SGLang 0.5.19 in Practice: Beam Search, a Faster Cold Start, and What Still Breaks
SGLang’s 0.5.19 headline is beam search, and it will not run on a disaggregated deployment with speculative decoding. The change I would…
Anthropic Cuts Fable 5.1 Cache Reads 75% to $0.25/M — and Breaks Three Prompt Patterns Developers Rely On
Anthropic just made reusing your system prompt 75% cheaper — and made editing it an HTTP 400. On Fable 5.1, the cheapest prompt is the one you…
The Harness, Explained: Why the Same Scaffold Scores 65% or 74% on SWE-bench
A 100-line Python harness scored 65% on SWE-bench Verified in July 2025. The same project now claims above 74% — and boots faster than Claude…
StackOne Defender 0.8.2 in Practice: An Apache-2.0 Prompt-Injection Filter for Tool Calls — What Works, What Does Not
A bundled ONNX classifier now sits between your agent and its tools, catching a vendor-reported 88.7% of prompt injections. The interesting…
Grok Still Leaks Full Chat Histories 11 Weeks After Disclosure—AES-256 Encrypted Prompts Beat Guardrails 40% of the Time
Grok’s safety classifier reads your prompt. It does not read what Grok decrypts afterwards. Adversa AI shipped attack instructions as…
Apiiro CLI Launches April 9 Turning AI Coding Assistants Into Security Engineers by Rewriting Developer Prompts in Real Time
Your AI coding assistant now has a shadow editor—one that rewrites your prompts before the AI model sees them, injecting security constraints…
PromptEng 2025 Workshop Accepts 7 Research Papers on January 24—First Academic Forum Dedicated to Prompt Engineering Techniques
While engineers debated whether prompt engineering was “real engineering,” seven research teams quietly got their papers…
Handshake58 Launches on Bittensor Subnet 58: $0.0001 Micropayments for AI Agents Using Polygon USDC and Zero-Gas Vouchers
The AI industry spent 2025 building agents that can reason, plan, and act—but forgot to give them wallets. Handshake58 just shipped the…
xAI Launches Grok 4.20 Beta with Four-Agent Architecture—65% Hallucination Reduction Shifts Prompt Engineering From Iterative Chat to Structured Contracts
The iterative prompt refinement loop that defined AI workflows for three years just became a liability. Grok 4.20’s four-agent system…
Calibrated Confidence Prompting: The Silent Shift from Asking Better to Trusting Smarter
Everyone’s arguing prompt engineering is dead. Meanwhile, engineers who actually ship LLMs to production discovered something more…
