Sridhar Ramakrishnan is a senior technology leader in the Developer Experience / Productivity area for the last 7+ years. He currently leads an organization of 90+ engineers across 15 teams at Slack focused on DevXP. He is an avid mentor and coach on different platforms helping engineering leaders to scale. Outside work, he is busy taking care of his fruits and vegetables, and arguing metrics with his kids on life.
Sridhar on LinkedIn
AI will Improve Developer Experience and Velocity by 10x, says Sridhar Ramakrishnan, who leads the Developer Experience team supporting thousands of developers at Slack.
Slack's engineering team published a detailed blog post on how they used an LLM to convert 15,000 unit and integration tests from Enzyme to React Testing Library (RTL). By combining AST transformations and AI-powered automation, they achieved an 80% success rate in code conversion, greatly minimizing manual work and showing how AI tools can be leveraged in software development life cycle beyond just coding assistants. Sridhar shares that this results from Slack’s two principles for leveraging AI tools in software engineering.
The first principle was to build a dedicated Infra team for AI experiments to provide secure and compliant infrastructure so developers can experience AI tools without worrying about the cost, scale, or security.
The second one was not to look for new problems to solve with AI, but to consider the existing, well-known pain points from developer surveys. That’s how they ended up experimenting with using AI for code migration from Enzyme to React Testing Library.
Not everyone is on board with using AI tools. We take extra care in communicating with developers so they don’t get a negative first impression, and it matters.
Like every time they start using new tools, that experiment also began with the success metrics and how they would measure them. In the case of code migration, it was the percentage of accuracy. The second important thing, Sridhar says, is keeping humans in the loop to give feedback.
What surprised us the most was how much experimentation and iteration is needed compared to a conventional software development life cycle.
They learn from each iteration, and every new experiment starts with talking to the team that did the last experiment, getting their mentorship and support, and applying their learning.
Input Data: Training AI with good examples and good documentation
Other tools: Use them to fine-tune it, like RAG
Evaluate: After each step you apply to assess whether it’s helping or not
Don’t aim for 100%: Once you are happy with the percentage of accuracy, that’s it
Most success stories like the one from the blog post probably started as failures.
We didn’t just plug the code into AI tools and everything worked right away. None of our AI experimentations was like that. So, the main failures were the ones when we tried doing something once and stopped.
At Slack they use both internal (Slack/Salesforce) and external models, and the infrastructure is set up so that it’s easy to swap between them. When they started experimenting with AI at the beginning of 2025, the LLM cost was almost insignificant, but very quickly, within one or two months, it became the most expensive.
First, we looked into how we could optimize cost with the assumption that we would continue to scale, and the team did an excellent job of reducing the cost by 90 percent while the usage has increased.
We also looked at every problem and analyzed if it was the right problem to solve with AI and what the ROI was, for example, developer time saved or improved velocity.
1. Educating and upskilling engineers in the DevEx space.
AI is a new area, Sridhar says, and engineers have to learn some fundamentals about how things work. AI will not solve all developers’ problems; even if it did, they’d have to understand how it had done it at least at a high level.
Also, once they know the basics of how AI works, they could come up with their own ideas on how to use it.
2. Good infrastructure strategy making sure it’s reliable and secure, will it scale, what the cost is and there are metrics and monitoring in place.
3. Don’t stop at code assistants, and don’t think your engineers are fully AI-enabled if they use copilots.
Sridhar recommends reading blog posts and research papers. Some can be very technical and complex, so he uses AI to read them. He also recommends learning from engineers and sharing experience with other companies and colleagues in the DevEx space.
We didn’t just plug the code into AI tools and everything worked right away.
.svg)
-1.svg)
00:00 Introduction to Developer Experience at Slack
09:40 AI beyond coding assistants
12:17 Accuracy with AI models
18:29C hallenges and Failures in AI Experiments
24:42 Cost and ROI of AI experiments
29:10 Advice for Implementing AI in Developer Experience
34:09 Future Roadmap for Developer Experience at Slack
38:59 Rapid Fire questions
