Presenters

Source

🚀 Beyond the Hype: How AI Agents and Serverless are Redefining Software Engineering

In a recent unscripted panel at the GoTo Conference, some of the brightest minds in tech gathered to pull back the curtain on how Generative AI (GenAI) and Serverless architecture are actually being used on the front lines. Moderated by Jeevan Dongre, the panel featured Nick Coult (Director at AWS), Robbie Kohler (VP at Yum!), David Anderson (Author and Architect at GP), Janak Agarwal (Product Manager at AWS), and Akshatha Laxmi (Software Engineer).

From 10 trillion function invocations to 15-minute prototypes, the conversation moved past the buzzwords to explore the gritty reality of modern development. 🌐✨

🏗️ AI as an Accelerant, Not a Replacement

For Nick Coult, AI isn’t just a toy; it is a daily productivity tool. He describes AI as an accelerant that changes not just how fast people work, but how they work.

  • The 15-Minute Prototype: Before AI, building a prototype might take 1 to 2 days. Now, using tools like Cursor or Hero, Nick can have a working prototype in 15 minutes. This speed allows leaders to demonstrate ideas visually rather than just talking about them.
  • The Human Driver: Despite the speed, the human remains firmly in the driver’s seat. Nick highlights a senior engineer who wrote a complex Rust prototype in a few hours—a task that previously would have taken weeks. The AI handled the boilerplate, while the human focused on the 10% of the code that was truly difficult. 🏎️💨

📈 Scaling to 10 Trillion: Where AI Meets Its Match

While AI is great at filling in standard patterns, Nick Coult raised a critical challenge regarding scale. AWS Lambda handles over 10 trillion function invocations per month.

  • The Knowledge Gap: LLMs are trained on public data like GitHub. However, you won’t find the architecture for a system that scales to 10 trillion requests on a public repository.
  • The Trade-off: AI excels at implementation and boilerplate, but for high-level system design at an “Amazon scale,” human expertise in availability and security is still mandatory. 🛠️🦾

🛠️ The Architect’s “Socratic” Sidekick

David Anderson and Robbie Kohler discussed how AI changes the role of technical leadership.

  • The Socratic Method: Instead of just generating code, David uses AI to inquire about his system. By feeding the AI well-documented architecture and infrastructure standards (built via AWS Control Tower), he can ask the AI to find problems or suggest improvements.
  • Staying Connected: For Robbie Kohler, AI acts as a bridge. It allows him to stay connected to the code without getting “stuck in the weeds.” He uses AI to summarize large Pull Requests (PRs) or explain code in languages he doesn’t use daily, like Go or Rust. This helps him maintain technical direction while focusing on high-level strategy. 🎯💾

🛡️ Modernization and the “Working Backwards” Philosophy

Modernizing legacy systems is a massive undertaking. Janak Agarwal emphasized that at AWS, the Working Backwards process is the secret sauce that prevents unnecessary refactoring.

  • The 5-Year Rewrite: Nick shared a story from his time as GM of ECR (Elastic Container Registry). Rewriting a critical backend system to handle future scale took 5 years because it had to be done with zero downtime for customers.
  • Developer Experience (DX): Janak views AI as a forcing function. As customers become more productive with AI, the volume of their requests increases by 10x to 100x. Product managers must use AI themselves just to keep up with the pace of innovation. 🛰️🚀

🎓 The Junior Developer’s Dilemma: Skill Atrophy?

A thought-provoking question from the audience sparked a debate: If we stop writing boilerplate, will our brains get lazy? 🧠💤

  • The Risk: Nick Coult acknowledged that if you don’t use a skill, it will atrophy. There is a real concern about how junior engineers will progress if they never have to struggle with the basics.
  • The Counter-Argument: Akshatha Laxmi argued that AI makes her a better engineer by uncovering edge cases she might have missed. She noted that while AI can scaffold an application in record time, a human must still understand the Big O complexity and business logic to ensure the code isn’t just fast to write, but fast to run.
  • The Verdict: You can’t fire an AI, but you can fire the human who blindly accepted the AI’s mistakes. Concepts still matter.

🤖 The Future: Agents, Guardrails, and Safety

As we move toward an agentic future, the panel addressed the challenges of non-determinism and safety.

  • Observability: Nick noted that 90% of observability for AI agents overlaps with traditional serverless monitoring (latency, error rates). However, new metrics like token budgets and GPU capacity are becoming essential.
  • Safety Critical Systems: For industries like aviation or the military, “hallucinations” are not an option. Nick highlighted AWS Bedrock Guardrails, which use automated reasoning and computational logic to verify the correctness of LLM outputs. This allows developers to use cutting-edge AI while ensuring the system remains safe and regulated. 🛡️👾

✨ Final Thoughts

The consensus is clear: AI is not a magic wand, but a powerful multi-tool. Whether it is Akshatha using it to master Typescript and Python patterns, or David using it to interrogate his architecture, the goal remains the same: delivering business value faster and with higher quality.

The future of tech belongs to those who can pair deep conceptual knowledge with the lightning speed of AI. Are you ready to ride the wave? 🌊👨‍💻🦾

Appendix