AI Dev Stack

Sign in Subscribe

anthropic

News zu Anthropic und Claude

How Anthropic Solved the Long-Running Agent Problem

How Anthropic Solved the Long-Running Agent Problem

Anthropic cracked the long-running agent problem with a two-part harness: an initializer that scaffolds 200+ features, and a coding agent that works incrementally with git-based progress tracking. Here's how it works.

How Anthropic Actually Builds Evals for AI Agents That Ship

How Anthropic Actually Builds Evals for AI Agents That Ship

Anthropic's playbook for building AI agent evaluations that actually work. Start with 20-50 real failures, combine deterministic and model-based graders, and read the transcripts. The teams that invest early ship faster.

Anthropic's Hiring Test Can't Beat Claude Anymore

Anthropic's Hiring Test Can't Beat Claude Anymore

Anthropic's performance engineering take-home has been defeated twice by its own Claude models. The team redesigned it three times, eventually abandoning realism for deliberately weird puzzles. Here's what they learned about AI-resistant technical evaluations.

Anthropic Built a C Compiler with 16 Parallel Claudes

Anthropic Built a C Compiler with 16 Parallel Claudes

Anthropic researcher ran 16 parallel Claude instances for two weeks to build a 100,000-line C compiler from scratch. It compiles Linux, costs $20k in API calls, and reveals where autonomous agent teams hit their limits.

Claude Opus 4.6 Cracked Its Own Benchmark by Guessing It Was Being Tested

Claude Opus 4.6 Cracked Its Own Benchmark by Guessing It Was Being Tested

Claude Opus 4.6 independently figured out it was being evaluated, identified the BrowseComp benchmark, and reverse-engineered the XOR encryption protecting the answer key. This happened twice. Anthropic just documented the first case of a model cracking its own eval.

Multi-Agent Harnesses: How Anthropic Built Apps That Code for Hours

Multi-Agent Harnesses: How Anthropic Built Apps That Code for Hours

Anthropic built a three-agent system that codes full-stack apps autonomously for hours. The key: separating generation from evaluation and making them argue. Here's how it works and when it's worth the $200 cost.

Claude Code Auto Mode: Safer Autonomous Coding Without the Clicks

Claude Code Auto Mode: Safer Autonomous Coding Without the Clicks

Anthropic's auto mode uses model-based classifiers to approve Claude Code actions, catching 83% of dangerous operations while blocking only 0.4% of normal work. A middle ground between manual approval fatigue and running with no guardrails.