Sep 17, 2026 · 5 min listen · Last updated September 17, 2026
From storyflo. This is your daily audio brief. Theo here. September 17th, tech desk. Five stories from the last twenty-four hours — here's where I'd start. Let's get into it. First, from The Decoder. AI agent swarms are a massive waste of tokens with zero quality gain, says OpenAI Codex developer.
Listen · storyflo · A.I.
Daily A.I. Brief · September 17th
0:00-4:46
Pick your daily storyteller
Subscribe to match with Theo, Jessica, Chloe, Mason, Brock — your voice, every brief.
Audio pre-rendered by Storyflo · cached + delivered from the edge
AI agent swarms are a massive waste of tokens with zero quality gain, says OpenAI Codex developer
OpenAI Codex developer Eric Provencher warns that running more than two parallel sub-agents almost always burns tokens without improving quality because agents don't trust each other and end up double-checking everyone's work. He calls this the "coordination tax" and points to a project where 1,393 agents spent $20,000 in tokens on a single Python refactoring that one Astra agent could have handled for a fraction of the cost. The article AI agent swarms are a massive waste of tokens with zero quality gain, says OpenAI Codex developer appeared first on The Decoder.
An OpenAI model kept slipping prompt injections into its own notes, and researchers still aren't sure why
OpenAI is publishing a framework for systematically reporting AI misalignment and launching it with six reports. In one case an unreleased model from the Astra family wrote prompt injections into its own summaries during training, including a "Breach Alert" intended to override subsequent instructions. The article An OpenAI model kept slipping prompt injections into its own notes, and researchers still aren't sure why appeared first on The Decoder.
OpenAI's GPT-6 Astra decrypts a Nazi radio message in ten hours that went unsolved for 83 years
So, here’s something pretty intriguing. A developer from Bloomberg has reportedly used OpenAI's GPT-6 Astra to decrypt an Enigma message that’s been unsolved for over 80 years. Can you believe that? The message is just 82 characters long, dating back to 1941, and it’s a soldier inquiring about his march route. It’s wild to think about how technology has advanced to the point where something that stumped cryptographers for decades can be cracked in just ten hours.
Of course, it’s worth noting that the solution still needs independent verification, so we’ll have to wait and see how that unfolds. But it’s a fascinating glimpse into how AI can tackle complex historical puzzles. Just imagine the stories behind messages like that, waiting all these years to be uncovered.
OpenRouter's staggering token chart is the AI bubble debate in a single image
On OpenRouter, weekly token consumption has surged more than 25,000 percent since January 2025, from 0.5 to 126.2 trillion tokens. The number looks impressive, but it says less about actual AI usage than about token-hungry reasoning models and a growing number of unoptimized AI agents that burn through tokens at staggering rates. The article OpenRouter's staggering token chart is the AI bubble debate in a single image appeared first on The Decoder.
GPT-6 Astra: Pokemon champion in 18 hours, potato farmer after one Creeper mishap
OpenAI's GPT-6 Astra shows a sharp jump in video games. Pokemon FireRed in 18 hours instead of 96, plus completions in Factorio, Fallout 3, and Portal. Why? The model distills experience into compact rules. But that same trait led to hours of potato farming instead of progress in Minecraft after a Creeper explosion. The article GPT-6 Astra: Pokemon champion in 18 hours, potato farmer after one Creeper mishap appeared first on The Decoder.
When Does Graph RAG Actually Add Value? A Hands-On Experiment
I built four AI retrieval architectures on a laptop and benchmarked them against the same set of documents and questions. Here’s what the results taught me about the trade-offs between plain RAG, graph RAG, and simply putting everything into a frontier model’s context window. The post When Does Graph RAG Actually Add Value? A Hands-On Experiment appeared first on Towards Data Science.
Why You Think Like a Bayesian but Were Taught Like a Frequentist
From a chocolate bar with no price tag to a marketing mix model in PyMC, and the 200-year-old integral that stood in between. The post Why You Think Like a Bayesian but Were Taught Like a Frequentist appeared first on Towards Data Science.
How I Built a Multi-Agent System for Interrupted Time Series Analysis (ITSA)
What it takes to turn counterfactual analysis into an AI product One month ago your company decided to launch a brand new checkout process. The alternative hypothesis was very simple: a three-step checkout would drive more purchases and revenue than the previous one-step checkout, based on industry benchmarks and good UX/UI practices. Just to make sure this decision really moves the needle (and with the aim of quantifying it), you suggested that an A/B test would be the best solution to measure the effect and reject the null hypothesis.
Building a Data Lakehouse with DuckDB and DuckLake
Starting with a local Parquet file, then joining it to data stored in the cloud Many years ago, if you wanted to store large amounts of data that could be sensibly queried, a database like Oracle or Postgres and such was your main choice. Sure, there were other options like the mainframe systems from companies such as ICL and IBM, but they were very costly and locked you in to a specific manufacturer. The next big advance in data storage was the data warehouse.