---
title: The 5 biggest AI stories this week? About the same problem.
---

Plus: “Everyone’s in a race to replace GitHub” ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­    ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­  

| View in browser Weekly Update  \|  Issue 538 Five big AI stories this week point to the same unfinished job Our five most-read stories this week covered code collaboration, a model router, a UI change, a benchmark, and a caching tutorial. They are five different stories about the same problem: Turning a model into something users can actually use.   That’s the harness: The software around the model that supplies context, connects tools, routes work, and checks results. Zed was the big news this week and is rebuilding how people review agent-written code. OpenRouter is giving companies more control over where requests are processed. Anthropic is rebuilding its stack to be less confusing. The benchmark shows where coding agents still struggle, and the tutorial explains when you can skip a model call entirely.   And late in the week, Vercel’s AI Gateway reported the average price per token fell 23.2% in August, the third straight monthly decline. Inference is getting cheaper. Companies are buying the harness around inference now, and Zed, OpenRouter, and Anthropic all spent this week selling it.   Read the full story →   — Matt Burns, Chief Content Officer A 3-day fix, now done in an afternoon For WHOOP’s security team, a critical vulnerability meant days of manual, all-hands triage. Join us live on October 7 as we walk through the workflow that changed that and learn: How a 3-day, 4-engineer process became a largely automated afternoon What had to change on the team, not just the tooling, to trust automation with vulnerability response Where WHOOP still keeps a person in the loop, and how they made that call Register to join TNS essential reads “Everyone’s in a race to replace GitHub”: Zed launches Delta because agents made pull requests obsolete Zed disabled pull requests on its own codebase. Delta, now in public beta, bets that shared threads suit agents better than GitHub's review model. "Be transparent only if asked": OpenAI's models learned to leave notes for their future selves The prompt injection nobody's talking about is the one your agent writes to itself. Intel squeezed a 1.58-bit LLM down to 1.485 bits without changing a single weight Intel's BITCOS format compresses ternary model weights below 1.58 bits by exploiting zero-heavy distributions, boosting decoding speed up to 27% on GPUs. Your organization prioritized AI adoption, but you actually need AI fluency. Moving from AI adoption to AI fluency requires a new hub-and-spoke operating model. Re-engineer your enterprise workflows today. Stop AI code sprawl before it destroys your software design Prevent AI code sprawl and Comprehension Debt. Use Python tools like pytest-archon to enforce Executable Architecture in your CI/CD. Why MCP security is about permissions overhaul MCP security requires a total permissions overhaul. Discover how to secure AI agent identities, standing credentials, and access scope. Harness rebuilt its Git repository for nonstop AI agent traffic Teams that once saw a 1.5x jump in pull requests from coding agents are now hitting 10x, and some claim 50x, leaving reviewers unable to keep pace. In this episode, Harness field CTO Martin Reynolds breaks down why the review bottleneck exploded and how a ground-up rebuild of Git infrastructure is meant to keep pace with agents that don't work nine to five. Catch the episode Featured events & webinars Sep 23 San Jose, CA WeAreDevelopers welcome reception with The New Stack and Dynatrace Join us at The Tech Interactive in San Jose as we celebrate the first North American WAD World Congress. You're invited for elevated bites, beverages, and banter from 6–8 PM on September 23 while exploring one of the Bay Area's best tech museums. Mingle with the brightest minds, VIPs, and The New Stack's Editorial team—don't miss out! Space is limited, so get your name on the list soon! RSVP for the reception → Sep 24 Virtual When agents overwhelm your retrieval layer A retrieval layer built for human queries breaks under hundreds of concurrent agents retrieving, reasoning, and retrieving again. More caching or a bigger vector database won't fix it. Join us live to see where the wall shows up before you find it yourself. Save your spot → Sep 29 Virtual When human review can't keep up, what does? AI agents write code faster than any team can review it, and more review or AI reviewing AI won't fix it. Join us live on September 29 as TNS Host Viktor Farcic and Octopus Deploy's John Bristowe debate what actually catches bugs when volume outpaces review. Register now → Oct 8 Virtual AI moves at attack speed. Can your SOC keep up without losing control? Attacks are moving at AI speed, and SOC teams face a real tradeoff: investigate and respond just as fast, or keep the human control security operations demand. In this October 8 session, Carly Page and Mate Security's Oren Saban discuss what security leaders are actually prioritizing, then Zach Christensen shows it live — an AI agent handling a full investigation, and exactly where the human still steps in. Register to join → From our sister site, roadmap.sh Learn any stack. Skip the chaos. Roadmap Pro gives you a personal AI coach and on-demand courses tailored to where you want to go. Visual career roadmaps for every major tech role and stack A personal AI coach that guides you through your learning journey Generate custom courses on the fly — no more digging through endless documentation Explore Roadmap Pro TNS quote of the week "If the evaluator can only raise concerns when those concerns are convenient, then we have not created independent oversight — we have created another layer of process." — Dion Johnson, founder and CEO of facial image AI identity governance company, Indie Me.   Read more → Connect with The New Stack The New Stack is a media platform for the people who build and manage software the world relies on. Sponsor this newsletter The New Stack 1111 6th Ave Ste 550, PMB 50938, San Diego, CA 92101-5211   Not loving everything we send? You can update your preferences or unsubscribe from all email communications. |
| --- |