--- title: "Ideas Worth a Longer Conversation: The JetBrains Research Podcast" url: https://daily.dev/posts/ideas-worth-a-longer-conversation-the-jetbrains-research-podcast-zljqodfra source_url: https://blog.jetbrains.com/research/2026/08/jbr-podcast type: article source: "JetBrains" published: 2026-08-25T10:16:53.004Z updated: 2026-08-25T10:17:21.056Z tags: ["ai-agents", "company-culture", "jetbrains"] reading_time: 10 upvotes: 1 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Ideas Worth a Longer Conversation: The JetBrains Research Podcast **[JetBrains](https://daily.dev/sources/jetbrains)** · 10 min read · 1 upvotes · 0 comments ## Summary JetBrains introduces its Research Podcast, which digs into deeper questions about software development beyond the usual AI hype. Episodes cover psychologist Cat Hicks on how team culture and belonging reduce AI-driven identity threat, philosopher Tomáš Petříček's five competing 'cultures' of programming and their differing takes on failures like the Knight Capital incident, Alexander Kulikov on teaching durable learning skills in an AI era, Ibragim Badertdinov on SWE-rebench (a decontaminated benchmark for coding agents that showed models score much lower than on SWE-bench Verified), and Anna Kogan on OpenCV's outsized impact despite minimal maintainer resources and slow AI adoption in sports analytics via her company FitWise. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why did models score much worse on SWE-rebench than on SWE-bench Verified? SWE-rebench uses decontamination, pulling GitHub tasks only after a model's release date so agents cannot have been trained on the answers. When this safeguard is applied, many models performed substantially worse than their reported SWE-bench Verified scores, suggesting those scores may reflect training data exposure rather than true problem-solving ability. SWE-rebench was accepted to NeurIPS 2025 and has over 12 million Hugging Face downloads. _Comparing coding agent benchmarks before picking a model gets easier when developers follow evaluation research on daily.dev._ ### How many developers maintain the OpenCV computer vision library? OpenCV, which runs on billions of devices and receives 32 million Python downloads a month on PyPI, has been maintained by only two to eight developers at any given time despite its massive reach. This mismatch between impact and maintainer resourcing is common among widely used open-source infrastructure projects. _Developers weighing dependency risk on widely used open-source libraries track sustainability stories like this on daily.dev._ --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#company-culture](https://daily.dev/tags/company-culture), [#jetbrains](https://daily.dev/tags/jetbrains) [View this post on daily.dev](https://daily.dev/posts/ideas-worth-a-longer-conversation-the-jetbrains-research-podcast-zljqodfra)