Over the past few weeks, I've been thinking about Sonar CEO Tariq Shaukat's recent blog post "Solving the Engineering Productivity Paradox". His insights really resonated with me, but from a different angle—one that's particularly relevant to what we're building at Dosu.
Tariq's engineering productivity paradox basically comes down to this: AI is helping organizations write significantly more code (Google reports that "well over 30%" of their new code is generated by AI), yet most organizations are seeing engineering velocity increase much more slowly than expected. Google, despite AI generating 30% of their code, only sees a 10% increase in overall engineering productivity.
As Tariq points out, the bottleneck has shifted to code review. But I think there's an even more impactful bottleneck that we're not talking enough about: code understanding.
Understanding code is a key component of making good decisions, and with the rate of code being produced increasing every day thanks to the help of AI tools, it’s becoming more and more important, but simultaneously harder and harder to achieve. For both humans and AI agents, this negatively impacts decision making, leads to duplicative work, and repeat mistakes.
The Hidden Bottleneck: Understanding What We Build
The above diagram illustrates what I see happening across engineering teams. The blue line represents the "stunning" increase in AI-generated code—it's moving fast and getting faster. The red line is the pace at which engineers can actually understand this code: how it works, why it was built this way, what it does, and how it fits into the broader system.
Unlike the security review bottleneck that Tariq describes, the code understanding gap is often invisible. Engineers merge AI-generated pull requests that pass tests and look reasonable, but months later when something breaks or needs to be modified, they realize no one on the team truly understands how that code works or why it’s implemented the way it is.
This creates a particularly insidious productivity paradox:
- MOVE FAST: Ship AI-generated code quickly, but accumulate "understanding debt" that will slow you down later
- UNDERSTAND FIRST: Take time to truly comprehend each change, but sacrifice the speed benefits that AI promises
This isn’t a new problem, writing code has never been the bottleneck. To reverse the code understanding paradox we need a system that evolves understanding as fast as code changes, even as the rate of change increases.
Move fast, understand everything
This is exactly why we built Dosu to focus on building knowledge from code, as it gets written. Dosu’s agentic knowledge system doesn't just triage issues or answer questions—it actively builds and maintains institutional knowledge about your codebase.
When someone asks Dosu on Slack, for example, "Why was this implemented this way?" or "What happens if I change this parameter?", Dosu can provide context not just from the code itself, but from the discussions, decisions, and documentation that led to that code existing in the first place.
Dosu is helping teams bridge the understanding gap for internal teams in a few ways:
Dosu helps answer questions from non engineer stakeholders like product managers, marketers and sales teams.
Dosu lets engineers focus on building by proactively generating and maintaining documentation from code as it changes.
Dosu improves velocity by empowering those same engineers to be able to trust that AI in IDEs like Cursor have fresh, accurate context ingested so they can make better decisions, requiring less troubleshooting and fixes later.
For open source projects and public teams, it similarly empowers maintainers to build while triaging issues contributors open and helping close issues proactively, fighting OSS burnout and increasing velocity.
Building for the Future of Engineering
The reality is that AI-generated code isn't going away—it's only going to accelerate. But if we don't solve the understanding problem, we'll end up with codebases that work but that no human can maintain, extend, or debug effectively.
The teams that will thrive in this new world are those that treat code understanding as a first-class concern, not an afterthought. They'll invest in tools and processes that help engineers build and maintain deep comprehension of their systems, even when much of the code is generated by AI.
At Dosu, we believe the future belongs to teams that can move at AI speed while maintaining human understanding. That means automating the knowledge work that traditionally fell through the cracks, and ensuring that every line of code—whether written by a human or generated by AI—comes with the context needed to understand, maintain, and evolve it.
The engineering productivity paradox isn't just about balancing speed and quality in code review. It's about ensuring that in our rush to build faster, we don't lose the understanding that lets us actually maintain and evolve our software.
Try Dosu for yourself to see how it can empower key stakeholders and builders, both human and AI agents, to make better decisions by having access to always up to date knowledge. And I’m always up for a chat, if you’re so inclined.