Mentoring the Machine — Eric Hou, Augment Code

Introduction and Personal Engineering Challenges 00:03

  • Eric Hou introduces himself and sets the context of sharing Augment Code's journey with AI in production software engineering
  • Describes high-complexity systems where no single engineer understands more than 5% of the codebase
  • Emphasizes routine chaos in engineering: frequent interruptions, firefighting, and context-switching are normalized
  • Notes that 23% of engineers' time is spent on code maintenance, amounting to $300 billion annually in lost productivity due to context switching

The Impact of AI Agents on Daily Engineering Work 04:48

  • Introduces the Augment Extension and its agent capabilities, which operate by setting boundaries rather than direct instructions
  • Details a typical, chaotic Tuesday: behind on a design system component, firefighting staging emergencies, and mentoring a new hire
  • Demonstrates parallelizing work by delegating tasks to AI agents: codebase exploration, RFC drafting, log parsing, git bisecting, and communication via Slackbot
  • AI agents provide personalized support to new hires through integrated knowledge infrastructure, freeing up senior engineers for critical tasks
  • AI completes remediation of a complex gRPC library issue, spanning 12 services and 20,000 lines of code, within half a day—far faster than traditional estimates
  • Human engineers shift from implementation to evaluation, focusing on critical decisions and polishing agent output

Mentoring AI: A New Paradigm for Engineering Teams 09:03

  • Maximizing AI utility requires mentoring it like a junior engineer: supplying context, outcomes, and constraints rather than just assigning tickets
  • Both AI and new hires lack organizational context and must work in structured environments to succeed
  • AI processes and implements quickly, but forgets between sessions, making it a perpetual "junior" requiring ongoing mentoring
  • Engineers, in turn, act as perpetual tech leads, orchestrating and mentoring their AI apprentices

Scaling AI Productivity Across Teams and Organizations 10:30

  • Individual engineers gain significant boosts from agentic AI, but scaling this across teams is challenging
  • Major blockers are context and knowledge gaps—the same factors that slow human onboarding and productivity
  • Four out of five engineers cite context deficit as their biggest blocker
  • Solving the knowledge infrastructure problem is critical for scaling AI benefits beyond individuals

Building Knowledge Infrastructure: A Three-Step Approach 12:23

  • Step 1: Knowledge gathering—audit current documentation and knowledge sources, map gaps, use meeting intelligence tools to capture undocumented decisions
  • Step 2: Familiarization—allow both humans and AI tools to learn about each other and the organization; introduce tools broadly and let teams train them in context
  • Step 3: Scaling—share successful workflows (memories, task lists) organization-wide so compound learning and productivity can multiply organically

Real-time Demo and the Evolving Economics of Software Development 15:02

  • Live demo of agent personality creation illustrates rapid feature implementation alongside ongoing tasks
  • With robust knowledge infrastructure, information transfer becomes instant and scalable
  • Organizations can move from linear, sequential software development to rapid, parallel prototyping and data-driven decision-making
  • Parallel exploration facilitates testing multiple approaches at once, converging on optimal solutions faster and with real metrics
  • The engineering process becomes more scientific, reducing guesswork and making informed decisions earlier
  • AI-enabled teams achieve higher productivity by mentoring agents and institutionalizing knowledge, fundamentally changing how software is developed

Conclusion and Call to Action 18:41

  • Encourages engineers and organizations to choose tools that best allow them to mentor and guide their AI systems
  • Invites attendees to try Augment Code and its remote agents to experience parallelized engineering work