Large codebases often come with a hidden problem. It is not always the code itself, but the knowledge around it. Over time, teams build systems that rely on decisions, fixes, and workarounds that are rarely written down in full. That “tribal knowledge” tends to live in scattered documents, old tickets, or in the minds of experienced engineers, which is now starting to be addressed by AI agents.
A recent post from Meta’s engineering team shows how this problem is starting to shift. Instead of relying on manual searches or asking colleagues, the firm has been using AI agents to map and retrieve internal knowledge across large-scale data systems. The goal is simple: reduce the time it takes for engineers to understand how things work.
According to Meta’s engineering blog, this approach has already led to measurable changes. Tasks that once took up to two days of investigation can now be completed in around 30 minutes. The system also reduced tool usage by roughly 40 per cent by preparing relevant context ahead of time. These figures come directly from Meta’s internal testing, as shared in its developer tools write-up.
The change points to a broader shift in how AI is used in software development. Much of the focus in recent years has been on code generation. Meta is showing a different use case: helping developers understand systems, not just write them.
Turning scattered knowledge into something searchable
In large engineering teams, knowledge often sits across version control history, design documents, chat logs, and issue trackers. Even when documentation exists, it may not explain why a system was built a certain way.
Meta’s approach uses AI agents to connect these sources and build structured representations of how systems work, including service relationships, dependencies, and past changes. The outcome is an internal map that developers can query.
Engineers can ask questions in plain language, and the system retrieves relevant context before returning an answer. This reduces the need to manually trace through logs or read long threads.
Meta said pre-computing context plays a key role. By preparing information in advance, the system can respond faster and with fewer calls to external tools.
From code assistants to knowledge assistants
The rise of AI coding tools has changed how developers write software. Many tools can now generate boilerplate code, suggest fixes, or help with syntax. These features save time, but they do not address a deeper issue: understanding existing systems.
Meta’s work suggests that the next step may focus more on context and reasoning. Instead of acting as a coding assistant, the AI agent becomes a knowledge assistant. It helps developers answer questions about system behaviour, dependency risks, and how parts of the codebase have changed over time.
These are questions that often require experience or access to past discussions. By structuring internal knowledge, AI agents can make this information easier to reach.
A shift in developer workflows
The impact goes beyond faster searches and may influence development workflows.
If knowledge retrieval becomes part of the development process, AI agents may be integrated into code review systems, CI/CD pipelines, and internal documentation tools. This means systems may need to continuously process and update knowledge rather than wait for user queries.
This raises concerns regarding data quality. AI agents depend on reliable sources, and incomplete documentation may limit their usefulness.
Limits and open questions
Meta’s results are based on internal use, and it is not clear how well the approach will work across different organisations or codebases.
Large companies often have the resources to build these systems, while smaller teams may face challenges. There are also concerns around accuracy and trust, as AI-generated summaries can miss details or misinterpret context. If the system produces inconsistent results, users may return to manual methods.
Meta does not present the system as a replacement for human expertise. Engineers still need to verify and apply what they find.
A broader trend taking shape
Meta’s work reflects a wider change in software development, where AI is being used less for code generation and more for understanding systems.
AI agents may become a standard part of development environments, helping developers navigate existing systems rather than write new code. Meta said the system cut research time from two days to 30 minutes and reduced tool usage by 40 per cent.
The approach is still evolving, but it points to a future where understanding a system may depend less on who you ask and more on how well your tools can map what your team already knows.
(Photo by Dima Solomin)
See also: AI coding tools are driving a surge in new app development
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.
AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.