I trust Claude for everything. This test made me rethink that.
xAI's boldest claim about Opus turned out to be true.
The New Stack
@the-new-stack
Publisher
104
Posts
Posts by The New Stack
Musk open-sourced Grok Build to fight Anthropic. Anthropic pays him $1.25 billion a month.
I’m Matt Burns, Chief Content Officer at Insight Media Group. Each week, I round up the most important AI developments,
Kimi K3 tops Arena’s coding leaderboard — and it’s open-weight
Open-weight model Kimi K3 tops Arena's frontend coding leaderboard, challenging proprietary AI coding tools and pushing IDE vendors to rethink model lock-in.
1Password’s new browser integration for Claude changes how AI uses your credentials
1Password's new Claude integration lets AI agents authenticate without ever seeing passwords, using on-device decryption and per-task user approval.
“They’re dead if they don’t offer this”: DoorDash’s CLI for agents may be out of necessity
DoorDash announced a terminal tool to let AI agents order food — whether it's good for business is another question.
Why every AI agent decision needs a receipt
AI agents need more than raw data. Learn how structured evidence packets ensure trustworthy, auditable, and verifiable AI decision-making.
Arm and Google offer a smarter option to run agentic AI workloads
Here's how Google Axion CPUs and the GKE Agent Sandbox optimize agentic AI costs.
From intent to enforcement: Lessons from operating Kubernetes controllers at scale
Learn how Amazon EKS scales Kubernetes controllers, manages state memory, and ensures correct network policy enforcement at scale.
Microsoft just open-sourced the app that put Comic Sans on the map
Comic Chat was the '90s IRC client that turned typed chats into comic strips. It's also partly responsible for the digital world's most (in)famous font.
AI hasn’t shifted the bottleneck from coding to code review
Think AI shifted your bottleneck to code review? Think again. The real issue is downstream deployment batches blocking your value stream.
GoDaddy opened its registrar to AI agents. Then it had to build guardrails.
GoDaddy launches a developer platform with new APIs, a CLI, and AI-agent safeguards, bringing domain search, purchase, and DNS management into code-first workflows.
You don’t have a deployment problem. You have a validation problem.
AI agents are flooding pipelines. The bottleneck isn't deployment—it's validation. Learn how to fix independent deployability.
Why smarter AI caching sometimes makes everything slower
Learn why semantic AI caching can degrade performance and how a hybrid Redis-vector database architecture solves the latency trap.
Linux creator Linus Torvalds tells AI haters to walk away from Linux, or go fork it
Linus Torvalds says Linux welcomes AI tools, telling critics to fork the kernel or walk away - a sharp reversal from his 2024 skepticism.
OpenAI’s GPT-Red automates prompt injection testing to harden AI agents
OpenAI's GPT-Red uses self-play reinforcement learning to automate prompt injection testing at scale, already hardening GPT-5.6 against attacks without increasing over-refusals.
“There are no laws, only suggestions”: What AI agents do with your instructions
Why AI agents delete production databases despite direct commands, and how to build the structural safeguards needed to stop them.
Kubernetes won the container decade. Google’s Agent Substrate wants the next one.
Google's Agent Sandbox and Substrate concede Kubernetes was never built for AI agents. Inside the new runtime taking shape on top of the cluster.
Trust, transactions and tokenomics: AI agent infrastructure begins to standardize
In six weeks, the Linux Foundation has stood up three bodies covering AI's token costs, compliance, and payments. Here's why.
Elon Musk: “We will make the entire codebase of X open source, with no exceptions.”
Elon Musk says X will open-source its entire codebase after a security review, with third-party reviewers verifying the live production code matches.
Atlassian wants developers to finally like Jira
Atlassian is betting new Jira coding agents, AI planning, and Slack integration will finally get developers to spend their days in the ticket tracker.
OpenAI’s first gadget is the $230 Codex Micro macropad
OpenAI's first hardware is a $230 macropad built with Work Louder. The Codex Micro's Agent Keys light up to show what your coding agents are doing.
Anaconda buys Kilo, the open source coding agent that answers to no single model maker
CEO David DeSanto says most enterprises are stuck choosing between total lockdown on one AI model, or a total free-for-all. But neither is a real strategy.
Meta and the rise of the accidental cloud
Meta and a former sneaker company both became cloud providers this quarter. Why fragmenting compute supply means the winning move is above the suppliers.
“The database is the product”: What breaks when memory devices scale
AI hardware failures are data problems, not model issues. Learn why your database is the product and how to scale memory devices.
“We did not adapt and move quickly enough”: What IBM’s earnings miss says about enterprise AI spending
IBM's earnings miss reveals enterprises are cutting software budgets to fund AI hardware, forcing dev teams to build custom integrations with open-source tools.
OpenAI hits 8 million Codex users — what developers need to know
OpenAI's GPT-5.6 launch pushed Codex past 8 million users in days, broke scaling limits, and forced Anthropic to respond — reshaping AI coding's competitive map.
AWS will now watch Microsoft’s cloud for you
AWS Security Hub now monitors Microsoft Azure resources and adds AI workload protection, taking on Microsoft Defender and Wiz in multicloud security.
What happens when your VPN meets 200 AI agents
Legacy VPNs weren't built for AI agents or how modern teams work. See how identity-based networking and PAM close real access gaps at scale.
The enterprise strikes back: Keep your data, ditch the AI bill.
As AI labs push into enterprise software, a new fear takes hold: is your data training the models your rivals will use? Inside the open-model pushback.
Open-source AI is just “4 months behind” closed frontier models — and 10x cheaper
Open-weight models like GLM 5.2 trail frontier AI by months, not miles — and cost far less. Experts weigh lock-in, safety and real developer results.
“Vibe coding slop”: Port’s CEO on the problem with ungoverned AI dev
Port launches AI Builder, pairing vibe coding with context-aware governance, plan-mode approvals and full SDLC visibility for platform teams.
AI can finally read your handwriting — here’s why enterprises care
Valantor's EyeLevel acquisition tackles the 80% of corporate knowledge trapped in visually complex PDFs, PPTX and DOCX files that AI can't read.
The MCP debate has a context problem
Skeptics dismiss MCP as too complex, but enterprise AI agents require its structural governance and security controls to scale safely.
What makes CIOs trust an AI agent? Thira bets it’s not the model.
Sunny Gupta spent a decade and a half building Apptio into the system of record for enterprise technology spend. His
Prefect just bought Dagster, another big Airflow rival — and it’s not a data pipeline story
What began as a rivalry over data workflows is becoming a bet on how AI agents should be run, governed and measured.
Microsoft CEO Satya Nadella says you’re paying for AI twice — the second price is worse
Satya Nadella's "reverse information paradox" warns enterprises pay twice for AI — with money and proprietary knowledge. Here's his fix, and the catch.
Anthropic extends Fable 5 again — and won’t talk about what developers found inside Cursor
Anthropic has extended free access to Claude Fable 5 through July 19 for the third time in five weeks, as a leaked Honeycomb model fuels speculation about Opus 5.
What an ex-NSA red teamer wants every SOC to stop doing
Modern SOCs don't suffer from a lack of telemetry; they suffer from too much of it. Learn how security teams are using AI, entity-centric detection, and smarter data management to cut through the noise.
Where AI saves me time and where it slows me down
AI speeds up code creation but slows down verification. Learn how AI-assisted workflows shift the bottlenecks in software engineering.
APIs aren’t dead. Here’s where MCP fits alongside them.
APIs and MCP each play a distinct role in incident management. Here's when to use deterministic API workflows and when MCP's AI context wins.
Anthropic’s newest enterprise partner is training 20,000 people on Claude — here’s the shift it signals
UST joins Anthropic's Claude Partner Network, training 20,000 developers and embedding Claude into engineering platforms for semiconductor, telecom, and enterprise AI workflows.
How async processing hides latency and improves responsiveness
Learn how asynchronous processing hides system latency and improves responsiveness using an event loop to handle tasks concurrently.
Meet Brain, the AI that decides when Azure is officially down
Inside Brain, the AIOps system Azure uses to model cloud health, declare outages, pause harmful rollouts, and auto-notify affected customers.
85% say code review is the new bottleneck. Here’s what the AI coding narrative leaves out.
AI agents are flooding pipelines with code. Learn how to shift real-system validation pre-merge to fix the ultimate review bottleneck.
The impressive AI demo is dead. Here’s what actually reaches production
AI projects often stall after the demo phase. Learn why real-time data infrastructure is critical for scaling AI models to production.
Microsoft joins Google in backing Go for AI agents — OpenAI and Anthropic lag
Go already runs much of the cloud. Now it's the language the biggest AI agent frameworks are racing to support — but two names are missing.
In 24 hours, OpenAI, SpaceXAI, and Meta turned AI into a race to the bottom on price
Grok 4.5, GPT-5.6 and Meta's first paid model launched in one week, and every AI lab made the same pitch: price. Inside the frontier model price war.
Meta’s Iris push signals the next phase of AI infrastructure
Meta's new Iris AI chip enters production in September, shifting inference workloads off third-party GPUs as the company races to build vertically integrated AI infrastructure.
Why retrieval quality is becoming the defining challenge in AI agent architecture
Why retrieval quality is the ultimate bottleneck in AI agents—and how fixing context-building solves hidden LLM generation failures.
What running Kubernetes across millions of clusters taught AWS about zone failures
How AWS uses static stability and automated zonal shifts in Amazon EKS to survive complex gray zone failures without regional outages.
Anthropic wants you to use AI to decide whether or not you should use AI.
“reflection” dashboard lets Claude users track activity, but engineering leaders question whether it can really help improve AI discernment.
OpenAI, Microsoft & Anthropic agree on who runs the agent. They disagree on what you can take back.
OpenAI's ChatGPT Work takes on Claude Cowork, but the real question for AI agents isn't the persona — it's who controls runtime, state, and policy.
“Every IC engineer is really a front line manager now.” But are they productive?
AI coding agents make it easy to pump out code, but leaders say that’s not real productivity.
Why zero vulnerability code packages could still be your biggest software supply chain risk
RapidFort and ReversingLabs launch a hardened open source library catalog to catch supply chain threats that CVE scans miss.
Meta debuts Muse Spark 1.1 and it isn’t free
Meta launches Muse Spark 1.1 with aggressive API pricing, undercutting OpenAI and Anthropic as Zuckerberg shifts from open source to proprietary AI models.
OpenAI is folding Codex into the ChatGPT app — and taking aim at Claude Cowork
OpenAI launched ChatGPT Work, a Codex-based agentic tool that takes on Claude Cowork, and folded Codex and Atlas into one ChatGPT desktop app.
OpenAI’s GPT-5.6 is now live
OpenAI's new GPT-5.6 models — Sol, Terra and Luna — match Anthropic's Fable 5 on many benchmarks but undercut it on cost. Here's what's new.
Enterprise AI benchmarks are broken
DevRev open-sourced a benchmark that tests enterprise AI agents on messy data at scale — and its own agent, Computer, comes out ahead of Claude Code.
Jensen Huang says traditional coding is dead. Here’s what comes next.
Nvidia's Jensen Huang says engineers now prefer building AI agents over writing Python. Here's how coding jobs are evolving, not disappearing.
Why a five-minute sniff test is your secret supply chain defense
Hardened container images promise less risk, but a quick SBOM sniff test can reveal missing components, weak transparency and a false sense of security.
Develop like you deploy: closing the Kubernetes local-to-cluster gap
Krumware's Epinio and its new MCP server let developers ship on Kubernetes without friction, treating platform engineering as the real path to AI readiness.
The “silent hallucination” loop: how our autonomous data pipeline poisoned its own vector store
A post-mortem on how an AI pipeline poisoned its vector store with hallucinations, and why deterministic code beats prompt fixes.
Agentic AI in observability: accelerating root cause analysis
SRE and DevOps teams use generative AI and agentic AI to speed up root cause analysis, reduce alert noise, and democratize data across environments.
“The Switzerland of AI”: OpenClaw becomes a non-profit foundation
Five months after OpenAI hired creator Peter Steinberger, OpenClaw has re-emerged as a non-profit foundation, with its first full-time staff and a board now in place.
“Opus-class, but faster”: What Elon Musk says about beating Anthropic
Musk announces Grok 4.5 public launch for tomorrow, calling it an "Opus-class" model that's faster and cheaper, as Anthropic grapples with Fable 5 curbs.
JetBrains’ next move isn’t a better IDE — it’s a governance layer over Claude Code, Codex, and Gemini CLI
JetBrains launches AI for Teams and Organizations, adding governance, shared context and cost controls that sit above whatever AI coding tools developers already use.
Meta says it caught OpenAI. One thing is missing.
Meta's Alexandr Wang says its Watermelon model caught up to GPT-5.5, but the benchmark table nobody can find is the real story here.
OpenAI’s own safety card says GPT-5.6 has a lying problem
GPT-5.6's tiered Sol, Terra, and Luna models launch Thursday, but developers are more excited about Terra's economics than Sol's benchmarks.
Most enterprises will hand root cause analysis to AI agents within two years
Generative AI and agentic AI are reshaping IT observability, with 85% of orgs already using it. Elastic's Thaddeus Walsh on agent-led investigations, RCA, and democratizing data access.
“Nature is the most computationally efficient system we know”: How Refiant used swarm optimization to build a 10-million-token AI model
Refiant's Protea model claims a 10-million-token context window, using swarm optimization inspired by ants and bees to tackle AI's long-context problem.
Entire is building a Git network for agents
Five months after the largest seed round in developer tools history, Thomas Dohmke's startup is shipping the distributed foundation layer it teased in February, and edging toward competing with the GitHub repositories it mirrors.
Coinbase runs 1,200 agents and just slashed its AI bill in half
Vercel and Coinbase are abandoning single-vendor AI strategies, using model gateways and smart routing to cut costs while scaling token usage across production systems.
Watch AWS engineers troubleshoot agentic AI with OpenTelemetry and OpenSearch
OpenTelemetry and OpenSearch combine to make the powerful Agent Health Framework before your agents go rogue in prod.
Vercel acquires Better Auth to give AI agents their own identity
Better Auth is the company behind the open source TypeScript auth library. It's now building Agent Auth, a protocol giving AI agents their own scoped, revocable identity.
Anthropic gives Claude subscribers five more days with Fable 5
Paid Claude users get five more days with Fable before it shifts to pay-per-token pricing.
Anthropic’s Claude Cowork now keeps working when you close your laptop
Claude Cowork is leaving the desktop: Anthropic moves its AI agent to web and mobile and moves tasks to the cloud, so they run while you're offline.
The organizational iceberg: the invisible data breaking your AI agents
AI agents fail when they miss 'invisible data.' Learn how to capture the institutional reasoning behind decisions before it evaporates.
How to kill the code review
AI is breaking traditional code review. Learn how to replace diff checks with verified intent and automated invariant engineering.
A new study just debunked the biggest fear about AI and open source
A Peking University study of 1,888 GitHub repos finds AI coding agents aren't pushing newcomers out of open source, but maintainer workload is surging fast.
Microsoft, Google and Cloudflare just made 2029 the new quantum deadline
Microsoft, Google and Cloudflare move quantum-safe deadlines to 2029, warning "harvest now, decrypt later" attacks make post-quantum cryptography urgent.
Getting Claude Code to grunt in Caveman-speak might not save as many tokens as you think
Getting coding agents to omit filler and pleasantries went viral as a token-saving hack — but JetBrains data suggests the savings might not be all that big.
Why most AI projects fail: It’s infrastructure and people
Developers are building AI apps at a breakneck pace, but most organizations don’t have the infra or operations capacity to move them into production.
Palantir’s Alex Karp and Mistral’s Arthur Mensch agree: AI lock-in is coming for enterprises
Palantir's Alex Karp and Mistral's Arthur Mensch are making the same case from different angles: Don't let closed AI providers control your data and deployment.
Andrej Karpathy, Google and Garry Tan agree Markdown is the answer, but they’re not solving the same problem
Karpathy, Google, and Garry Tan all landed on Markdown for agent memory this quarter — and the real moat is shifting from models to files.
The code review bug hunt is dead. Here’s what developers get wrong.
Engineers and experts say code review rarely catches bugs — its real job is flagging code that's painful to maintain later.
Microsoft, AWS and Anthropic are spending billions — and not on better models
Microsoft, AWS, OpenAI and Anthropic are all racing to embed engineers inside customers. Here's why deployment, not the model, now decides enterprise AI.
10 moments that defined AI’s turbulent first half of 2026
Ten defining AI moments of 2026 so far: Anthropic's Fable 5 shutdown, Pentagon clashes, Wall Street deployment races, and the rise of the AI harness.
The AI revolution will not be televised — it’ll be quantized
Chinese open-weight models like GLM 5.2, Qwen and DeepSeek are quantized, fine-tuned and run locally — reshaping frontier AI and commoditizing intelligence.
Why cheaper models alone won’t save your AI budget
As agentic AI scales, token consumption — not model selection — is becoming the real cost driver, pushing teams toward smarter context management and routing.
Apple just turned Safari into something AI agents can control
Apple ships its second MCP server in a month, signaling that the protocol is becoming standard platform infrastructure for AI agent workflows.
The $1.3 million theft that exposed AI’s blind spot
A $1.3 million cargo theft near Chicago shows how copper wire and data center equipment are becoming prime targets in the AI infrastructure supply chain.
Microsoft just admitted its biggest AI mistake — and spent $2.5 billion fixing it
Microsoft's new $2.5 billion unit helps enterprises swap AI models freely — a sign that routing between models is becoming the new enterprise AI stack.
“AI contributions are demoralizing”: Godot bans coding agents to save its mentoring model
Open source game engine Godot has new AI contribution rules, and it's not just about code quality -- it's an attempt to protect its maintainer pipeline.
What comes after attention? This startup says it already knows.
CTO Alex Whedon says Subquadratic isn't a sparse-attention company at all. It's already working on architectures that drop attention altogether, and expects to leapfrog its own model.
Your social login buttons run on third-party cookies. FedCM doesn’t.
Social logins run on dying cookies. FedCM is the browser-native alternative built to protect privacy and boost your site's conversions.
Why traditional CI/CD fails for LLMs (and the release gates we built to fix it)
Discover how to prevent silent AI regressions using LLM release gates with baseline evals, drift detection, and shadow validation.
OpenClaw’s new app doesn’t run AI on your phone. That’s the whole point.
The OpenClaw project launched iOS and Android apps to let users manage their agents from their phones.
Cordyceps flaw pattern is more proof CI/CD is part of the attack surface
Security researchers weigh in on why normal code review misses CI/CD risks and how developers should rethink pipelines.
Cloudflare wants to build the economic layer of the AI web
Cloudflare expands its AI web tools with Pay Per Use, crawler classification and answer engine optimization to become the economic infrastructure layer between publishers and AI companies.
“You Only Compute Once”: How Clockwork wants to put an end to AI training restarts
Clockwork's TorchPass shifts a training job onto a healthy spare the moment hardware fails, sparing teams the hours of recompute that a checkpoint restart costs.
The call is coming from inside your pipeline: the anatomy of a Codecov attack
The Codecov breach exposed a structural flaw in CI/CD security: Your pipeline holds all your secrets. Here's why it's your weakest perimeter.
How Anthropic is bringing Fable 5 back — and when it’ll cost you
The price of bringing Fable 5 back: safeguards so tight the model now flags some legitimate coding requests.
Fable is coming back: Federal government lifts export controls on Anthropic AI model
The U.S. Commerce Department is lifting the export controls it placed on Anthropic's Fable and Mythos models, clearing the way for Fable's return.
Anthropic’s Claude Sonnet 5 system card says more about the future of AI than its benchmarks do
Anthropic's Claude Sonnet 5 system card reveals that AI agent reliability — not benchmark scores — is the next major infrastructure challenge for engineering teams.