ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
NewsMaterial3 min readDisconfirmAI Editorial

Angie Jones of the Agentic AI Foundation on How AI Agents Reshaped the Junior-Senior Developer Exchange

Angie Jones, former lead developer at Block and current vice president of the Agentic AI Foundation, describes how AI agents changed the junior-senior developer exchange at Block and how the Goose project shifted to an issue-first contribution process to handle agent-generated pull requests.

2026-11-22 05:58 GMT+8

Only a few years ago, it would have been impossible to picture this scenario, Angie Jones told The New Stack. A lead developer at Block, she once gave 10 new hires access to AI agents, handed them a codebase they had never seen, and set them loose on problems that customers had brought to them in person for a day and a half. It was their first day. They all delivered. For teams wondering where junior developers fit in an AI-assisted workflow, Jones says this experiment raises a question: what might experienced engineers learn from them?

Jones, who has since become vice president of the Agentic AI Foundation, draws a line between two activities that often get conflated. Learning to work with an agent, pointing it at a repository, iterating on its output, getting something running, is one skill. Learning to build an agent that others can depend on is another. The second still requires understanding the loop that lets an agent choose an action, examine the result, and decide what to do next, plus deciding where to constrain that loop and what access it should have to customer data.

The Goose project, an open-source coding agent Jones helped maintain at Block, offers a maintainer's view of what happens when the first skill spreads. Coding agents made it easy for people to propose fixes and features, which meant each pull request still took someone's time to review. The project shifted to an issue-first process: contributors and maintainers agree on the problem and the design before anyone generates a patch. "Let's talk about the problem and the direction that we want to take with the maintainers, then collaborate on the design," Jones said. Goose's maintainers also use agents to triage issues, review incoming code, and run security scans. The tools that make it easy to open a pull request can help maintainers manage the queue, while issue discussions give contributors a place to explain why a change belongs in the project.

Sign in / Sign up
ReadingResearchRadarInvestment framework

At Block, Jones approached adoption by asking a small group of engineers to make the development environment easier for agents to use. Baking that work into the system meant developers could point an agent at a repository and benefit from what the group had learned, even if they had missed last week's lesson in prompting. She sees a similar exchange between junior and senior engineers at work. "A new hire may know how to get an agent moving quickly. An experienced reviewer can help them see which of those changes are worth shipping," Jones said.

Jones teaches using the Model Context Protocol to connect agents to applications and uses Agent2Agent as an example of a standard for coordination between agents. With multiple companies agreeing on how to tackle common problems, developers have interfaces they can build around as the tools evolve. "You can bet on it and say, 'I can build on top of this now,'" she told The New Stack. That still leaves plenty of engineering to do, but the standards give teams a baseline rather than a moving target.

The Block scenario is a single anecdote with no control group. We do not know whether the same new hires would have delivered without agents, or whether "delivered" means production-quality code or working prototypes. Jones' distinction between using and building agents is intuitive but not rigorously defined, and the issue-first process she describes is Block's internal practice reported by Block's own lead developer. The reader should treat these as attributed observations from a practitioner navigating the problem in real time, not as validated benchmarks or causal findings.

Sources

  • thenewstack.ioThe New Stack↗
Continue researching →