English text is being prepared.

DeepLearning.AI Launches Context Engineering Course, Teaching Agent Context Management

DeepLearning.AI has launched a new Context Engineering course, taught by Chroma CEO Jeff Huber and CMPND CEO Drew Breunig, framing context engineering as the main thread for keeping Agents reliable and token-efficient in complex multi-step tasks. The course builds a harness from scratch in Python and uses traces and evals to quantify each change, covering three capabilities—Agentic Search, Context Management, and Agentic Memory—across four modules, with labs mostly consisting of about 1-hour coding exercises. The course assumes intermediate Python, familiarity with LLM applications and tool calling, and basic retrieval.

Baseten Optimizes Qwen3-ASR Batch Transcription, Boosting Single-GPU Throughput by 7x

The Baseten engineering team, through native audio decoding, KV cache FP8 quantization, and reduced host-device synchronization, increased batch transcription throughput for Qwen3-ASR 1.7B on a single RTX PRO 6000 by 7x and reduced latency by 6.9x. After optimization, the transcription cost for one hour of audio is approximately $0.0013, i.e., 1/10 of a cent per audio hour, with no significant change in word error rate. The optimized version has been made the default deployment in the Baseten Model Library and can be called through an OpenAI-compatible transcription client.

Pydantic compares code mode and tool calling across 5,276 runs

Michael Pfaffenberger of Pydantic used 5 small Pydantic AI applications and 8 models to conduct 5,276 scored runs, comparing code mode with direct tool calling. On long tasks, code mode reduced average cost by 58%, latency by 51%, and increased success rate by 16 percentage points; however, for short tasks with 5 items, 3 out of 5 tasks were instead 21% to 38% more expensive. The author argues that models are better at writing loops than at acting as the loop, and the gap widens as tasks get longer.

NVIDIA Explains How to Fine-Tune Nemotron 3.5 Lightning with the Nemotron-Personas Dataset

NVIDIA has released the Nemotron-Personas France and Belgium datasets and provided a complete post-training workflow for fine-tuning Nemotron 3.5 Lightning with them. The workflow includes defining behavior contracts, generating synthetic conversations with personas, fine-tuning with LoRA on NVIDIA NeMo, merging adapters, and then comparing and evaluating item by item against the base model on a frozen test set. The France dataset was built in collaboration with Pleias, and the Belgium dataset involved Pleias and KU Leuven; the Nemotron repository provides reusable steps and agent skills for data generation, training, and evaluation.

How LangChain Built the Design Agent Content Studio

LangChain's GTM engineering team, working with the design team, built Content Studio, a design agent inside Slack, on Managed Deep Agents. It can generate videos, social graphics, UI mockups, and slide decks, and has already produced more than 3,000 assets and over 80 slide decks. The article breaks down its four-layer knowledge structure (skills, wiki, code validation, memory), the dedicated LangSmith Sandbox for each Slack thread, the two rendering engines HyperFrames and the in-house motion-studio, and how the main agent orchestrates asynchronous sub-agents. The team says that after migrating to Managed Deep Agents, its own code shrank by about 80%, the median model cost per request fell from about $8 in mid-September to about 50 cents, and the entire harness has been open-sourced.

Browser Use Tutorial: How to Log In Daily and Download Invoices on Websites Without an API

Browser Use has published a tutorial introducing three approaches for automatically logging in and downloading invoices daily on supplier portals that lack an API: Playwright scripts, RPA tools such as UiPath or Power Automate, and browser agents with saved configurations. The article provides complete code for Playwright and Browser Use Cloud, covering reusing login sessions, filling in TOTP verification codes with 1Password, scheduling runs via GitHub Actions, and handling failure scenarios such as session expiration, 2FA blocks, page redesigns, and empty downloads. The author recommends first asking suppliers for exports, APIs, or emailed invoices, and reminds readers to first confirm whether the portal's terms permit automated access.

AWS explains prompt caching patterns for using OpenAI models on Amazon Bedrock

AWS has published a tutorial explaining how to use prompt caching to reduce the cost of agents repeatedly reading context when calling OpenAI models on Amazon Bedrock, distinguishing between two approaches—implicit caching and explicit prompt_cache_breakpoint—and recommending their combined use. The article provides code examples for tool and instruction caching, rolling breakpoints across multi-turn history, adding tools mid-session, and shared prefixes across sub-agents, and states that in a six-turn workload on GPT-6 Astra, implicit caching reduces input costs by about 62%, while explicit rolling breakpoints reduce them by about 61%. Cacheable prefixes must be at least 1024 tokens, and cache writes are billed at 1.25 times the normal input price.

LangChain summarizes the five components for building company-level agents

Harrison Chase and Caspar Broekhuizen of LangChain published an article summarizing the five core components shared by its three internal agents for marketing, sales, and engineering: agents live in existing workplaces such as Slack, a distinction between acting on behalf of users and a fixed identity, global and user-level memory, code execution within a sandbox, and a large number of runs triggered by schedules or events. The article uses GTM Agent, OpenSWE, and Content Studio as examples to explain their specific approaches to identity, memory, and code execution, all three of which are built on Managed Deep Agents.

Sebastian Raschka releases an 11-hour video course on building a reasoning model from scratch

Sebastian Raschka has released 8 programming videos on building a reasoning model from scratch, totaling 11 hours, using Python and PyTorch to implement evaluation, inference-time scaling, reinforcement learning, and distillation starting from a pretrained Qwen3 0.6B model. The course first builds a math answer verifier and evaluates on MATH-500, then implements temperature sampling, top-p filtering, and self-consistency, followed by RLVR training with a simplified version of GRPO, and finally distills a small model using answers generated by DeepSeek R1. The content accompanies his book "Build a Reasoning Model (From Scratch)"; the early text generation and evaluation examples can run on a CPU, while later training experiments are recommended to use a GPU.

SemiAnalysis benchmarks Trainium3's PyTorch training and Neuron Explorer profiler

The SemiAnalysis engineering team benchmarked the native PyTorch training stack TorchNeuron and TorchTitan on Trainium3, training Qwen3-32B on 16 Trainium2 chips; eager mode reached 11,099 tokens/s and 23.89% MFU, while compiled mode reached 14,156 tokens/s and 30.47% MFU. The author considers the Neuron Explorer profiler superior to Nvidia's and AMD's solutions, but first-time compilation takes too long: the first-step compilation for the gpt-oss training recipe took 27 minutes 17 seconds. FlexAttention tests passed 89 of 96 cases, with all 7 failures due to compilation timeouts rather than numerical errors.

TinyFish Breaks Down Production-Grade Agent Search Architecture and Deep Research Experience

TinyFish released two technical guides: one defines a production-grade agent search architecture, and the other summarizes lessons from building a deep research system. The architecture is built on four web primitives—TinySearch, TinyFetch, TinyBrowser, and TinyAgent—and runs a seven-stage control loop around a foundation model, covering task definition, retrieval planning, tool routing, evidence evaluation, and failure retry. In OpenBenchmarks tests with a fixed gpt-5.6-sol model and 100 tasks, TinyFish search plus fetch achieved a 79.0% completion rate, with a median task cost of $0.050 and a runtime of 24 seconds; under a search-only configuration, the completion rate was 59.3%.

Fireworks Offers Free Training of Muse Glimmer 30B: Using SFT to Predict Kalshi Market Outcomes

Fireworks AI announced that Muse Glimmer 30B can be trained with SFT for free until November 30, and released a complete recipe for a prediction experiment using historical Kalshi market data. The fine-tuned Glimmer achieved 58.3% accuracy on unseen topics, 11.2 percentage points higher than Jev's 47.1%; in a historical backtest, a hypothetical $1,000 grew to $1,071 after accounting for simulated fees, while both base Glimmer and Jev lost money. Training used LoRA adapters and a K-way cross-entropy loss, ran through the training data only once, and separately used a calibration set to fit a temperature parameter for each question category.

Aviator CEO shares how to replace line-by-line code review with intent verification

Ankit Jain, founder and CEO of Aviator, explains that as the share of AI-generated code rose and PR volume reached four times the level of a year earlier, his team abandoned reading diffs line by line. Instead, they review acceptance criteria, constraints, and edge cases before code generation, and give a deterministic conclusion with evidence for each criterion. The team also turned recurring review comments from historical PRs into reusable invariant rules, enforced by machines on every change—for example, always use the Money type for anything involving currency, and never write directly to the users table. The author says this transition took about three months, after which code review was no longer a delivery bottleneck.

Simon Willison built Battle Drones, an audio-networked RTS game, with Claude Code

Simon Willison released Battle Drones, a real-time strategy game developed entirely through vibe coding with Claude Code. It went through 12 major versions and 321 commits and runs in mobile and desktop browsers. The game includes 9 building types, 7 ground vehicles, 6 air vehicles and 3 infantry units, and supports fog of war, random map generation, AI versus AI, and an Agent battle mode in which two LLMs fight each other. He also implemented a multiplayer method in which two devices synchronize game state via microphone audio, and because the audio caused headaches, he added online multiplayer based on a WebSockets relay server.

AWS Publishes Agent Registry Runtime Tool Discovery Practice

AWS introduces the Agent Registry runtime tool discovery approach: agents use hybrid semantic and keyword search to find approved MCP servers and A2A agents, then instantly build clients from descriptors to execute requests. The sample code is based on Amazon Bedrock AgentCore Gateway and runtime, integrates an order management MCP server and two A2A agents for pricing and customer service, and demonstrates a return and refund scenario using pure semantic matching. The registry returns only connection metadata; invocation credentials must be obtained separately through Cognito OAuth or IAM, and new capabilities require no changes to orchestrator code.

How to run n8n AI workflows on NVIDIA DGX Spark without calling cloud APIs

n8n Blog published a tutorial on how to deploy a complete model stack locally on a single NVIDIA DGX Spark, enabling n8n workflows to perform transcription, OCR, document Q&A, and image generation without calling cloud APIs. The entire stack exposes an OpenAI-compatible endpoint through LiteLLM; the main model is the NVFP4 version of Qwen3.6-35B-A3B, with a single-user generation speed of 120 tokens/s and an aggregate of 359 tokens/s across eight users. The author's testing found that image generation reduces chat speed to about one-third, and recommends that small teams queue image tasks or schedule them outside working hours.

RTX Spark laptop tested: How does Nvidia's first local AI laptop chip perform?

heise online's c't 3003 tested the Asus ProArt P14, which is equipped with Nvidia's first laptop chip, the N1X (brand name RTX Spark). The machine supports up to 128 GB of unified memory; the test unit had 32 GB of memory plus a 1 TB SSD and cost 3,300 euros. In llama-bench tests with Qwen 3.8-27B, generation speed was about 14.5 tokens/s, slightly faster than Apple M6 and Strix Halo but slower than M5 Pro; with the MoE model Qwen 3.6-35B-A3B, it reached nearly 61 tokens/s. Prompt processing speed was better than Strix Halo and the M5 Pro and M6, but fan noise under full load reached up to 2.7 sone, and the fans also spin up when idle. The author also tried installing Ubuntu; after manually adding the drivers, both models ran 14% and 9% faster under Linux than under Windows.

English translation is being prepared.