Tim Sweeney, head of Epic Games, is fundamentally restructuring his company's game development and aligning it with AI tools. His expectation: a golden age for computer games. The report does not provide more details – and that is precisely why it is worth looking less at the announcement itself than at the question of what such a restructuring practically triggers in a development organisation.
Because the move is not an isolated case. The fact that an established software manufacturer is not selectively supplementing its production method but fundamentally re-cutting it is the actual event. For small and medium-sized enterprises that operate web platforms, shops or TYPO3 installations, the same question arises on a smaller scale: where does AI change only the tool – and where does it change the division of labour?
From aid to component of the process
The first stage of AI use in development is usually inconspicuous: developers have code suggestions generated, design tests or have error messages explained. This remains individual, voluntary and without impact on the project flow.
The second stage looks different. Here, AI agents – i.e. programs that independently execute several work steps in succession – are firmly integrated into the development process: they generate pull requests, i.e. change proposals to the source code, update dependencies or migrate code between versions. From this point on, AI is no longer a personal work facilitation but a production factor that must be planned, monitored and accounted for.
Anyone who does not consciously shape this transition gets it unplanned: in many teams, the first stage has long been everyday practice without anyone having regulated what of it may go into productive code.
Roles shift, they do not disappear
When code is created faster, it is not writing that becomes the bottleneck but reviewing. Review, architecture decisions and understanding of the technical requirements gain weight. At the same time, the need for people who can precisely describe what is to be built increases – an imprecise requirement leads faster to a lot of unusable result with AI support than without.
For project managers, this means concretely:
- Senior competence becomes more important, not superfluous – someone must be able to evaluate generated code.
- Entry-level tasks change; classic routine work as a learning field for juniors partly falls away and must be replaced.
- Technical specification moves closer to development because it flows directly into prompts and agent orders.
- Responsibility for the result remains with the team, regardless of who or what generated the code.
Quality assurance becomes the bottleneck
AI-generated code often looks plausible. That is precisely the risk: errors are less often obvious but subtle – an incorrectly handled special rule, an unchecked input, an access without permission check. In TYPO3 projects, typical pitfalls are added, such as the correct handling of caching, of multilingual content or of extensions whose interfaces change between major versions.
What carries in this situation is no longer control in retrospect but a robust technical foundation:
- Automated tests that map technical rules and do not only check syntax.
- Static code analysis and uniform coding standards as a hard barrier in the pipeline.
- Mandatory review by humans for everything that goes into the productive branch.
- Traceability: document which changes were created by machine.
- Clear rules on which data and which source code may be handed over to external AI services.
The last point is not a formality. Anyone who processes customer data, internal business logic or licensed third-party code needs a documented decision on which tools are permissible – also with regard to data protection and copyright.
Why speed alone is not a goal
Speed gains in development fizzle out if the rest of the organisation does not keep up. If coordination, approvals or content maintenance still take weeks, faster code changes little about the delivery date. The lever therefore lies less often in the tool than in the overall process: in short decision paths, cleanly demarcated responsibilities and in deployments that work without manual intermediate steps.
In addition, there is a point that rarely appears in restructuring announcements: every automation creates new maintenance effort. Agents need configuration, access, guardrails and regular review. Anyone who does not plan for this exchanges manual work for a tool chain that is difficult to see through.
A realistic entry
A radical restructuring like that of a large game developer is neither necessary nor sensible for most companies. More sustainable is a limited entry: a clearly defined use case, such as support with a TYPO3 version migration, with test coverage or with recurring maintenance tasks. Afterwards, it is measured whether throughput time and error rate actually improve – and only then expanded.
It is important to define the evaluation beforehand. Without defined metrics, the assessment remains a feeling, and feelings regularly turn out too positive with new tools.
Classification
For web and software projects, the focus shifts from code creation to specification, review and automated quality assurance. Anyone who wants to use AI productively therefore first needs a clean technical basis: tests, pipelines, version states and clear responsibilities.
The announcement from the games industry is above all an occasion to review one's own processes – not a reason to restructure the development organisation at short notice.