Sopra Steria: Augmenting, rather than Replacing Software Engineers

PAC recently talked with Yves Nicolas, Sopra Steria’s Deputy CTO Group, about using GenAI to transform its software development activities. At Sopra Steria, the emphasis is on augmenting developers and satisfaction, rather than targeting only productivity gains. Replacing software developers with LLMs is not the company’s strategy. Our conversation provided a reality check on the productivity gains that can be achieved in the short term and a perspective on how LLMs will transform the software development lifecycle (SDLC).

Productivity Increments for Now, Rather than Massive Cost Savings

Sopra Steria started experimenting with LLMs in 2023 and purchased 10k GitHub Copilot licenses in 2024. It has now fully deployed the tool with its developer workforce that uses Sopra Steria’s development toolkit. The company progresses with customers using their own development environments.

Despite several competitors’ claims that LLMs bring up to 60% productivity gains, Sopra Steria takes a more nuanced approach.

The company indeed finds massive gains for specific and fragmented tasks. However, it has not seen evidence of such high gains across the SDLC. To achieve such gains, it would take a systematic automation of all SDLC activities.

The company conducted a pilot and evaluated the performance of two development teams, one using Copilot and the other not. And indeed, the results show faster development in the LLM-enabled team. Yet, the company found increased developer satisfaction with this team. As a result, it has positioned LLM-based automation around developer augmentation.

Development of Higher Quality Code and Developer Satisfaction are Priority #1

The company’s early polls find that 80% of developers see GenAI as helping them develop code better and augment their interest in their work. Code development remains the #1 use case.

Sopra Steria also sees some behavioral change among developers, with more appetite for creating unit test scripts (and conducting unit testing) based on user stories. This is a significant change, as most developers traditionally had no time or interest in performing unit testing. Early experiments show that GenAI makes unit testing easier for developers. Sopra Steria finds that test coverage has increased. Code quality is rising, and Sopra Steria uses the open-source static code analysis tool SonarQube as part of its processes to assess code quality.

Sopra Steria’s use case #2 is onboarding new members in a project. The company finds LLMs helpful for new joiners on the project’s technology and functional sides. Sopra Steria finds that LLM-based onboarding is faster and puts less pressure on existing teams, minimizing disruption.

Next Is Application Modernization

Outside of pure software development, Sopra Steria targets application modernization projects. In the short term, the company has taken the incremental route and targets limited-scope technical tasks. Examples of such tasks easily automated by LLMs range from creating screens on SAP in ABAP to moving code from one Java SE version to another to conducting reverse engineering and creating documentation for stored procedures (embedding SQL queries within code). Expect Sopra Steria to keep on expanding its application modernization portfolio with such point solutions.

The company works in the long term around themes such as COBOL code conversion, helped by R&D partnerships with INRIA. The application modernization opportunities are huge: the company finds that technologies such as LC-NC, 4G languages, and computer-aided design, or initiatives such as citizen development, have significantly increased technical debt. Eventually, clients will need to address this technical debt and LLMs will help there.

Excessive Productivity Gains Expectations

This is the beginning of the journey, and Sopra Steria highlights that LLMs bring many opportunities and also challenges.

For instance, the company finds that the past GenAI hype has raised cost savings expectations from customers’ procurement offices. Sopra Steria is now educating them on realistic gains and selectively accepting gain-share contracts.

Clients expect high savings. Yet the traditional SDLC KPIs used in software development are best-suited to capture the actual gains. The company is still working on identifying the right KPIs to demonstrate those gains on both the productivity and the software quality side.”

Deploying GenAI also raises questions around the impact of LLMs on energy and electricity consumption. Despite the recent pushback in the US, sustainability remains a priority for most European firms. Sopra Steria advocates a nuanced approach to using LLMs while considering ESG goals.

Despite this nuanced approach to LLMs, Sopra Steria absolutely believes that GenAI will reshape the SDLC. Right now, the company’s strategy is to augment engineers. However, there is no doubt that LLMs will drive significant productivity gains. Agentic AI is coming with early use case identification.

Yet, as always, automation will not come only from tool usage. Process automation is another priority across development, away from piecemeal automation. The next step is the integration of development and testing/quality engineering. The two activities are siloed with different tools, processes, and skills. Integrating the two activities will come from making functional testing easily accessible to developers, and GenAI will help. Welcome back to the world of shift-left.