What shipped in AI engineering
Catch up on any month: model releases, agent tooling, inference and RAG, grouped by area with one-line summaries.
November 2026 (so far)
339 posts
- Q&A: Enabling more human-like interactions and understanding in multimodal foundation models
Researchers from MIT‑IBM have introduced data‑efficient benchmarks and techniques to surface latent multimodal knowledge in foundation models without heavy annotation.
- The agent evaluation playbook: Built-in, DeepEval, and AutoEval on Amazon Bedrock AgentCore Evaluations
The article outlines how to evaluate Amazon Bedrock agents using AgentCore’s built‑in evaluators (e.g., Builtin.GoalSuccessRate, Builtin.Helpfulness) together with third‑party DeepEval and AutoEval evaluators hosted in managed mode.
- Amazon SageMaker AI now detects latency drift, idle GPUs, and missed optimizations on your inference endpoints
Amazon SageMaker AI now automatically monitors generative AI inference endpoints (vLLM containers on GPU instances) for latency drift, idle GPU capacity, and missing optimizations, surfacing findings via AWS Health.
October 2026
380 posts
- The Hidden Cost of LLM Routing: Why Switching Models Costs More
The article investigates LLM routing and finds that switching models during a session can negate the cost savings expected from using cheaper inference models.
- Introducing Reflection Beam: an open-weight model for agentic coding on AWS
Reflection AI unveiled Beam, a 501 B‑parameter sparse mixture‑of‑experts model (23 B active parameters) designed for agentic coding and long‑horizon software tasks, now released with open weights on Hugging Face under Apache 2.0.
- Push-based KV transfer to improve TTFT and TPOT for disaggregated inference in vLLM
vLLM’s new push‑based NIXL connector moves KV‑cache transfer off the critical path of disaggregated prefill/decode inference, cutting time‑to‑first‑token (TTFT) by up to 3× and improving time‑per‑output‑token (TPOT) by up to 30 % for…
September 2026
216 posts
- Anthropic's Claude Fable 5.1 and Mythos 5.1 arrive with a 75% cost reduction for Fable cache reads
Anthropic unveiled Claude Fable 5.1 (GA) and Claude Mythos 5.1 (restricted)—the same model with differing safeguards—and announced a 75 % cost reduction for cached‑context reads, enabling far cheaper persistent agents.
- Automated agent evaluation with Amazon Bedrock AgentCore and GitHub Actions
This article demonstrates how to build a CI/CD quality gate for Amazon Bedrock AgentCore agents by automatically deploying the agent with role‑based MCP tools via CDK, running on‑demand evaluations through the AgentCore Evaluate API,…
- Benchmarking small LLM inference on SageMaker AI: G7 vs G5 and G6
This article benchmarks two 30B Mixture‑of‑Experts models on Amazon SageMaker AI inference across G5, G6, and the newer G7 GPU instance families, showing that the Blackwell‑based G7 (RTX PRO 4500) can achieve superior throughput,…
August 2026
59 posts
- The fix for the AI agent that hijacked a company's DNS: it can propose the change, but it can't approve it
The article reveals that an AI agent, designed to automate infrastructure changes, was exploited via a prompt‑injection payload that rewrote a company’s DNS records.
- Perplexity partners with Nvidia to launch Portable Computer, a fully local AI agent with zero token costs
Perplexity’s new Portable Computer lets its “Computer” agent run entirely on‑premise using Nvidia hardware—specifically Nvidia DGX Spark desktops or Linux machines equipped with Nvidia RTX GPUs—by loading a 27‑billion‑parameter Qwen…
- Can an LLM Forget the Right Things?
The article introduces a hand‑written CUDA inference runtime tailored for Vision‑Language‑Action (VLA) robots that explicitly manages memory by deciding what to remember, what to forget, and when inference should be aborted.