The battle to replace Github, building assembly lines for software, and why no one finishes projects anymore

The podcast discusses the evolving landscape of AI adoption in software development, emphasizing a shift from measuring usage by tokens to focusing on tangible business outcomes. Many companies, including large tech firms, are still in the early stages of integrating AI, often struggling with infrastructure challenges as they scale. GitHub's recent outages - driven by increased demand from AI automation - are highlighted as symptoms of this growing strain, prompting developers to explore alternatives like self-hosted solutions and new AI-native platforms such as Cursor. This shift reflects a broader trend where engineering teams are outpacing existing tools, leading to custom-built workflows and a reevaluation of reliance on incumbent services.

A central theme is the emergence of "software factories" - structured, factory-like systems for scalable and automated software delivery. These systems leverage AI agents to generate code and manage pull requests but face bottlenecks in review cycles and long-term maintenance. The discussion also addresses the difficulty of completing projects, particularly the final 10%, due to scope creep and mismatches between initial development skills and those needed for hardening and refinement. As AI accelerates the pace of creation, the value of skilled builders who can finalize and stabilize projects becomes increasingly critical, underscoring the need for better processes, handoffs, and leadership to deliver reliable, high-quality software.

• What if you shifted your AI adoption from tracking usage to measuring business outcomes?

  • Move: Audit your current AI tooling (e.g., Copilot, Cursor, custom agents) and replace token-based metrics (e.g., lines generated, prompts sent) with outcome-focused KPIs like PR merge speed, bug reduction in AI-generated code, or feature delivery cycle time.
  • Why Now?: The industry is moving past "token maxing" - early adopters are already seeing diminishing returns from raw usage; now is the time to differentiate by proving real engineering impact.
  • Expected Upside: You'll build a defensible, value-driven AI strategy that attracts clients or investors looking for efficiency gains, not just flashy dashboards.

• What if you migrated your core project off GitHub to a self-hosted or emerging alternative?

  • Move: Within two weeks, set up a mirrored repo on a resilient alternative (e.g., self-hosted GitLab, SourceHut, or Cursor's environment), automate sync workflows, and test failover during local outages.
  • Why Now?: GitHub's recent cascading outages disrupt AI-driven workflows that depend on constant agent-repo interaction - reliability is now a competitive advantage.
  • Expected Upside: Gain operational independence, reduce downtime risk, and position yourself as resilient to infrastructure shocks, enabling 24/7 automation pipelines.

• What if you treated your solo dev workflow as a software factory and systematized its final 10%?

  • Move: Map your project lifecycle, identify where work stalls (e.g., testing, docs, edge cases), then build reusable templates, checklists, or bots to automate completion tasks (e.g., auto-changelog, CI hardening scripts).
  • Why Now?: High-velocity AI tools let you start fast, but the bottleneck has shifted to finishing - those who ship complete products win trust and retention.
  • Expected Upside: Increase shipped project velocity by 30 - 50%, turning prototypes into reliable offerings faster, and establish a personal brand for delivery excellence.

• Implement outcome-based metrics for AI adoption in your projects, focusing on measurable business value rather than usage volume like token counts.
• Set up a self-hosted or decentralized code repository as a backup to GitHub, especially if you rely on continuous integration or AI-driven automation that demands high uptime.
• Designate a "project closeout" phase in your development workflow, assigning specific time and responsibility for hardening, edge cases, and final testing to avoid perpetual scope creep.
• Build lightweight automation workflows (e.g., using open-source tools) to mimic software factory principles, such as automated PR generation and review guardrails, tailored to your solo operation.
• Monitor long-term system health in automated processes by scheduling regular audits for drift, rot, or degradation - especially after initial AI-generated code deployment.