How XCL turned real business problems into AI-powered products with Udemy

The results varied by project but shared the same pattern: AI compressed the mechanical, time-consuming part of the work, and the skills people picked up on Udemy compressed the ramp-up curve to apply it. On Aptsense, technical recruiters had been losing 8 to 12 minutes per candidate on manual work like searching LinkedIn, enriching data across disconnected tools, logging information in the ATS by hand, and screening without standardized criteria — with 300+ active candidates a month, hundreds of hours of operational work that never produced any systemic learning. 

The discovery and structuring phases recovered 60% to 70% of the time normally spent on them, and that time was reinvested into more user interviews and validation with engineering. The product manager leading Aptsense had gone deep on prompt engineering, the highest-rated course in the AI catalog at 4.9 out of 5, and applied its patterns to synthesize six recruiter interviews that would have taken two to three days to analyze manually. The team also caught and corrected a case where AI misread an organizational process gap as a tool usability issue — a reminder that AI is strong at spotting patterns but weak at interpreting organizational context, and that human judgment has to stay in the loop.

A custom AI assistant was built to question assumptions rather than validate them. AI drafted the initial PRD and user stories, covering roughly 70% of what was needed; the remaining 30% — acceptance criteria, edge cases, and business-specific nuance — only emerged through iteration with engineering and grounding in what recruiters had actually said. Engineering iteration ran fluidly because the developers had already ramped through the IA pra Devs and Amazon Q Developer courses.

All candidate data was anonymized before reaching any external AI tool, in line with Brazilian data protection (LGPD) requirements.

On XCL Pulse, a monthly financial closing that used to take hours of manual downloading, cross-checking, and calculating now runs in minutes, with standardized calculations that previously varied depending on who did the work. On the marketing side, landing pages and campaign assets that used to take about a week to ship through the development queue now go live same-day or next-day, freeing up developer capacity for higher-complexity work. Across all three projects, the time saved did not just make people faster. It created room to focus on the actual customer problem instead of pure execution. 27 active Udemy courses remained in rotation, with more than 450 hours of content consumed in the same quarter.

“Udemy didn’t just make people faster,” Freitas said. “It created room to focus on the actual customer problem instead of pure execution.”