company replicates Delivery Hero’s results, the supply of software increases by an order of magnitude while the cost of producing it collapses. Software becomes a commodity input rather than a scarce capability. The companies that win in that world are not the ones that write the best code. They are the ones that know what to build. Product judgment, customer understanding, and strategic clarity become the scarce resources. The engineering department does not disappear. It becomes a utility.
Mann’s council of agents is itself a preview of this future. The multi-model review system that Herogen uses, where different LLMs from different providers examine the code from different angles before the human signs off, has already moved the developer’s role from production to judgment. The human checking the final pull request is not verifying whether the code works. The machines have already done that, across multiple models. The human is checking whether the code should have been written at all. Whether it solves the right problem. Whether it fits the product strategy. The judgment layer is the last layer that remains human. Everything below it, at one company in Berlin, is already automated.
What shifts the probabilities is what happens in the next two earnings cycles.
Watch for Delivery Hero’s H1 2026 results, expected in August. The number to track is not revenue or adjusted EBITDA. It is the Herogen adoption percentage and the share of total code changes. If the 18 percent has become 35 percent and the merge rate has held above 80 percent, Mann’s press release was not a marketing exercise. It was a leading indicator.
Watch for the next major technology company to publish equivalent metrics. Atlassian, which replaced its CTO with two AI-focused co-CTOs in Q1 2026, is the most likely candidate. If a second company publishes comparable autonomous coding numbers before year-end, the thesis moves from anecdote to trend.
Watch for GitHub’s next product announcement, expected at Universe in Q4 2026. GitHub has already introduced Agent HQ, a system for running multiple coding agents simultaneously. If GitHub ships Herogen-equivalent autonomous capabilities for Copilot, the addressable market shifts from companies large enough to build their own systems to every company with a GitHub subscription.
Watch for junior software engineering salary data from Glassdoor and Levels.fyi in Q3. If median compensation for entry-level roles in the United States declines by more than 10 percent year over year, the headcount compression has reached the labour market.
The 130-engineer machine sits in a server room in Berlin, writing code, testing code, and submitting code for review by other machines. Benjamin Mann’s developers review the output. They approve or reject. They no longer write. The number on the pull request dashboard reads 9 percent today. By year-end, the company expects it to read 20. The rest of the industry is doing the arithmetic. The denominator is the same everywhere. Only the numerator varies, and it only moves in one direction.
ANNEX: WHAT HAPPENS TO YOUR ENGINEERING COST STRUCTURE WHEN THE MACHINES WRITE THE CODE?
Delivery Hero published the first public proof point for enterprise-scale autonomous AI coding. Four paths for how the industry responds sum to 100 percent.
Industry Standard Within 18 Months – 40%
If you are running a technology portfolio and you are watching the next four quarterly earnings cycles, this is the scenario that reshapes your operating margin assumptions for every software company you hold. Every major tech firm builds or licenses a Herogen-equivalent system by the end of 2027. Engineering headcount growth goes to zero across the sector. AI-written code exceeds 30 percent of production output industry-wide. Tech company margins expand by 10 to 15 percentage points as the single largest cost centre compresses. The companies that execute fastest gain a durable cost advantage that compounds each quarter. Junior engineering hiring collapses while senior engineers who can orchestrate AI systems command premium compensation. The bifurcation is permanent.
The variable to watch: Delivery Hero’s Herogen adoption rate, currently 18 percent. At H1 2026 results in August, if adoption exceeds 30 percent with merge rates above 80 percent, this scenario gains probability. At 3 months, look for at least one additional public company disclosing equivalent autonomous coding metrics. At 12 months, the confirmation signal is AI-generated code exceeding 30 percent of production output at three or more public companies, tracked through earnings call disclosures and developer survey data from Anthropic and GitHub.
Gradual Adoption With Quality Drag – 25%
If you are evaluating AI coding vendors for your engineering organisation and you are deciding how much autonomy to grant the system, this is the scenario where caution pays. One or more high-profile incidents of AI-generated code causing production outages, data breaches, or security vulnerabilities force the industry to add expensive human oversight layers. Adoption reaches 40 to 50 percent of major tech companies by end of 2027, but the headcount savings are smaller than the optimists project because every autonomous pull request requires deeper human review than Delivery Hero’s current model suggests. The 130-engineer headline proves optimistic when review overhead is fully accounted for. Merge rates at scale settle closer to 65 percent than 85 percent.
The variable to watch: major production incidents attributed to AI-generated code, tracked through public incident reports, CVE databases, and post-mortems published by affected companies. At 1 month, if a Tier 1 tech company discloses an AI-code-related outage, this scenario gains 5 to 8 percentage points from the base case. At 3 months, if no incidents have surfaced, this scenario loses probability. At 12 months, the industry-wide average merge rate for autonomous coding agents, tracked through GitHub and Anthropic developer surveys, is the definitive metric: above 80 percent confirms the base case, below 70 percent confirms this path.
Acceleration Beyond Code – 25%
If you are a venture capital partner underwriting the next generation of software startups, this is the scenario that breaks your model. Autonomous agents expand from code writing into product design, project management, quality assurance, and technical documentation. The 130-engineer equivalent becomes a 300-engineer equivalent as agents handle the full software delivery lifecycle. A new category of AI-native companies emerges with 10-person teams delivering products that previously required 100. The capital required to launch a software startup drops by 60 to 80 percent. Mega-rounds become artefacts. Seed-stage valuations compress because the burn rate to reach product-market fit halves.
The variable to watch: the number of startups reaching one million dollars in annual recurring revenue with fewer than 10 employees, tracked through Y Combinator batch data and Carta cap table analytics. At 3 months, if GitHub ships autonomous capabilities beyond code at Universe, this scenario gains 5 percentage points. At 12 months, if more than 15 percent of the next Y Combinator batch has fewer than 5 full-time employees, the structural shift is underway. Also track Delivery Hero’s own roadmap disclosures: if Herogen expands into QA and documentation by Q4, the template extends beyond coding.
Regulatory Intervention – 10%
If you are a policy advisor tracking AI governance and you are drafting rules for autonomous systems in production, this is the scenario where the labour market forces the regulator’s hand. Labour unions, government bodies, or industry standards organisations impose restrictions on AI-generated code in production for regulated industries. Liability frameworks for autonomous code are introduced in the EU or individual US states. Adoption splits into a two-speed market: consumer technology and unregulated sectors accelerate, while healthcare, finance, defence, and critical infrastructure maintain human-written code requirements. The compliance cost of operating in both worlds favours large incumbents over startups in regulated sectors.
The variable to watch: legislative proposals in the EU AI Act implementation process or US state-level AI liability bills that specifically address autonomous code generation in production systems. At 3 months, if the European Commission’s AI Office issues guidance on autonomous coding agents, this scenario gains probability. At 6 months, track the US Senate Commerce Committee hearing schedule for AI workforce impact sessions. At 12 months, if any G7 country enacts specific restrictions on autonomous code in critical infrastructure, this scenario doubles to 20 percent with corresponding reductions across all other paths.
Sources:
Delivery Hero, “Delivery Hero Unveils Herogen, Autonomous AI Agent Unlocks 130-Person Engineering Output,” April 24, 2026.
Delivery Hero, “Delivery Hero promotes Benjamin Mann to Chief Technology Officer,” July 3, 2024.
Anthropic, “2026 Agentic Coding Trends Report,” March 2026.
Challenger, Gray and Christmas, Q1 2026 tech sector layoff data, April 2026.
Statista, global software developer population estimate, 2025.
Glassdoor, senior software engineer salary data, United States, April 2026.
CompaniesMarketCap.com, Delivery Hero SE market capitalisation data, 2026.
Unify, Delivery Hero employee headcount and department breakdown, February 2026.
Disclaimer: This report is published by Scenarica Intelligence for informational purposes only. It does not constitute investment advice, a solicitation to buy or sell any financial instrument, or a recommendation regarding any particular investment strategy. Scenarica Intelligence is not a registered investment adviser or broker-dealer. All scenario probabilities and assessments represent the analytical judgment of Scenarica Intelligence and are subject to change without notice. Past performance of any asset or strategy discussed does not guarantee future results. Readers should conduct their own due diligence and consult with qualified financial advisers before making investment decisions.
Scenarica Premium: The full Scenarica suite includes Geopolitics, Economy, Bitcoin, AI, and Sunday Edition.
Scenarica Intelligence
We don’t predict the future. We price it.