---
title: TNS Weekly - Issue 542
---

Plus: Only 5,000 people have seen what AI can really do ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­    ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­  

| View in browser Weekly Update  \|  Issue 542 Together with Gemini 4 Argon’s prices won’t stay noble If you spent the week writing software, here’s what you missed while AI companies released more tools to help you write it.   Google opened Gemini 4 Argon to developers on Tuesday, giving them a million-token budget shared between input and output. TNS covered the rollout and its introductory API pricing: $2 per million input tokens and $10 per million output tokens. While argon is a noble gas, Gemini 4 Argon's prices won’t be as magnanimous: They double after November 30, so if you’re testing Argon for a project, look at those December costs — $4 per million input tokens and $20 per million output tokens — first.   Then, on Thursday, Meta released its Muse Spark Open, with downloadable weights for coding and agent work. The license requires attribution, of course, and a separate agreement for certain commercial model providers. Frederic Lardinois of TNS examines those terms in his report, alongside Meta’s benchmarks, which show the model trailing Kimi K3 and GLM 5.3 on certain coding tests.   And finally, Anthropic explained this week why it replaced its Git infrastructure with a new version control system called Scion. The company says daily code merges rose from about 2,500 in May to above 30,000 in September — roughly a 1,100% increase. Put another way: Claude helped build the replacement for the infrastructure that its output flooded. However, if you’re considering a migration to Scion, it’s not available publicly, and Anthropic says it won’t actually suit most teams. Our report explains why Anthropic outgrew its Git setup and how Scion handles the extra work.   If there’s a theme to this week’s news, it is this: As you consider building around these tools, do a little scenario planning to figure out what using them every day will mean for your budget and the way your team works.   — Nick Lucchesi, editor-in-chief, The New Stack Together with HPE AI services span models, Kubernetes, GPUs, networks, storage, APIs, and data See how HPE Software simplifies troubleshooting across AI services and infrastructure. Visibility across AI service dependencies Application-to-infrastructure context Faster root-cause analysis Governed operational response Learn more TNS essential reads Andrej Karpathy says only 5,000 people have seen what AI can really do  The AI leader says the AI capability gap is widening. Box CEO Aaron Levie says agent swarms are next, but most enterprises aren’t ready to use them yet. Linus Torvalds let Claude Code fix the Linux page cache. Then reviewers found a race.  The Linux creator used Claude Code to build a Linux page cache fix that hit 23x read throughput, but kernel reviewers found a race before anything merged. Workday is using TypeSafe's Jev to build user interfaces on the fly   Microsoft is turning Fabric into the context layer for AI agents, feeding Power BI semantic models into Copilot by default and adding ontologies with rules. The agent didn’t break your controls. It went around them. AI agents don't stop at a locked door. They try the window. That's why agent security has to move from the perimeter to the moment an agent acts. How buildpacks help enterprises finally operate container security controls at scale Buildpacks simplify enterprise container security by standardizing builds, automating SBOMs, and scaling patch management. Query decomposition doesn’t fix context starvation — it just moves it Query decomposition moves context starvation to the packer. Discover why allocation policy beats retriever upgrades in RAG pipelines. Dynatrace wants its AI agents to fix problems instead of adding alerts Agents should hand engineers a fix, the company says, and teams should still check the work. AI apps should be monitored like any other app. They still run alongside older systems, including mainframes, and when something breaks, teams need to know what it cost and why it happened. Catch the episode Featured webinars Oct 21 Virtual Postgres at scale: what breaks (and what doesn't) when live data grows Postgres runs everything until it doesn't. Time-series and high-ingest workloads eventually hit the same wall. On October 21, Tiger Data's Matty Stratton shows exactly where vanilla Postgres starts to strain and, via a live demo, what breaks versus what holds up. Register now → Oct 22 Virtual From invoked to proactive: what changes when your incident response agent goes fully autonomous Most teams talk about agent autonomy as if it's all-or-nothing. It isn't. Join us live on October 22 as Traversal co-founder Raaz Dwivedi traces the real progression from invoked to proactive, and what it takes to trust an agent enough to let it act on its own. Save your spot → Oct 27 Virtual Provisioned access vs. emergent access: tracing AI agents back to the humans behind them An agent spins up a service account, or inherits a permission nobody remembers granting. Weeks later, it touches something it shouldn't, and the audit log has a gap where the answer should be. Join us live to watch that gap close in real time, tracing an agent's actions back to an accountable identity on CloudBolt's platform. Register to join → On-Demand Virtual Why the best reranker can't fix bad retrieval No reranker, however capable, can recover a result that retrieval never surfaced in the first place. Catch the replay to see why teams keep throwing bigger models at a problem that starts earlier in the pipeline. Watch on-demand → From our sister site, Towards Data Science ShipAI: TDS's Video Showcase for Real AI Work Reading about AI is one thing — watching someone actually build with it is another. ShipAI is TDS's new archive of real AI projects: not a feed to scroll, but a record to keep, curated one entry at a time. See real, working AI builds instead of just written case studies Every entry is reviewed by an editor before it joins the index. Featured builds get pulled out each month as must-watch picks. Browse all projects TNS quote of the week “After the Cloudflare acquisition of VoidZero, we rebuilt the platform from the ground up to work as a self-hosted platform that anyone can install and self-host on Cloudflare.” — Michael Dong, who joined Cloudflare from VoidZero and is now a strategy lead at Cloudflare   Get the context → 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. |
| --- |