Peter Dedene
16.5K posts

Peter Dedene
@dedene
AI & Tech Explorer · 15+ years building meaningful software · Digital Artisan & Musician · Founder & Engineer at @zenjoy · Building https://t.co/zkaZly9NVM



The latest CodexBar update renders API costs wayyyy nicer. codex.bar











A huge issue with Grok Build: in a few hours of light experimentation I burned through 5% of my MONTHLY usage quota for Grok Heavy. Normal use would be 3-5x that. Completely absurd quota.


Elon wants you to pay another $300/mo (or a “67% off” $100 promo) just to test the Grok Build terminal harness. I tried claiming that promo over and over, and it never applied. Even the official CLI starts by demanding SuperGrok Heavy validation. Screw that. I’d rather use the subscription I already pay for. So I built Groc (with a C): run the full Grok Build CLI locally with your existing ChatGPT/Codex auth. One command. Interactive mode. GPT-5.5/5.4/5.3/Spark model picker. And yes: in coding benchmarks and real-world agent runs, GPT-5.5 has been stronger for me than Grok’s best models. Great harness + better model = obvious combo. OSS: github.com/softmaxed/groc


Spent the evening with Grok Build. First thoughts 👇 It is FAST. The CLI is surprisingly polished for a beta. You can tell they put a *lot* of work into the details. The workflow needs some tuning. After planning, it stopped a bunch of times, asking me to run commands it should be able to run itself. Compared to Codex or Claude, it definitely needs more hand-holding. Context compaction is on the slower side, feeling a lot like Opus. Codex is still amazingly fast here. But the intelligence level surprised me. Not quite GPT-5.5 or Opus tier, but definitely not dumb. The final verdict? I'm impressed. It built the entire agent integration in Zentty fully on its own. That update is dropping in one of our next releases 🔥








