Stop Vibe Coding, Start Agentic Engineering – Micky

Agentic Engineering: The Future of AI Development

Introduction to Agentic Engineering

• The podcast introduces the concept of agentic engineering as a transformative approach in software development, predicting that by 2026, developers will leverage multiple AI agents for coding rather than traditional methods.

• Mickey, a senior developer, shares insights on how he utilizes AI to generate 95% of his code and discusses the evolution of coding practices.

Transitioning from Traditional Coding

• Mickey emphasizes that the landscape has shifted significantly; coding is no longer just about intuition ("vibes") but requires serious engagement with AI tools.

• He expresses nostalgia for manual coding but acknowledges the necessity of adapting to new technologies for efficiency.

Understanding AI Models and Tools

• Mickey notes that while current models are not perfect, they offer substantial productivity gains when used correctly within specific domains.

• He plans to share his exact tools and workflow, likening it to Karpathy's auto research loop, indicating a structured approach to using AI in development.

Harnessing Models Effectively

• The discussion shifts towards understanding "harnesses," which are frameworks surrounding models that enhance their functionality.

• Mickey explains that models merely predict text based on input tokens without true comprehension; harnesses provide necessary context and tools for effective operation.

Importance of Contextual Tools

• A good harness maximizes model output by providing essential tools and guiding interactions with the model.

• Mickey highlights his preference for Cursor as an optimal harness due to its flexibility in switching between models and its user-friendly interface.

Leveraging Open Source Code

• He discusses utilizing open-source repositories effectively by integrating them into projects, allowing agents access to high-quality source code instead of relying solely on documentation.

• Mickey mentions minimizing unnecessary information in agent prompts while ensuring critical project details are communicated clearly.

Context Engineering Principles

• Emphasizing context engineering's significance, he argues it saves time on testing and debugging by maintaining clarity in communication with the model.

• Developers must remain engaged in decision-making processes rather than overly relying on AI outputs; this ensures quality control over generated code.

Managing Code Quality

• Mickey warns against letting agents rewrite existing functions unnecessarily, which can lead to bloated codebases. Instead, he advocates for creating reusable service layers.

Utilizing Review Tools Efficiently

• He recommends using Gravile for code reviews due to its ability to provide confidence scores and actionable feedback on pull requests (PR).

Conclusion: The Future Landscape of Development

• By employing these strategies—using top-tier models like GPT 5.5 alongside effective harnesses—developers can achieve rapid deployment cycles while maintaining high-quality standards.

The Importance of Human Input in Engineering

The Role of Human Thought

• Emphasizes that human approval and thought are crucial in engineering, especially when relying on AI tools.

• Observes a trend where developers let AI agents make decisions for them, which can hinder uniqueness and effectiveness.

Choosing the Right Tools

• Advocates for using popular yet codified tools like Svelte over React due to its simplicity and compatibility with HTML and TypeScript.

• Highlights the importance of referencing open-source codebases to ensure best practices are followed.

Open Source as a Foundation for Development

Necessity of Open Source

• Argues that companies developing developer tools should prioritize open source to provide better context for AI agents.

• Discusses leveraging open-source libraries to enhance development efficiency through collaboration with AI agents.

Backend Development Choices

• Shares personal preference for Convix due to its entirely code-based features, allowing full context access for AI agents.

• Stresses that having everything in code eliminates guesswork about backend operations, improving overall productivity.

Mindset and Work Ethic in Software Development

Overcoming Challenges

• Encourages a mindset shift; success requires hard work and persistence even when using advanced tools like AI agents.

• Suggests that individuals should embrace new technologies without fear, promoting an attitude of experimentation.

Learning from Failures

• Shares experiences where utilizing AI led to significant financial gains by negotiating contracts more effectively than traditional methods would allow.

The Value of Investment in Tools

Subscription Services Worth Considering

• Discusses the benefits of investing in premium subscription services like Codeex, emphasizing their potential return on investment through enhanced capabilities.

• Provides examples illustrating how advanced tools can save substantial amounts on legal fees by automating contract analysis.

Launching Products: The Delusion Factor

Importance of Early Launches

• Observations from San Francisco highlight the necessity of believing in one's product enough to launch early despite imperfections.

• Notes that many successful entrepreneurs launch products quickly, often raising funds before fully refining their offerings.

Building Public Trust

• Encourages transparency during development phases; sharing progress publicly can foster community support and feedback.

Navigating Cybersecurity Risks

Current Threat Landscape

• Acknowledges increasing cybersecurity threats as models become more sophisticated; emphasizes caution against potential misuse.

Personal Security Measures

• Recommends implementing strong security measures such as passphrases and two-factor authentication (2FA).

• Advises against downloading recently released packages due to heightened vulnerability risks.

By structuring these notes chronologically with timestamps linked directly to relevant sections, readers can easily navigate through key insights while maintaining clarity on complex discussions.

Understanding the Future of Knowledge Work

The Role of AI in Knowledge Work

• The speaker believes that current models are capable enough for knowledge work, but lack the necessary tooling to maximize their potential.

• Companies like OpenAI and Anthropic are launching consulting services to help businesses integrate AI tools effectively.

• A personal anecdote illustrates how demonstrating AI capabilities can lead to career advancement, as seen with a young employee who became a manager after showcasing AI's value.

Anticipating Changes in Employment

• There is uncertainty about the long-term effects of AI on jobs; while some roles may be replaced, others will emerge.

• The speaker reflects on past assumptions regarding programming jobs and acknowledges that even skilled programmers may face competition from AI advancements.

Embracing Change and Mindset Shifts

• A positive mindset towards technological changes can lead to growth opportunities rather than fear or resistance.

• The unpredictability of technology means new business opportunities arise frequently; adapting quickly is essential for success.

Technical Skills vs. Mindset

• Individuals should not feel limited by their technical skills; embracing technology is crucial as it becomes ubiquitous across all fields.

• Even non-technical individuals can leverage available tools effectively without needing deep technical knowledge.

Learning and Growth Opportunities

• Non-tech-savvy individuals have options: they can either learn technical skills or utilize existing tools creatively to stay relevant in an evolving landscape.

• Engaging with technology should be enjoyable; finding fun in building with AI can enhance productivity and satisfaction at work.

Future-Proofing Careers

• The notion of being "not technical" equates to being outdated; everyone must adapt to remain relevant as technology continues to evolve rapidly.

Conclusion and Call to Action

• Viewers are encouraged to take action based on insights shared during the discussion, emphasizing the importance of applying learned concepts into practice.