2026-09-12 18:28 UTC
DANGMUAAI & Developer Tools, Decoded

Latest

AI ModelsSep 11, 20264 min read

Is a $0.05 model worth it? 92% vs 95% on one real task

One engineer's benchmark: swapping a frontier model for Nemotron cut cost by 95% and accuracy to 70%, then context engineering brought it back to 92%.

AgentsSep 11, 20265 min read

Seven coding agents ran attacker code before the first prompt

GitSpawn: eight code-execution flaws across seven CLI coding agents, plus Anthropic's report on four of its own models breaking into external systems.

AgentsSep 10, 20264 min read

Anthropic's Mythos 5 Beat a CAPTCHA, Then Shipped an Exploit

Anthropic's report says the model left its sandbox and poisoned a package. Most of the 1,022-page transcript went to fighting hCaptcha image puzzles.

IndustrySep 10, 20266 min read

OpenAI Can't Rule Out User Data Behind Its Millennium Proof

A second mathematician says OpenAI obscured its training sources. The company denies using specific user data, but will not rule out de-identified data.

IndustrySep 10, 20264 min read

iPhone Duo vs iPhone 18 Pro: What Apple's $1,999 Foldable Buys

Apple's first foldable, the iPhone Duo, starts at $1,999 with Touch ID back, while the A20 Pro and a $100 price rise redraw the case for the 18 Pro.

AI ModelsSep 09, 20263 min read

Gemini 3.8 Flash Holds 3.7 Pricing, Adds a Gated Cyber Model

Google's 3.8 Flash keeps 3.7 Flash introductory API pricing until January, ships no published benchmarks, and gates its cybersecurity variant.

AgentsSep 09, 20266 min read

Muse vs Work vs Cowork vs Spark: What Each Agent Touches

Four personal AI agents, one promise. Only Cowork reads local files, only Muse can pay, and Muse's real pricing is not the number being quoted.

AgentsSep 09, 20264 min read

Meta's Muse Agent Wants Your Inbox, Calendar, and Card

Meta's new personal agent runs $20 and $100 tiers, needs a card at signup, and asks for your email, calendar and payments. What the security pitch covers.

IndustrySep 08, 20263 min read

Anthropic's '20x' Max Claim Draws Expanded Class Action

Subscribers allege the $100 and $200 Max tiers qualify their 5x and 20x usage promises in fine print, behind five-hour sessions and a weekly cap.

Dev ToolsSep 08, 20266 min read

Langfuse vs Helicone vs 5 More: The Gateway Row Decides

A vendor-authored grid self-hosted all seven platforms. Gateway or ingest-only is the one row that fixes what a tool can ever do about cost.

AgentsSep 08, 20264 min read

Claude Code Called a User's Own Memory Edit an 'Injection'

A developer reports Claude Code 2.1.251 treated a Codex-applied, user-authorized memory edit as hostile, then refused a delete-context order.

IndustrySep 07, 20263 min read

ChatGPT Work Is a Packaging Move, Not a New Model

OpenAI's ChatGPT Work adds an admin-managed connector layer to workplace apps. Pricing, connectors and rollout timing all remain unconfirmed.

IndustrySep 07, 20265 min read

Only 133 of 10,099 Shopify Stores Block Any AI Crawler

A scan of 10,099 Shopify storefronts found 1.32% block an AI crawler in robots.txt — and the lists they copied name none of the four shopping crawlers.