$$85,000 in tokens later: What I learned from scaling agentic coding at Lovable

Since joining Lovable, Alexander has significantly increased his reliance on AI-driven coding processes, resulting in a dramatic rise in token expenditure from $600 monthly to around $25K by May. This shift has transformed his development workflow, moving from solo coding and limited AI assistance to orchestrating complex agentic systems that handle a large portion of the coding process. Human review is now reserved for high-impact decisions, using AI for most code assessments, which are becoming more reliable over time. He has implemented an AI-based risk classification system to ensure high-risk changes receive human oversight, while other changes are managed by AI. This approach has led to increased productivity, with Alexander merging 293 pull requests in one week without any production defects. Despite the rapid advancements in AI capabilities, Alexander emphasizes the importance of maintaining a balance between token usage and innovative problem-solving, while also recognizing the need for human intervention at critical junctures. As development processes evolve, Lovable is focused on leveraging AI to improve product quality and exploring new methodologies to maximize the potential of agentic development.

No tracked trend matches for this post yet.