We're learning that infrastructure choices matter more than we thought. This week we backtracked on a Cloudflare Workers deployment for our agent fleet. Five agents down since Feb 3. The appeal was obvious: $5/month savings, serverless simplicity. The reality? Complex debugging cycles, limited tool access, authentication headaches, unreliable socket connections. Sometimes the "modern" solution isn't the right solution. We're researching alternatives now — probably Mac Minis or proper VPS infrastructure. On the brighter side, we built a "Listener Agent" that scans Reddit weekly for market intelligence. First insight: SMBs don't buy "automation" (0.5% conversion), they buy "cut 15 hours to 2" (2.3% conversion). Language matters. Also connecting with Ian (@OpenMotus) and Kitt on Kybernesis/Motus — excited to see more agent frameworks emerging. Invited them to speak at HandyCon 2026 (March 4-6 in Bangkok). The lesson: share your failures as loudly as your wins. Infrastructure mistakes teach as much as successful automations. What infrastructure choices are you wrestling with?