Guest: Yang Li, Chief Operating Officer of Cosine
Host: Seth Earley, CEO at Earley Information Science
Published on: May 27, 2025
In this episode of the Earley AI Podcast, host Seth Earley sits down with Yang Li, a leading figure in AI and software innovation. Yang is the Chief Operating Officer of Cosine, an advanced AI development firm, with deep experience driving startups, scaling organizations, and pioneering advancements in engineering and software development. Yang’s work focuses on leveraging AI to empower the next generation of developers, especially in navigating the increasingly complex landscape of modern and legacy codebases.
Yang and Seth dive into how AI is reshaping the role of software engineers, the evolving challenges of handling massive backlogs and legacy systems, and what creativity and efficiency really look like in an age of AI-powered software development.
Key Takeaways:
Insightful Quotes:
"Previously you had to use words and language to describe your idea, you can now show people your idea... The time between you having thought of an idea to actually be able to show people that idea has now reduced almost to zero because of vibe coding." - Yang Li
"AI is not replacing engineers. It's changing what it means to be an engineer. The craft is evolving from writing code to orchestrating systems and making strategic decisions." - Yang Li
"The biggest challenge with legacy code isn't the code itself—it's that nobody understands it anymore. AI can help bridge that knowledge gap." - Yang Li
Tune in to discover how to make AI practical, actionable, and intelligent for your organization.
Links
LinkedIn: https://www.linkedin.com/in/yangli92/
Website: https://cosine.sh
Ways to Tune In:
Earley AI Podcast: https://www.earley.com/earley-ai-podcast-home
Apple Podcast: https://podcasts.apple.com/podcast/id1586654770
Spotify: https://open.spotify.com/show/5nkcZvVYjHHj6wtBABqLbE?si=73cd5d5fc89f4781
iHeart Radio: https://www.iheart.com/podcast/269-earley-ai-podcast-87108370/
Stitcher: https://www.stitcher.com/show/earley-ai-podcast
Amazon Music: https://music.amazon.com/podcasts/18524b67-09cf-433f-82db-07b6213ad3ba/earley-ai-podcast
Buzzsprout: https://earleyai.buzzsprout.com/
Podcast Transcript: AI-Powered Software Development, Legacy Code, and Vibe Coding
Transcript introduction
This transcript captures a conversation between Seth Earley and Yang Li on how AI is transforming software engineering. Topics include the changing role of developers, tackling legacy codebases, augmenting rather than replacing engineers, the limitations of coding benchmarks, the emergence of vibe coding, and the future skills needed in an AI-powered development world.
Transcript
Seth Earley:
Welcome to the Earley AI Podcast. I'm your host, Seth Earley, and today I have Yang Li joining us. Yang is the Chief Operating Officer of Cosine, an advanced AI development firm. He has deep experience in driving startups, scaling organizations, and pioneering advancements in software development. Yang's work focuses on leveraging AI to empower developers. Yang, welcome to the show!
Yang Li:
Thank you, Seth. Great to be here.
Seth Earley:
So Yang, let's start with the big picture. How is AI transforming software engineering?
Yang Li:
It's fundamentally changing what it means to be a software engineer. Traditionally, being an engineer meant writing code—lots of code. But AI is shifting that. Now, engineers are spending less time on the mechanics of coding and more time on creative problem-solving, architectural decisions, and strategic thinking. AI is handling more of the routine coding tasks, which allows engineers to focus on higher-level challenges.
Seth Earley:
So it's not replacing engineers, it's changing their role.
Yang Li:
Exactly. AI is not replacing engineers. It's changing what it means to be an engineer. The craft is evolving from writing code to orchestrating systems and making strategic decisions. You still need deep technical understanding, but you're applying it differently.
Seth Earley:
Let's talk about Cosine and what you're building. What problem are you solving?
Yang Li:
We're focused on helping engineers deal with complexity—especially the complexity of large, legacy codebases. Most companies have massive amounts of code that was written years or decades ago. Sometimes it's in languages like COBOL or Fortran that few people understand anymore. And even modern codebases can be overwhelming. Our tool, Genie, helps engineers understand these codebases faster, navigate them more effectively, and make changes with more confidence.
Seth Earley:
Legacy code is such a huge problem. Why is it so challenging?
Yang Li:
The biggest challenge with legacy code isn't the code itself—it's that nobody understands it anymore. The original developers are gone. The documentation is outdated or nonexistent. The code has been patched and modified over the years. So when you need to make a change, you're working with something you don't fully understand. That's risky and time-consuming. AI can help bridge that knowledge gap by analyzing the code, understanding its structure and behavior, and helping engineers navigate it.
Seth Earley:
How does Genie work in practice?
Yang Li:
Genie acts as an AI assistant that understands your codebase. You can ask it questions—"What does this function do?" "Where is this feature implemented?" "What would happen if I changed this?" It can also help you make changes—refactoring code, fixing bugs, implementing new features. The key is that it understands the context of your specific codebase, not just general programming knowledge.
Seth Earley:
You mentioned reducing ramp-up time for engineers. That's a huge pain point for organizations.
Yang Li:
Absolutely. When a new engineer joins a team, it can take months for them to become productive. They need to understand the codebase, the architecture, the conventions, the business logic. Genie dramatically reduces that time. Instead of spending weeks reading code and asking questions, engineers can get answers immediately. They can start contributing much faster.
Seth Earley:
Let's talk about benchmarks. There are a lot of coding benchmarks out there. How useful are they?
Yang Li:
They're useful for research, but they can be misleading when it comes to real-world enterprise applications. Most benchmarks test on relatively simple, well-defined problems with clean codebases. But enterprise reality is messy. You're dealing with legacy code, complex dependencies, undocumented business logic, integration with multiple systems. The benchmarks don't capture that complexity. So a tool that scores 95% on a benchmark might struggle in a real enterprise environment.
Seth Earley:
That's an important distinction. What should organizations be evaluating instead?
Yang Li:
Evaluate on your own code, with your own team. Don't just look at whether the AI can write correct code—look at whether it helps your engineers be more productive. Does it reduce ramp-up time? Does it help with debugging? Does it make refactoring safer? Does it help with documentation? These are the metrics that matter in practice.
Seth Earley:
You mentioned "vibe coding" earlier. What does that mean?
Yang Li:
Vibe coding is this idea that you can go from concept to prototype incredibly quickly with AI. Previously, if you had an idea, you had to describe it in words, get buy-in, write requirements, wait for developers to build it. Now, you can just work with an AI assistant and build a prototype yourself, even if you're not technical. The time between having an idea and showing people that idea has reduced almost to zero. That's vibe coding.
Seth Earley:
That's powerful, but also potentially risky. What are the downsides?
Yang Li:
The main risk is over-reliance. If you just accept whatever the AI produces without understanding it, you can end up with code that works but isn't maintainable, or that has subtle bugs, or that doesn't follow best practices. You still need engineers who understand what's happening under the hood. AI is a tool, not a replacement for expertise.
Seth Earley:
How should organizations think about integrating AI into their development workflows?
Yang Li:
Start with specific use cases where AI can provide clear value—like onboarding new engineers, documentation, refactoring legacy code. Don't try to automate everything at once. Give your team time to learn how to work effectively with AI. Establish guidelines and best practices. Measure the impact. And most importantly, maintain human oversight and accountability.
Seth Earley:
What skills will engineers need in this AI-powered future?
Yang Li:
Several things. First, the ability to work effectively with AI—what some people call prompt engineering. Second, stronger architectural and system thinking, since you'll be orchestrating at a higher level. Third, adaptability and willingness to learn, because the tools are evolving rapidly. Fourth, creativity and problem-solving, since the routine work is being automated. And fifth, the ability to take calculated risks and experiment.
Seth Earley:
How are organizations rethinking career ladders and evaluations in this new world?
Yang Li:
That's a great question, and honestly, most organizations are still figuring it out. The traditional metrics—lines of code written, number of commits—don't make sense anymore. We need new ways to evaluate productivity and contribution. Maybe it's about the complexity of problems solved, or the quality of architectural decisions, or the ability to leverage AI effectively. This is an area that needs a lot of thought.
Seth Earley:
Any final advice for engineering leaders navigating this transition?
Yang Li:
Embrace the change, but do it thoughtfully. AI is going to transform software development—that's inevitable. But how you manage that transition will determine whether it's a positive transformation for your organization. Invest in your people. Give them the tools and training they need. Experiment. Learn. And don't be afraid to rethink how you do things.
Seth Earley:
Well, Yang, thank you so much for joining us and sharing your insights.
Yang Li:
Thank you, Seth. It's been a pleasure.
Seth Earley:
And thank you to our listeners. You can find Yang on LinkedIn and learn more about Cosine at cosine.sh. Thanks for tuning in to the Earley AI Podcast, and we'll see you next time!