From Skeptic to Superpower: Real‑World AI Coding Workflows That Scale | BRK229
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Two practitioners from Think, an energy management company, share their real-world journey adopting AI-assisted coding workflows. Starting skeptically, they evolved from basic AI tool usage to a sophisticated multi-agent 'software factory' using GitHub Copilot CLI, Claude, and custom orchestration. Their system breaks product requirements into PRDs, features, and parallelized tasks executed by specialized agents (security auditor, resilience auditor, API expert) overnight, with human review in the morning. Key lessons include: AI code drift compounds rapidly and requires architectural guardrails (hexagonal architecture, DDD, immutable specs); vendor lock-in is a real risk after Anthropic banned their org for 10 days; context management is critical for keeping LLMs on track; and using multiple models (Claude for implementation, GPT for QA review) catches more issues. The talk includes a live demo where an autonomous agent implemented a SQLite repository, passed adversarial review, and merged a PR during the presentation.