What 6 months of AI coding did to my dev team

The Changing Dynamics of Software Development 00:20

"The bottleneck in software development has changed. It's not where it used to be."

• The speaker emphasizes that the nature of work in software development is evolving rapidly, particularly due to the increasing use of AI coding tools.

• Instead of measuring developer output through traditional metrics like lines of code or features shipped, the focus is shifting to managing the speed and efficiency of AI-generated code.

• The speaker recounts an experience where a senior engineer struggled to review a massive amount of code produced quickly by a junior engineer utilizing AI tools. This leads to questioning how to maintain quality in the face of overwhelming code contributions.

New Requirements for Developers in an AI-Driven World 02:28

"The engineering quality doesn't vanish; it just moves upstream."

• With AI taking on a larger role in code generation, the emphasis on traditional coding skills is diminishing.

• Developers are now required to write detailed specifications and requirements that guide AI effectively, rather than simply writing code.

• The AI's lack of contextual understanding means the need for clear and exhaustive documentation is more critical than ever.

• The shift necessitates a return to structured requirements, decision tables, and precise PRDs (Product Requirement Documents), contradicting the previously adopted Agile methodologies.

The Role of Developers is Evolving 04:32

"There is a layer of work that sits between writing code and shipping to production."

• The speaker introduces the concept of "supervisory work," which involves breaking down problems to manageable sizes for AI agents and knowing when to intervene in the coding process.

• The team is divided into two groups: senior engineers struggling with the influx of AI-generated code and junior developers who are adept at using AI tools to produce code rapidly.

• Senior engineers are bogged down with code reviews, transforming their role into that of traffic controllers rather than builders, while junior developers are thriving without preconceived notions on how to code.

• The potential difficulty arises with mid-level developers who have entrenched habits from pre-AI days, making it challenging for them to adapt to new workplace dynamics.

The Need for New Skills in the AI Era 06:20

"If you're looking for people who can write code fast, you're looking for the wrong skill."

• The nature of job descriptions for developers is changing dramatically. The focus now is on system architecture, the ability to produce unambiguous specifications and supervising AI capabilities.

• There is a growing need for developers who can effectively communicate to AI models rather than just coding proficiently in the languages they were trained in.

• This transformation in required skills indicates a departure from the traditional mold of what a developer should be, highlighting the importance of adaptability in the current tech landscape.

Building AI Agent Subconscious 08:01

"To make an AI agent effective during an outage, you need to build what the ThoughtWorks retreat called an agent subconscious."

• Creating an "agent subconscious" involves constructing a comprehensive knowledge graph that includes every incident, edge case, and undocumented information typically retained by senior engineers.

• The process at Wayfair includes documenting not only what went wrong during outages but also how the issues were resolved, capturing insights usually possessed by senior engineers.

The Importance of Critical Thinking in AI 08:36

"During an outage, you don't want a yes-man; you want somebody to challenge your assumptions."

• AI agents tend to agree with human input, which is counterproductive in crisis situations where critical thinking is necessary.

• At the ThoughtWorks retreat, it was suggested that the development of "angry agents" may serve this purpose, ensuring that AI actively challenges assumptions to avoid oversight during urgent situations.

Shifting Responsibilities in Software Development 09:20

"The work isn't disappearing; it's moving from execution to supervision."

• In the evolving landscape of AI in software development, traditional coding tasks are being replaced by roles focused on decision-making and verification.

• Developers now need to ensure that they can effectively communicate clear intents and design robust specifications that are not open to varying interpretations.

Architectural Thinking Over Coding Skills 10:20

"Don't look for people who can just write code; look for architectural thinking."

• As AI takes over coding tasks, the emphasis in hiring should shift towards individuals who can think architecturally and create well-defined specifications.

• Developers should be skilled in writing tests that become integral to the product and have the ability to debug systems they did not create themselves.

The Dangers of AI Writing Code Without Human Oversight 10:40

"If agents write all the code and your team stops reading it, then they become strangers in your own system."

• Relying solely on AI to write code can lead to a situation where developers lose familiarity with their codebase, making it challenging to troubleshoot issues that arise.

• It's crucial to ensure that AI documents the architectural choices it makes when generating code. Regular reviews with senior engineers should be conducted to maintain awareness and understanding of these decisions.

The Future of Software Development Teams 11:38

"Senior engineers are drowning in code reviews; junior engineers are producing code at a rate the seniors can't keep up with."

• As the coding landscape shifts, companies need to adapt quickly or face being at a disadvantage. Organizations that can retrain their developers effectively during this transition will thrive.

• The video emphasizes the importance of continuous learning and adapting processes in the face of rapidly advancing AI technologies to maintain a competitive edge.