Angie Jones, Global Vice President of Developer Relations at Block Incorporated, award-winning educator, and international keynote speaker, shared her insights on agentic AI with the Sauce Labs Developer Relations team on the latest episode of the Sauce Labs podcast, Test Case Scenario. This Q&A explores how this emerging technology is changing and evolving the software development and testing fields.
What exactly is agentic AI, and how does it differ from traditional LLMs like ChatGPT?
Jones: With ChatGPT, you're interacting with a large language model through a chat interface using natural language. If I ask ChatGPT to help debug a failing test, it will offer possible solutions based on the knowledge it has. Then I need to manually implement those solutions, switching between windows and feeding it more information as needed.
Agents perform the tasks on your behalf by connecting to your systems. Instead of pasting error messages into ChatGPT, I can connect my agent to my IDE or codebase and say, "I'm getting an error. Can you figure out what's going on?" Because it's connected to my tools, it has full context. It tells me how to solve the problem and implements the solution. For example, if I'm missing a browser executable, it can download that for me and make sure the error is resolved.
How does MCP (Model Context Protocol) fit into the larger agentic AI conversation?
Jones: MCP takes agentic AI even further. Companies can build a wrapper around their existing APIs using this protocol, specifically designed for AI agents, allowing agents to interact with their products. For example, when planning a trip, you need to book a hotel, flights, passes to attractions, and so on. Instead of just getting a plan, your agent can actually book everything for you by working with various vendors' APIs through the protocol (MCP). As an end user, you don't care about these behind-the-scenes integrations. The agent handles all the details.
With the DORA report showing 40% more bugs introduced by AI, are we really becoming more productive with these tools?
Jones: This really answers the question, "Is AI going to take my job?" The answer is no. If you leave everything to AI, you end up with buggy software. There's definitely a need for humans in the loop.
I think people are misusing these tools. You don't just put the tool there and say, "Do the job." You wouldn’t just leave a hammer next to a nail and expect it to work. You need to guide it. Similarly, if you tell an agent to build a feature for you, it'll give it a shot, but is that the best code possible? Probably not.
Quite rarely—I want to say never—have I been able to "one-shot" something, meaning I ask it to build something and it's absolutely perfect with no changes needed. I still say, "No, no, we need to change that" or "I don't like that approach." You're still involved, but you can move much quicker.
What does this mean for junior developers and testers who might not have the experience to properly check AI-generated code?
Jones: I'm concerned about what this looks like for junior developers. They might get something working and not recognize the weaker areas of the code that need improvement. Juniors should definitely continue to learn the basic foundations and fundamentals.
We take it for granted that everyone works on a big team with senior people who can check their work. Some people might be the first tech hire at a company, and they're excellent with AI tools, but their code might not scale well.
I question my AI tools a lot. Even if I'm not sure why something looks wrong, I'll ask, "Is this right? It looks a little wonky." The AI will often take a closer look and say, "Yeah, you're right, that's a little lazy. Let me fix it." I'll ask about security aspects even though I'm not a security expert. Just having the knowledge to know what questions to ask is incredibly valuable.
How might education evolve to prepare people for this new reality?
Jones: When I was learning to code, we had to write our code with pen and paper in the classroom, even though computers existed. There was something about writing that code down that helped it stick. During tests, we had no resources—no books, no internet. You had to know it for yourself.
Even as tools evolve, we might still need traditional learning approaches to build that foundation, supplemented by courses that introduce AI.The fundamentals matter.
Tell us about Codename Goose and how it's changing the way people work.
Jones: Goose is an open source MCP client I'm working on. Unlike proprietary agents that only know their specific applications, Goose can connect with any of the 4,000+ MCP servers out there.
It's LLM-agnostic, so you can connect it with any large language model you want. So you can craft your agent to do what you need at the time. If you're planning a trip, you might add MCP servers for airlines, hotels, or travel booking sites, then, in natural language, ask it to handle all the details for you. We initially developed it for developers, but when we started using it internally, we found great productivity benefits across all departments, which is why we decided to open source it.
How will this technology change the future of work, particularly for testers?
Jones: It completely changes how much we can accomplish. In my Selenium conference keynote "Air Fryers, Automation, and AI," I built an entire automation framework in about a minute. Even with some code review and tweaking, I was done in five minutes. That raises the question: what do I do with the other 7 hours and 55 minutes of my work day?
I shared some ideas for what automation engineers could do with that extra time: things we've always dreamed about or had on our backlog forever, but never had time to implement because we were constantly writing tests. Now that we have help with the routine tasks, we can focus on more innovative, high value work.
This technology is pushing the boundaries of what's possible and how much we can accomplish in a short period, freeing us to do more meaningful work.
If you’d like to watch the full Test Case Scenario episode, visit us on LinkedIn.
