Delivery and Nearshoring

Extensions to developer IDEs using LLMs to assist with coding tasks, increasingly automated into agentic workflows, are changing the way software is developed. SAP and ABAP as a language are not leading this development, but they are moving now as well.

Once requirements have passed through the tactical layer, and the development is sufficiently specified, the effort needed for development, unit testing and functional testing will continue to shrink rapidly.

Where specifications are not specific enough, AI coding assistants will find the solution that is most likely to be desired given the context. The time to present a first prototype to the business may shrink from days to hours.

This is both an opportunity and a risk.

Avoiding scope creep has been discussed as a key challenge on the tactical layer, and it will remain one of the challenges for consultants and developers at this level to differentiate clarifications and bug fixing from scope creep, and delegate the latter back to the tactical layer.

Should AI-assisted delivery be left on its own without qualified human supervision, severe scope creep will be likely if the AI is not provided with very strong and intelligent guardrails.

On the positive side, AI will also be able to produce structured documentation at this level, so AIs and humans working on maintenance will be able to reference the why and the how of a development.

This is not a minor benefit. In many projects, documentation is the first thing to suffer under time pressure, although it is exactly what the next project phase, the support team and future AI systems will urgently need.

The development tooling around this is moving quickly. In the wider software world, VS Code-based environments, coding assistants and increasingly agentic developer workflows are already changing daily work. In SAP, ABAP Development Tools in Eclipse remain central for many development tasks, while VS Code-based workflows and SAP’s own AI capabilities are developing as well.

SAP’s Joule for developers capabilities seem to point in the right direction. Personally, we have not tested it, but have been excited with the results of testing Marian Zeis‘ MCP servers for ADT (https://github.com/marianfoo/abap-mcp-server) and abap docs (https://github.com/marianfoo/abap-docs) together on our accelet.ai platform.

AI-assisted development is not only about the model writing code. It is about the model being able to access the right context and the right tools. In the SAP world, this includes documentation, system metadata, custom code, transport context, test results and, where appropriate, live system information. There is still a long way to go, but we believe that more automated development is around the corner.

Our accelet.ai platform adds further capabilities around SAP context, documentation, custom-code analysis and support processes, and it gives us the freedom to integrate new tools and features that help us in our daily work.

For consultants and developers, it will bei important to control the development process and make sure that implicit assumptions are clarified and made explicit to the AI, that unit tests and functional tests are meaningful, and all client processes are adhered to. This discipline will become more important the more the the amount of generated or AI-assisted work increases.

A severe risk is that if the restraint on requirements on the tactical layer breaks down, custom developments might become so plentiful and complex that total cost of ownership becomes higher than before.

During implementation, huge repositories of custom development might cause high AI context and token costs. The same may apply to maintenance. This may sound strange at first, because implementation itself becomes cheaper. But custom software does not only have to be built. It has to be understood, tested, documented, monitored, changed and maintained. If AI makes it much easier to create custom developments, but the organization does not become better at deciding which developments should exist in the first place, the result may be more complexity, not less.

At acceletail, this is also where we see a strong role for AI-supported nearshore delivery with our partner Arteis, who provide us with excellent SAP retail consulting and development skills.

Nearshore delivery has always depended on the quality of communication, documentation and handover. AI improves exactly these points. It helps with translation, meeting summaries, specification quality, test preparation, documentation and shared access to context.

The local consultant still has to understand the customer, the requirement, the SAP standard and the consequences of a solution. The nearshore delivery team still has to implement carefully, test properly and document what was done. But AI can reduce friction between both sides, make context easier to share, and help ensure that delivery remains connected to the agreed requirement.

The aim is to accelerate delivery where the requirement is clear enough, while keeping the context, documentation and review process strong enough that the result can still be maintained later.

Faster delivery will not remove the need for discipline. It will increase it. The unclear requirement, the small addition, the misunderstood clarification and the prototype that quietly becomes productive software will all have to be watched carefully.

AI will be a powerful tool on this layer. But it should remain tied to sufficiently clear requirements, qualified review, structured documentation, careful handling of customer data and code, and a tactical layer strong enough to say no when delivery becomes too easy.

What acceletail offers: Together with our nearshore partner Arteis, we use AI to make delivery faster and more controllable: better specifications, faster implementation, translation, documentation, test preparation and handover, delivered by our experts.