Pioneers of AI: How fast can you upskill in AI? We did a sprint to find out.

What It Covers

WaitWhat, the 40-person production company behind Masters of Scale, paused all operations for three days to run a company-wide AI sprint guided by AI engineer Parth Patel. The episode documents what worked, what failed, and the three core lessons extracted from building real AI tools using Claude and Replit.

Key Questions Answered

• •AI Sprint Structure: Pause operations for three focused days, divide staff into 12 small groups each tackling one specific workflow problem, provide universal tool access (Claude and Replit seats for every employee), and require each group to present a working prototype — not just ideas — at the end of day three.
• •Conversational Prompting: Ask AI to interview you before building anything. Team Gatorade reversed their workflow by requesting Claude ask clarifying questions first, then build. This back-and-forth approach produced a functional guest-speaker dashboard with tagged profiles and contextual notes, outperforming single-prompt attempts significantly.
• •Targeting Repetitive Click Work: Identify tasks involving manual data transfer across multiple platforms — spreadsheets, email, Slack, text — and build AI dashboards to consolidate them. Summit project manager D'Angela Napier built a real-time hotel operations dashboard that replaced scattered spreadsheet management, earning the sprint's standout reception from the full company.
• •Build vs. Buy Decision Framework: Before automating any workflow, evaluate off-the-shelf tools against custom-built agents on three dimensions: time, cost, and output quality. WaitWhat's video team tested multiple AI clip-generation products and found none met podcast-specific requirements like audio transcription, suggesting custom builds may outperform commercial options for specialized workflows.
• •Post-Sprint Integration Is the Hard Part: A three-day sprint generates roughly 30 ideas but converting them requires a dedicated task force, security infrastructure (WaitWhat built an AI agent called Warden to prevent prompt injection and protect financial data), and ROI measurement — complicated by unpredictable token costs that can jump from $1,000 to $1,500 per person daily within a week.