Lenny Rachitsky · 8 min read · Gen AI Tools
🎙️ This week on How I AI: GPT 5.5, Claude Design, and GPT Images 2.0 hands-on reviews—plus an inside look at Memelord
Read the original on lennysnewsletter.com →
This episode puts GPT-5.5 through real-world, hard problems—from a six-hour autonomous data migration to hacking a proprietary Bluetooth protocol—showing its value when it can operate autonomously over long runs. The takeaway is that GPT-5.5 is incredibly capable but most everyday tasks don’t justify its cost; its ROI comes from solving genuinely tough technical problems. The show also tests Claude Design and GPT Images 2.0 and offers practical tweaks like using the /personality command to improve Codex interactions during long sessions.
Key Takeaways
- GPT-5.5 excels on genuinely hard technical problems and autonomous multi-hour workflows, not routine tasks.
- It can instantiate and manage subagents to perform data migrations, testing, repairs, and production prep with minimal prompting.
- The model demonstrated the ability to reverse-engineer a proprietary hardware protocol (Bluetooth bitmap encoding) after providing comprehensive context.
- Price point is high (GPT-5.5 Pro around $30 per million input tokens and $180 for outputs), but can be cheaper than human engineering time given the ROI on time saved.
- Small UX tweak like the /personality command improves Codex interactions and long-session experience.
Excerpt
GPT 5.5 just did what no other model could Listen now on YouTube • Spotify • Apple Podcasts Claire put GPT 5.5 to the test on real, messy problems—from a six-hour autonomous migration to a hardware hack no other model could crack. Biggest takeaways: GPT 5.5 is incredibly smart, but most ChatGPT users don’t have problems complex enough to justify its intelligence or cost. Claire struggled to find meaningful use cases in her personal ChatGPT account because everyday tasks don’t require super-intelligence. The model spent 17 minutes thinking about how to build a simple subtraction app for her first-grader—impressive, but…