Adam Okoń's first big-stage talk, delivered at wroclove.rb 2026. Argues that complex forms are a bottleneck for non-technical B2B users, that undocumented form paths accumulate over a product's life, and that developers rarely see the daily pain of end users. Proposes an augmented-form workflow: an extra first step where users paste requirements in natural language (e.g. from a client email) and an LLM analyzes them to produce a pre-filled form state that the user audits and finalizes — giving non-technical people a 'code review' step suited to regulated markets. Compares monolith (Rails + Sidekiq/Solid Queue + Ruby LLM + Turbo Streams) and microservice (message-broker based) designs. Highlights using async and fibers for heavy IO-bound LLM workloads, with live statistics showing the throughput advantage over thread-based jobs, and recommends OpenTelemetry for observability and tracking baseline metrics before and after introducing LLM features. Closes with production case study: complex form with incomplete data sets saw dramatic time reduction, higher adoption of supporting features, and strong positive user feedback. Q&A covers measured async speedups, why AI instead of splitting UI into steps, back-end vs. browser-based WebLLM (blocked by security/regulated markets), preserving user options in constrained forms, and guiding users through nested conditional forms.