I’ve been using the OpenAI Codex since it came out and it’s great! Here are some early product feedback for OpenAI Developers and the team: 1. There is no way to update a PR with new code. Closing PR for each update is tiresome (use case here, I have a GitHub action pipeline that

This golden rule with the new Anthropic Claude 4 on Cursor is literally magic. I used Opus for planning and Sonnet for Executor and it’s been consistently delivering. It felt like it is working so much better (when it’s not overloaded). I honestly don’t know what to say.

I’m in Addis for the next week and I’m curious to learn what AI projects you are working on and how you’re using AI. Let’s get together for lunch/dinner and share experiences. If you are interested, sign up: forms.gle/SSoG7NtfpWCaWk… Space is limited.

Earlier this week, we organized an AI discussion event in Addis Ababa to explore AI adoption in Ethiopia, and I’m incredibly excited to share how it went. The diversity of ideas the different teams are working on was fascinating - founders are building solutions from AI enabled

If you want true superpowers when using Claude Code, do this in planning mode. Simply add “please make sure ultrathink my request and use sub agents to parallelize your review” after your prompt. You’re welcome. It’s bonkers how life before Claude Code feels like the dark ages

I think Figma Make has a very high chance of displacing Lovable. It’s just a really well built product. If they add support for Claude Code through the SDK, it could corner the whole market. What an incredible time to be alive!

.Gebeta Maps is building a fundamental stack to digitization. Many think mapping is a solved problem because of Google Maps, but having a digital map that’s owned locally and responsive to the changes is so critical as countries digitize. I’m really bullish on what Bemhreth Gezahegn and

Spent the last 24 hours using the GPT 5 models in different contexts. 1. The GPT 5 Thinking and Pro models are really amazing for any type of analysis. Really better than anything OpenAI has released before. 2. The coding is okay (better than before) but on par with Claude yet.

I am noticing a difference between using Claude Code in a standalone terminal and in Cursor terminal. The standalone terminal is able to keep context for a long time whereas the other loses context. Anthropic is there something to this? eric zakariasson - thoughts?

This is more true today than it was in 2023 and I’m afraid America is making a mistake by pushing India away. Siding with Pakistan to push India might work, but it might be more risky.

This week, I switched from Claude Code to OpenAI Codex. I have no idea what happened to Claude Code over the last two weeks, but as of now, Codex is producing better quality code more regularly than it. Are you seeing the same thing? Curious to hear your experience

This is amazing! Way to go to both teams. Congrats Beka and team

I have no idea what OpenAI to Codex, but it the improvement is eye watering. The difference between using Claude Code Sonnet 4.5 next to Codex is just so stark. Sam Altman was right about Codex. It feels like magic.

Reminder to apply to AddisCoder, a one month free algorithms and coding camp for high schoolers in Ethiopia. Deadline is December 31. Spread the word to those who may benefit from the program. addiscoder.com

Change happens slowly and then suddenly. The new generation of Ethiopian kids are built different and the country will change because of them. Incredible excited to see the progress and the work the Better Auth team is doing in Ethiopia. Amazing job!