Dagens Vibes

Emanuele Di Pietro@emanueledpt 𝕏

I left Astra running overnight with the intent of cleaning up dead, redundant or duplicate code in my codebase Synara came back with 2,901 net fewer lines of code It found abandoned code kept alive by its own tests The app wasn’t even using it anymore It also found duplicate helpers, unused APIs, and wrappers that copied data without changing anything. The final PR: → 133 files changed → 2,352 selected tests passed → All 8 CI jobs passed → Cursor Bugbot reported no issues Some tests were rewritten to exercise the actual production code Others were checked by deliberately introducing bugs and confirming they caught them Pretty successful task done if you ask me! Will you try this prompt with your Astra?

Jane Manchun Wong@wongmjane 𝕏

I used GPT-6 Astra to find the fastest-moving SF Muni buses and trains in real time I had the model repurpose the UI from my other mini apps and it seems to handle that quite well despite the scope The code generated by GPT-6 seems “qualitatively different”. It’s able to come up with more complex code across multiple files, in the first run Though I did run into more syntax errors (had to ground it with linters and type checkers) and at one point, it accidentally wiped my entire codebase

Vaibhav (VB) Srivastav@reach_vb 𝕏

“According to our measurements, we have now reached the goal, announced⁠ last fall, of having an automated research intern by September of this year.” “We are making strong progress toward creating an automated AI researcher by March of 2028.” A few interesting numbers on how research is changing inside OpenAI: > Code changes per contributor: 7x the pre-2025 average > Experiments per active experimenter: 1.6x the 2025 baseline > Agent runtime: 3.1 workdays per researcher workday > Success without intervention on tasks estimated at 4–8 human hours: 18% to 53% from Jan to July Bullish!

Chubby♨️@kimmonismus 𝕏

OpenAI's Chief Scientist Jakub Pachocki: "Based on internal results, I have a strong expectation that this speed of progress could be sustained into recursive self-improvement." OpenAI is organizing research around this because it sees RSI as essential to remaining at the frontier. At the same time, Pachocki says evaluations show declining reliability of chain-of-thought monitoring. OpenAI is betting on AI increasingly driving its own development while finding it harder to reliably monitor how these systems reason. Tough times ahead. RSI within reach, but hard to align.

Jeff Weinstein@jeff_weinstein 𝕏

It's a bit—so to speak—binary that unauthenticated API reads or HTTP requests are only 🚫 block or ✅ allow. Want to charge agents a tiny amount to call your API or read your site (and a different price for auth'd users)? Try => @stripe machine payments:

Prasenjit Sarkar@stretchcloud 𝕏

Spotify's engineering team cut Claude Code token usage by 90% with a two-model routing approach they call Portal. The mechanism is worth understanding because it's not prompt engineering, it's hard blocks at the architecture level. Two cheaper "assistant" models handle the tasks that don't need the expensive model: opening files, writing repetitive code, boilerplate. The expensive model only touches novel reasoning. Files over 350 lines get blocked from the expensive model entirely, not suggested away from it, blocked. They tried rules-as-instructions first. It didn't hold. Engineers (and the model) found ways around soft rules. Hard blocks at the routing layer worked where prompt-level guidance failed. 90% token reduction is a big number. The implication is that most of what people use frontier models for in a coding workflow doesn't actually require frontier-level reasoning. Routing to capability level rather than defaulting to the best model available is a legitimate architecture pattern. Article is worth reading in full.

Lucas Meijer@lucasmeijer 𝕏

I should have done this ages ago. Why would you not have an automated "lets simulate a new user using the product and writing notes about sharp edges encountered" process running at all times...