Medium: I used AI every day for 30 days in 2026 — the workflow that actually works

Adam Regiaba ran a 30-day experiment in early 2026: use AI across writing, research, planning, and administrative work — but discontinue any use that degraded quality. The article documents what he kept, what he cut, and how he structures AI involvement now.

The central insight is spare: “AI doesn’t replace thinking. It replaces friction.” The experiment confirmed that AI performs well in the middle of a process and poorly at the beginning. Delegating the initial draft to AI consistently produced output that felt technically adequate but lacked the specific angle, voice, or judgment that came from the writer thinking first.

Regiaba’s workflow has five steps. First, human-first ideation: write messy initial thoughts before asking AI anything. This forces the writer to form an opinion rather than react to AI output. Second, limited research: use AI to summarize sources and identify gaps in coverage, but verify all facts independently — AI citations are unreliable. Third, outlining: AI is most useful here, improving structure and flagging redundancies in a draft outline. Fourth, human-driven drafting: the writer drafts first and AI enters only to tighten sentences or flag unclear passages, not to write sections. Fifth, rigorous verification: fact-check everything AI touched, evaluate tone and authenticity, and treat AI-generated sentences with the same skepticism applied to unverified quotes.

The seven mistakes Regiaba identifies are concrete: letting AI write the initial draft, trusting unverified citations, over-automating personal or voice-dependent work, using vague prompts, ignoring privacy risks with sensitive material, prioritizing speed over clarity, and confusing fluent sentences with accurate ones.

The most effective prompt he found in 30 days: “Rewrite this so an intelligent non-expert could understand it.” Narrow, specific, and evaluable.

Useful for writers and journalists who have access to AI tools but are uncertain where to deploy them in a professional workflow. The framework applies to long-form writing, editorial work, and any context where voice and accuracy matter more than volume.