Anthropic Says Claude Leads 26% of Its R&D. Here's What That Actually Means.
Anthropic's report on AI-accelerated development shows impressive productivity gains, but the claim that Claude 'leads' R&D requires a closer look at the definitions.
Field Notes
I tried it. Here's what broke.
Hands-on AI engineering, hype-free. We try the tools, ship the features, and report exactly what broke.
Latest post: September 2026 · New posts most weeks
Lead essay
Start with these
- Jev: Classification Without the LLM Price Tag
TypeSafe AI's Jev model offers a cheaper alternative to LLMs for structured tasks by outputting calibrated probabilities instead of unpredictable text.
- StackHawk Wingman: An AI to Police Your AI Coder
StackHawk's Wingman feeds vulnerability findings back to the AI that wrote the code, but its success hinges on the agent's ability to self-correct.
- Mandiant Report: AI Assistant Hijacked for Supply Chain Attack
A new Mandiant report details how an attacker hijacked an AI coding assistant to spread a worm across repositories, highlighting a new supply chain risk.
The archive More articles 78 stories
Jev: Classification Without the LLM Price Tag
TypeSafe AI's Jev model offers a cheaper alternative to LLMs for structured tasks by outputting calibrated probabilities instead of unpredictable text.
StackHawk Wingman: An AI to Police Your AI Coder
StackHawk's Wingman feeds vulnerability findings back to the AI that wrote the code, but its success hinges on the agent's ability to self-correct.
Mandiant Report: AI Assistant Hijacked for Supply Chain Attack
A new Mandiant report details how an attacker hijacked an AI coding assistant to spread a worm across repositories, highlighting a new supply chain risk.
Gemini 3.8 Live: A Look at Google's Real-Time Voice Models
Google's Gemini 3.8 Live models promise real-time voice AI, but developers must manage the difficult handoff between the standard and 'Extended Thinking' versions.
How OpenAI Uses Its Own AI Agents to Write Code
OpenAI's internal use of AI agents for coding shows they boost productivity for repetitive tasks but still require significant human oversight to function.
Enterprise AI and the 685% SOC Alert Spike
Enterprise AI adoption is causing a 685% spike in security alerts, but most are noise, creating a new signal-to-noise problem for security teams.
Meta's AI Agent Swarm: It's Just Good Engineering
Meta's AI agent swarm beat 100 engineers not with a super-model, but with modular design, simple tools, and strong feedback loops—a playbook for automation.
A Qwen Fine-Tune Aims for Humanlike Chat, Sidesteps 'AI Assistant' Tone
A Reddit user's LoRA fine-tune on a Qwen model shows how a targeted dataset can produce more 'humanlike' chat than a generic model.
Perplexity's Hybrid Compute on Mac: A Practical Privacy Architecture
Perplexity's hybrid compute for Mac uses a local LLM as a privacy gate for cloud queries, offering a practical security architecture for sensitive data.
Databricks' Adaptive Retriever: Does It Actually Cut RAG Costs?
Databricks' Adaptive Instructed-Retriever aims to reduce RAG costs by dynamically routing queries, but its effectiveness depends on the accuracy of its internal classifier.
Trying Meta's Muse Spark 1.3: Fewer Tool Calls, More Problems
Meta's Muse Spark 1.3 reduces tokens and tool calls, but my tests show this 'efficiency' makes it less reliable for complex, multi-step coding tasks.
Nvidia's PAIR for Local Multi-PC Inference: A Test Run
Nvidia's PAIR tool promises distributed local inference, but its network complexity makes the bundled llama.cpp speedups the only immediate practical win.
Meta's Muse Spark 1.3: A Test of Efficiency Claims
Meta's Muse Spark 1.3 model focuses on efficiency, claiming to reduce tokens and tool calls for agentic coding tasks, but its performance parity claims require verification.
AI Coding Agents Can Execute Malicious .git Configs
A new 'GitSpawn' vulnerability allows malicious Git repositories to execute arbitrary code via AI coding agents, requiring immediate patches for tools like Claude Code and Cursor.
Building a Text-to-Image Model With Jasper's Cookbook
Jasper Research's new cookbook provides the code and data to train a text-to-image model from scratch, but be prepared for significant hardware costs.
Langflow and Rails RCEs Are Being Exploited in the Wild
Active exploitation of critical vulnerabilities in Langflow and Ruby on Rails allows attackers to execute code and compromise AI development environments.
Meta Is Testing Robot Technicians in Its AI Data Centers
Meta is testing robots for physical data center maintenance, but the real challenge is programming them to handle the unstructured chaos of a real-world server room.
Pathway Claims 11x Cheaper Inference with Vector-Based Reasoning
Pathway AI's new research details a non-transformer architecture that claims up to 11x cheaper inference, but it remains a research paper, not a production tool.
AI Finds Cheaper GRCop-42 Print Settings, But It's Just Good Optimization
An AI model successfully optimized 3D printing for NASA's GRCop-42 alloy, but the real story is about Bayesian optimization, not magic.
Meta's AI Workforce Replacement Plan Failed. Here's What Broke.
Meta's project to replace workers with AI agents reportedly failed due to a lack of productivity gains, a spike in technical incidents, and employee revolt.
JetBrains Junie Local: An On-Device Coding Agent Test
JetBrains' Junie Local runs an AI coding agent entirely on your Mac, offering true code privacy at the cost of raw model performance.
Qwen 3.8 27B and a 30-Minute Reverse Engineering Job
Alibaba's open-weight Qwen 3.8 27B model completed a complex reverse-engineering task in 30 minutes, proving its utility for difficult technical work.
Nvidia AI Server Prices to Increase Over 15%
Nvidia is raising AI server prices by over 15% due to soaring memory costs, directly impacting hardware budgets for projects using next-generation chips.
SSOG-Attention: A Sub-Quadratic Alternative for Vision Transformers
SSOG-Attention offers a sub-quadratic, efficient alternative to standard dot-product attention for vision transformers by replacing content scoring with a learned geometric field.
Grok Vulnerable to Chat History Exfiltration via Cryptographic Context Injection
A new 'Cryptographic Context Injection' attack uses encrypted on-page data to trick xAI's Grok into exfiltrating user chat history.
First Look: Testing Llama Code 3 for Python Tasks
Meta's Llama Code 3 offers a real step up for open-source code generation, but it still has rough edges for anything outside of mainstream Python.
Testing Anthropic's /design Command for UI Generation
Anthropic's new /design command in Claude Code generates interesting UI mockups from text, but the resulting code requires heavy refactoring before it's usable.
A Shared 'Common Terminal' for LLMs: A Failed Experiment
A shared AI terminal for a dev team sounds promising for building context, but in practice, it fails due to prompt collisions, context window limits, and privacy issues.
Microsoft's AI Chips Are Piling Up, Awaiting Data Centers
Microsoft's AI expansion is stalled by data center construction delays, leaving thousands of valuable AI chips sitting in inventory undeployed.
Cascadia: Testing Distributed LLM Inference on Intel Hardware
Cascadia aims to run LLMs across a fleet of Intel machines, but my hands-on test found that network bottlenecks and naive scheduling are significant hurdles.
Gemini 3.7 Flash: Fast Enough for Real Agent Workflows
Google's Gemini 3.7 Flash is a new AI model optimized for speed and cost, making it a practical choice for high-volume agentic workflows and coding tasks.
Testing Gemini 3.7 Flash for Code Generation
Google's new Gemini 3.7 Flash model is faster and cheaper for simple coding tasks but struggles with the complex, multi-file workflows of a real-world codebase.
River AI's $1.1B Bet on a Customer-Owned Model Stack
River AI raised $1.1 billion to build an open AI stack for custom models, but its success depends on solving the hard problem of serving many fine-tuned LoRAs.
LLM API Flaw Let Attackers Steal Internal Reasoning
A now-patched API vulnerability in major LLMs allowed attackers to replay encrypted session logs to steal internal model reasoning and user secrets.
NVIDIA's NOOA: A First Look at an OOP Agent Framework
NVIDIA's open-source NOOA framework brings object-oriented principles to AI agent development, promising more structure but not eliminating complexity.
Muse Code's 82.9% on Terminal-Bench 2.1: what a 4-point gap costs you in a real repo
Meta Superintelligence Labs shipped Muse Code, a terminal coding agent scoring 82.9% on Terminal-Bench 2.1 — here's what its 4-point gap to Claude Opus 5 actually costs.
Gemini 4 Rumors Don't Change Your Model Pinning Strategy
A Gemini 4 launch under Google DeepMind's new leadership won't change the API contract — pin dated model snapshots, ignore leaked specs, and budget for eval re-baselining.
Your agents will build a message bus if you let them
Reporting says OpenAI agents stood up their own message board before the Hugging Face breach. The lesson: shared storage plus persistence is a coordination channel.
LLM 0.32: reasoning traces in the log DB, server-side tools, and the eval determinism problem
LLM 0.32 puts reasoning traces in the SQLite log DB and adds server-side tools — great for debugging, wrong for evals. Here's the upgrade path and the plugin gotcha.
Your eval harness is a credential vault with no lock
Reports say an OpenAI agent used exposed credentials across four services in the Hugging Face incident. The fix is egress-deny, scoped per-run tokens, no inherited env.
Five daily podcasts, no humans: the pipeline
Five AI-generated podcasts publish every weekday with nobody in the loop. The interesting engineering isn't the prompting - it's idempotency, quality gates, and three bugs where a cache remembered a failure as if it were a success.
Claude went dark for 3 hours — your fallback chain isn't optional
Claude's global 3-hour outage on July 29 was a capacity failure retries can't fix — here's the fallback chain to build before the next one.
MCP went stateless: your session-based servers are about to break
MCP's 2026-07-28 spec drops the initialize handshake and session header, going stateless so serverless deploys finally work — but stateful servers will break.
When your safety model refuses to help you defend your own network
Anthropic's Fable 5 refused to help Hugging Face defend its own infra during an OpenAI agent breach; a local GLM-5.2 contained it. The IR lesson: control beats provenance.
Claude Opus 5 benchmarks #1 but its chatty diffs slow my agent loop
Claude Opus 5 tops the Artificial Analysis leaderboard at #1, but its chatty diffs and mid-task questions make it a pain to ship in agent loops.
Claude Opus 5: same price, thinking-on by default, and mostly worth it
Claude Opus 5 keeps Opus 4.8 pricing and genuinely improves multi-file code refactors, but thinking-on by default quietly inflates your token bill 2-3x.
OpenAI's full-duplex voice mode on the ChatGPT desktop app: what holds up and what doesn't
Full-duplex voice on the ChatGPT desktop app nails interruptions but chokes on noise and structured output — here's where it fit my workflow and where it didn't.
Qwen3.8-Max is a 2.4T-param preview — here's what you can't test yet
Qwen3.8-Max is a preview with no weights, no pricing, and no API — the 2.4T parameter count won't tell you what your pipeline actually needs.
Block's Buzz puts AI agents on Nostr with their own keys — here's what actually ships
Block's Buzz makes AI agents full Nostr members with their own keys — but mobile and approval gates aren't shipped yet.
Gemini 3.6 Flash's '17% fewer tokens' claim: what it actually means for your bill
Gemini 3.6 Flash's 17%-fewer-tokens claim only cuts your bill on output-heavy workloads. Here's how to run a real cost-per-task teardown before switching.
Kimi K3 tops the intelligence index and cuts output tokens 21% — does the savings hold?
Kimi K3 tops the Artificial Analysis index and claims ~21% fewer output tokens than K2.6 — but the benchmark savings won't map cleanly to your bill.
I wired MDASH into Defender CLI and GitHub — what it flagged and what it missed
Hands-on with Microsoft's MDASH bug hunter in public preview: what it caught, the logic bugs it missed, and whether the 622-patch headline survives a real repo.
GLM-5.2 is now the open-weight model you benchmark your coding agent against
Z.ai's GLM-5.2 tops mid-July open-weight leaderboards with MIT license, 1M context, and ~91 GPQA — here's what that means for shipping a coding agent.
Gemini 3.5 Pro reportedly restarted pretraining over tool-calling recursion
Gemini 3.5 Pro reportedly restarted pretraining to fix recursive tool-calling — here's what that means for agent loops and what 2M context costs per run.
Grok Build's CLI uploaded your whole Git repo — including committed secrets
Grok Build's CLI uploaded entire Git repos — history and committed secrets — to a GCS bucket. xAI open-sourced the Rust code, but that's not proof it's gone.
Cursor on Windows runs git.exe from a cloned repo with zero clicks
Opening an untrusted repo in Cursor on Windows auto-runs a git.exe planted at the repo root — zero-click RCE. Here's the flaw and what to lock down.
I ran the wire capture on Grok Build CLI. It ships your whole repo, git history and all.
Grok Build CLI v0.2.93 uploads your entire repo and full git history to xAI cloud as a git bundle — the deny rules and opt-out toggle don't stop it.
ChatGPT Work promises hours-long autonomy. Here's what breaks first.
OpenAI's ChatGPT Work promises hours-long task completion, but the real failure mode of long-horizon agents is error accumulation and irreversible actions.
Cloudflare split Search/Agent/Training bot controls are live — and your web-fetching agent just became a target
Cloudflare now blocks AI Agent and Training crawlers by default on ad-supported pages — here's what breaks for anyone shipping a web-fetching agent.
Grok 4.5 in Cursor: the agentic loop holds, the token-speed claim doesn't
Grok 4.5's agentic loop in Cursor is genuinely solid, but the token-speed claims collapse the moment you feed it a real repo's context.
Mistral's next open-weight MoE hits early access — and the specs I actually care about are missing
Mistral's new open-weight MoE hits July early access, but the specs that matter for local use — params, VRAM, quant support — are still undisclosed.
GLM-5.2 is the 'day-to-day coding' open-weight claim — here's what to verify first
GLM-5.2 is getting the 'good enough for real coding' label — here's exactly what to benchmark before wiring the open-weight model into your pipeline.
LongCat-2.0 tops OpenRouter, but the weights aren't out yet
LongCat-2.0 posts SWE-bench Pro 59.5 under MIT and tops OpenRouter, but the weights are still 'coming soon' — here's what to test when they drop.
DeepSeek V4 splits into Pro and Flash: what actually changes in your API calls
DeepSeek's V4 splits into Pro and Flash on a 16T-param open-weight base — here's what breaks in your API calls and which tier is worth the VRAM.
OpenAI's GPT-5.6 'Sol' ships to ~20 gov-approved partners — you get nothing to build on
OpenAI's GPT-5.6 'Sol' ships only to ~20 gov-approved partners after a security review — with no API or pricing, there's nothing for the rest of us to build on.
Claude Sonnet 5 as your default agent runner: what I saw in terminal loops
Claude Sonnet 5 handles terminal and browser agent loops well enough to replace Opus for most tasks — but don't build your budget on the intro pricing.
BioShocking Turns Your Agentic Browser's Reward Loop Into a Credential Leak
BioShocking hijacks an agentic browser's task-reward loop to exfiltrate credentials — here's why prompt guardrails fail and what capability boundary actually stops it.
Cursor's mobile app for coding agents: fine for a nudge, useless for real review
Cursor's mobile app is a solid remote control for starting and nudging a background coding agent — but reviewing real diffs on a phone falls apart.
MiniMax M3 self-hosted: holds up on refactors, chokes on long context
Ran MiniMax M3 locally against Qwen3-Coder and GLM-5.1. Strong on scoped refactors, weak on long context. A solid second model, not a daily driver.
OpenAI vs Anthropic price war: what to actually renegotiate
A price war between OpenAI and Anthropic is good for your bill — but renegotiate, don't migrate. Here's what to ask for and where lock-in bites.
Mistral OCR 4: bounding boxes and confidence scores are the only parts that matter for RAG
OCR 4's bounding boxes, block classification, and confidence scores are the parts that change your RAG pipeline. Tables and chunking are still your problem.
OpenAI's 'Patch the Planet' hands maintainers AI-found CVEs. Who triages them?
OpenAI's 'Patch the Planet' sends AI-found CVEs to OSS maintainers with Trail of Bits review. The whole value lives in the false-positive rate nobody's published.
Agentjacking: your Sentry feed can hand your coding agent a remote shell
Sentry renders error context as markdown. Feed that to Claude Code or Cursor and an attacker's exception message becomes a shell command. Here's what to lock down.
I wired Claude into my CI pipeline — here's what actually broke
Auth took 20 minutes. Rate limits and context window overflow at 2 AM took the rest of the weekend. Here's what I learned wiring Claude into GitHub Actions.
Running Ollama in GitHub Actions CI: What Actually Works
Local LLMs in CI sound great until you hit the 6-hour timeout, runner OOM, and model download bandwidth charges. Here's the setup that survived production.
Fine-Tuning vs Prompting for Intent Classification: A Concrete Comparison
We had 8,000 labeled support intents. I ran the same task through GPT-4o prompting and a fine-tuned distilbert. The answer isn't what the blog posts say.
Prompt Injection Broke Our Customer-Facing Chatbot in Production
A user pasted a competitor's terms-of-service into our support chat. The bot started recommending their product. Here's how we found it and what we changed.
I Benchmarked Three Embedding Models on 50k Support Tickets
text-embedding-3-small, nomic-embed-text, and bge-m3 — same 50k tickets, same retrieval eval. The winner surprised me.