🧠 IDEs as we know them are dead
I’ve been using Cursor as my IDE, and I’ve realized that I barely use most of its features anymore—simply because I’m writing less code manually.
I think IDEs can afford to be much more lightweight when it comes to editing features. Instead, they should be designed around agents, workflows, and review. GitHub actually feels closer to this setup already. You have an Agent tab, you can orchestrate tasks through GitHub Actions, review through PRs, and spin up a Codespace that’s sandboxed to your repository.
I can even imagine “projects” becoming a way to launch agent tasks and track their progress.
So yeah, in the long run, I think GitHub is well positioned here. Editors built on VS Code are going to have a hard time—or will need some drastic changes.
🛠️ Deepfakes are too easy now
Last week I had to get creative for some marketing content. If you haven’t seen it:
Beyond being a fun experiment, it gave me a firsthand look at the challenges involved. How good and easy is it to create a deepfake?
First, you have to realize that most platforms have security checks—so deepfaking a celebrity is harder if you’re not using open-source tools. At least, that’s what I thought.
Turns out, many of these tools have barriers when you use the AI directly, but far fewer when you use their API. 😅
Once you have a voice clone and a video, there are tools like sync.so that sync a given voice audio with video footage. It does a decent job.
But speaking of open source—cloning a voice is actually really easy now.
Jeff Geerling did a video about how scary the new Qwen3-TTS (text-to-speech) model is. No long speech sample needed—just a small voice extract and you get really good results. Sure, intonation and emotion still need work. But if you want to change a few words in a given speech? Really easy.
Try it yourself on Hugging Face: https://huggingface.co/spaces/Qwen/Qwen3-TTS
All in all, just chaining a few tools with some open-source models can get you surprisingly far in a matter of minutes—at okayish quality. It’s only a matter of time before the results become unrecognizable from real footage. 😬
📚What I Read/Watched
ElevenLabs just got nuked by open source — Jeff Geerling shares his story about people faking his voice and the evolution of these open-source models
Inside OpenAI’s in-house data agent — this will be the playbook for companies building data agents to lower the technical barrier for internal analytics
GitHub just announced Claude & Codex available as coding agents
Maxime Beauchemin (creator of Airflow) on the semantic layer’s role in the context of AI
AI Expert: Here Is What The World Looks Like In 2 Years! — ex-Google insider Tristan Harris shares some nuggets about the AI race and its consequences for our economy and possibly democracy