Mentoring the Machine — Eric Hou, Augment Code

Key Concepts

• Agentic AI
• Knowledge Infrastructure
• Context Gap
• Mentoring the Machine
• Parallel Exploration
• Prototyping
• Augment Extension
• AI Coding Assistants
• Remote Agents

My Journey to Realization

The speaker, Eric from Augment Code, shares his personal experience of using AI agents to improve software engineering productivity. He recounts a typical Tuesday morning scenario involving a design system component deadline, a staging emergency, and a new hire needing assistance. He highlights the common problem of context switching and firefighting, which costs the industry $300 billion annually.

• The Problem: Engineers spend 23% of their time maintaining code instead of building new features due to constant interruptions and context switching.
• The Solution: Using AI agents to parallelize work and focus on critical tasks.

He then demonstrates the Augment Extension, an AI coding assistant, and explains how it transformed his workday.

• Example: He shows how he used an agent to scope out the design system component, parse through logs for the staging emergency, and assist the new hire through the Augment Slackbot.
• Result: He completed all tasks, including a gRPC library upgrade affecting 12 services and 20,000 lines of code, in half a day, which would normally take three weeks.

Mentoring the Machine

The core realization is that AI should be treated like a junior engineer, requiring mentoring and guidance.

• Analogy: AI and junior engineers lack context, organizational knowledge, and experience.
• Difference: AI learns and executes quickly but forgets information, while junior engineers learn slowly but retain knowledge.
• Conclusion: AI is a perpetually junior engineer that can work on multiple tasks simultaneously.

To effectively use AI, engineers need to become "perpetual tech leads" and mentor their AI apprentices.

Scaling AI Across Teams

The speaker addresses the challenge of scaling individual AI success across teams and organizations.

• Problem: Replicating individual productivity gains with AI across teams is difficult.
• Root Cause: The context or knowledge gap, which also affects new hires and causes context deficit for engineers.

The solution is to institutionalize knowledge infrastructure by choosing the right tools and systems.

Three Steps to Get Started

The speaker outlines three steps for companies to successfully adopt AI tools:

1. Knowledge Gathering:
   • Explore existing knowledge bases (Notion, Google Docs, GitHub).
   • Fill knowledge gaps with meeting intelligence tools (e.g., Granola AI) to capture meeting decisions.
   • Integrate data sources using MCP and Augment native integrations.
   • Example: Using Granola AI to record meetings and generate task lists.
2. Gaining Familiarity:
   • Introduce AI tools across teams.
   • Explore the strengths and weaknesses of AI in specific contexts.
   • Teach the platform about coding patterns, architectural decisions, and business logic.
3. Leaning In:
   • Expand successful patterns.
   • Entrust more complex tasks as trust and confidence grow.
   • Share successful memories and task lists across teams (e.g., using Augment's "memories" feature).

Parallel Exploration and the Future

Solving the knowledge infrastructure problem unlocks AI's true economic potential, enabling parallel exploration of business opportunities.

• Traditional Approach: Design, build, test (sequential).
• AI-Enabled Approach: Rapid prototyping, iteration, testing, and convergence based on real metrics.
• Example: Augment has prototypes of a VS code fork and agents, allowing for experimentation and data-driven decision-making.

Parallel exploration allows for measuring and testing divergent approaches early on, leading to better-informed decisions.

• Key Argument: Using AI effectively can make software creation more of a science.

The speaker concludes by encouraging the audience to visit the Augment booth or website to try out their tools, including the new Remote Agents feature.

Notable Quotes

• "To make the most use out of AI, we need to work with it as we would work with junior engineers. Not assigning tickets, but mentoring."
• "AI is effectively a perpetually junior engineer, but one that can work on multiple tasks simultaneously and incredibly quickly."
• "If we use AI effectively to augment our organizations, we can make the creation of software more of a science, not less."

Technical Terms and Concepts

• Agentic AI: AI systems that can autonomously perform tasks and make decisions.
• Knowledge Infrastructure: The systems and processes for capturing, organizing, and sharing knowledge within an organization.
• Context Gap: The lack of understanding and information that hinders effective task completion.
• Parallel Exploration: Simultaneously exploring multiple approaches to a problem to identify the best solution.
• gRPC: A high-performance, open-source universal RPC framework.
• RFC: Request for Comments, a document describing proposed standards or technologies.
• Git Bisect: A Git command used to find the commit that introduced a bug.

Synthesis/Conclusion

The presentation argues that AI's potential in software engineering is best realized through a "mentoring" approach, treating AI agents as junior engineers needing guidance and context. Overcoming the knowledge infrastructure gap is crucial for scaling AI adoption across teams and enabling parallel exploration, ultimately transforming software creation into a more data-driven and scientific process. The Augment platform provides tools and features to facilitate this transformation.