Replit now offers domain-specific AI agents that automate tasks within the development environment. A Growth Agent surfaces SEO issues, while a Security Agent identifies potential vulnerabilities. Users can select all issues and fix them with a single "Fix with Agent" action. The feature integrates directly into Replit’s editor, streamlining code improvement workflows. It was demonstrated by Replit founder Amjad Masad as a key productivity enhancement.
Ethan Mollick posted a hypothetical scenario on X. He posits that if AGI is achievable and labs can only be banned from using a model internally when they release it publicly, the Big Three labs may decide to capture all value from AGI themselves through expansion and acquisition. Sharing AI access with other firms would trigger the risk of internal bans, so avoiding public release eliminates that vulnerability. The tweet highlights a potential perverse incentive in AI governance. No concrete event or decision is reported; it is purely speculative.
Together AI shared the three main optimizations applied to accelerate GLM 5.1 inference. They rewrote the indexer topk kernel and fused the indexer kernel to reduce memory and launch overhead. Additionally, CPU overhead that was bottlenecking prefill throughput was eliminated. The indexer changes yielded the largest performance gain. GLM 5.1 is now available on the Together AI platform.
A social media post humorously notes the tendency of language models to frequently mention 'smoke tests', implying an exaggerated affinity for the term.
Pieter Levels, known as @levelsio, shared an X post asking what indie developers should build in 2026. The tweet includes two t.co links to unstated external content. No further details or context are given in the post.
Pieter Levels (levelsio) revealed that a vintage app concept called 'Nomads for Windows 3.11,' originally mocked up using Sora 2 video generation last year, has been turned into a real application. He credits recent advances in AI coding, stating that AI is now capable of developing software for vintage operating systems. Additionally, he announced that PietGPT 1.0 is now functional. The project highlights how AI tools can now produce working retro software from visual concepts.