A design essay argues that AI agents must distinguish a successful tool call from a completed customer task, using a hypothetical order A17 address change where the service returns HTTP 202 with reque…
AI research agents can attach real, DOI-resolvable papers to claims those papers never made, and the proposed fix is an "AI citation verification receipt" — a durable, machine-readable record of which…
Developers routinely underestimate LLM costs because they misunderstand three mechanics: tokens are statistical text fragments rather than words, tool-definition JSON schemas are re-sent and re-billed…
Databricks now offers a managed agentic memory store built on Lakebase, letting developers configure memory with a few inputs instead of manually defining a traditional Postgres schema, as of Septembe…
MemCell launched as persistent memory infrastructure for coding agents, letting two agents from different vendors — Antigravity and Claude Code — read from and write to the same versioned project memo…
A 20-case synthetic expense-claim experiment found that a single Jev question applying an entire reimbursement policy matched 17 expected decisions, while a graph using the same model for narrow judgm…
The article's body is unavailable, blocked by a Cloudflare security page, so no verifiable details can be reported about OpenAI's agent accessing Australia's Medicare portal or the claimed 54-day dete…
QLoRA fine-tuning on 150 synthetic examples raised a base Qwen3.5-4B model's adversarial-prompt refusal rate from 79.5% (31 of 39) to 89.7% (35 of 39), while GPTQ and AWQ 4-bit quantization added roug…
A 2025 METR study of 16 experienced developers completing 246 real tasks on codebases exceeding a million lines found that developers using AI tools were 19% slower, despite predicting a 24% speedup b…
Healthcare organizations can violate HIPAA's 45 CFR § 164.502(b) Minimum Necessary Standard without any external data breach, because naive monolithic RAG pipelines retrieve and inject unrelated clini…
Enterprise data breaches involving unsanctioned Shadow AI cost organizations an average of $650,000 more than conventional security incidents, and 20.0% of global enterprises have already suffered a p…
Google's Antigravity SDK now supports local models, letting developers run Gemma on-device via LiteRT or connect to OpenAI-compatible local servers such as Ollama, LM Studio, or vLLM. The LiteRT path …
The July 2026 OpenAI-Hugging Face cybersecurity incident marked the first publicly documented production intrusion executed end-to-end by an autonomous agentic framework, according to Larcher et al. (…
Microsoft Fabric's T-SQL COPY INTO command ingests structured CSV and Parquet files directly from Azure Data Lake Storage Gen2 or Amazon S3 into Fabric Warehouse tables without provisioning a Spark se…
A guide on handling unsatisfactory LLM output argues that systems should validate responses against a strict schema using Pydantic's Literal types and ConfigDict(strict=True, extra="forbid"), since wi…
A developer walkthrough published in the VersusControl devops-ai-guidelines repo shows how to build an AI agent evaluation harness that separates what an incident agent did (tool calls) from what it s…
A University of Cambridge and Google DeepMind team published XConf on 15 September 2026, a confidence-estimation method that reads confidence from a bank of past graded episodes rather than from the m…
TypeSafe released Jev in September 2026, a model that returns probabilities for typed questions instead of generating text, scoring 67.8% versus GPT-5.6 Terra's 67.9% on TypeSafe's own four-workflow e…
OpenAI has introduced Prompt Cache Diagnostics in the Responses API for GPT-5.6 and later supported models, letting teams compare a current response against a recent baseline via prompt_cache_options.…
Google's September Antigravity Agent preview update introduces a new Gemini agent identifier and changes built-in local-tool shapes, moving file operations toward explicit PascalCase parameters and li…