Better Than Vibes

Episode Summary

Jonathan Hall sits down with veteran software engineer and continuous delivery expert Brian Finster to discuss how to move beyond basic "vibe coding" and build disciplined, reliable workflows with AI. They explore why traditional CD and behavior-driven development (BDD) principles are more critical than ever, how to manage AI context windows and tokenomics, and why building an agentic scrum team changes the role of the developer into a technical lead.

Key Topics & Takeaways

• From Late Adopter to Agentic Workflows: Both Jonathan and Brian share their initial skepticism of AI coding tools, moving from treating LLMs as novelty toys to leveraging platforms like Claude Code for major refactoring and multi-file development.
• Continuous Delivery Meets AI: Brian explains how CD habits—small batch sizes, frequent commits, and rigorous automated fitness functions—prevent developers from accelerating their risk when using AI.
• Tokenomics and Code Structure: Clean, well-structured code isn’t just for humans anymore; modular, well-architected code is essential for optimizing token utilization and preventing agents from going off the rails.
• Building an Agentic Development Team: Brian discusses his work on agenticdevteam, setting up specialized agents for domain-driven design, hexagonal architecture, and adversarial code reviews.
• Specification-Driven Development (The Rebrand of BDD): Why writing clear acceptance criteria and focusing on behavior over implementation details dramatically improves first-pass accuracy for coding agents.
• The New "Home Computing" Era: Why nobody is a true expert yet, and why waiting on the sidelines means falling behind in an ecosystem that is evolving almost daily.

Timestamps

• 00:00 – Introduction and Welcome: Jonathan and Brian share their journeys into AI coding.
• 06:45 – Applying Continuous Delivery and BDD principles to AI workflows.
• 14:30 – Managing context windows, cognitive load, and preventing agents from breaking tests.
• 21:15 – Tokenomics: Why token costs demand better-structured, modular code.
• 28:00 – Real-world use cases: Using agents for massive repository analysis, architectural mapping, and refactoring.
• 37:50 – Building specialized AI agents for code reviews, domain-driven design, and testing.
• 46:10 – Is "Spec-Driven Development" just Waterfall in disguise? (Agile, intent, and hypotheses).
• 54:30 – Closing thoughts: Why you shouldn't sit on the sidelines of the AI revolution.

Links & Mentions

• Brian Finster's Resources: minimumcd.org
• GitHub Repository: bdfinst/agentic-dev-team
• Podcast Mentioned: Cup o' Go

In this episode, host Jonathan Hall sits down with fellow veteran software developer and early adopter Paul Hammond to discuss how to successfully integrate AI coding tools—like Claude Code—into a rigorous, professional engineering workflow without losing control or code quality.

Key Topics Discussed

• The Reality of AI Coding Assistants: Why early experiences with AI often feel like an "impressive parlor trick" that breaks more things than it fixes, and how to transition from chaotic "vibe coding" to disciplined engineering.
• TDD (Test-Driven Development) with AI: How Paul and Jonathan leverage Test-Driven Development practices to constrain AI scope, provide instant validation loops, and build comprehensive regression suites.
• Managing AI Context & Compaction: Strategies for dealing with context loss and compaction events, including the use of incremental plans (PLAN.md), architectural decision records (ADRs), and specialized custom agents or skills.
• Mutation Testing for AIs: Utilizing custom AI skills inspired by mutation testing frameworks like Stryker to ensure that AI-generated tests are actually robust and catching real defects.
• The Future of the Software Industry: An honest look at job security, the impact of AI on junior developers, the rising barrier to entry for the profession, and why core engineering fundamentals matter more than ever.

Guest Links & Resources

• Paul Hammond on LinkedIn
• Paul Hammond's GitHub Dotfiles: Explore Paul's custom configurations, skills, and agents on GitHub.
• Scenarist: Paul's open-source testing tool designed to help navigate modern frontend frameworks like Next.js (scenarist.io).
• Chippin: Check out Paul's upcoming micro-SaaS framework and application project at chippin.online.
• Ian Cooper's TDD Video: Referenced during the episode regarding behavior-based testing over rigid unit isolation (TDD Where Did It All Go Wrong).

If you enjoyed this episode, make sure to subscribe, leave a review, and check out Jonathan's work over at Cup o' Go and Better than Vibes!