AI news for builders and product teamsUpdated Nov 18, 2026, 10:01 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Podium used LangSmith to test and fine-tune its AI Employee agent, improving F1 response quality from 91.7% to 98.6% and reducing the need for engineering intervention by 90%. The company also enabled its Technical Product Specialists to troubleshoot agent behavior without engineering help.

Factory used self-hosted LangSmith to export traces to AWS CloudWatch, link feedback to individual LLM calls, and automate prompt optimization, reporting a 2x improvement in iteration speed. Factory also reports about a 20% reduction in open-to-merge time and 3x less code churn on Droid-impacted code in the first 90 days.

LangSmith LLM Gateway is now in public beta for Plus and Enterprise plans, offering centralized runtime controls for production agents including spend caps, rate limits, model fallbacks, and PII redaction. It supports multiple model providers and is BYOK-first, with PII redaction limited to Enterprise users.

LangChain outlines how LangSmith and LangChain OSS can help teams meet EU AI Act requirements ahead of the August 2, 2026 compliance deadline for high-risk AI systems. It maps observability, evaluation, human oversight, and data residency features to specific Act articles.

LangChain announced a self-improving evaluator feature in LangSmith that stores human corrections to LLM-as-a-Judge outputs as few-shot examples, which are then fed back into future prompts to align evaluations with human preferences without manual prompt engineering.

LangChain released a stable LangGraph v0.1 framework for building agentic and multi-agent applications and announced LangGraph Cloud, infrastructure for scalable, fault-tolerant agent deployment, now in closed beta. A note says LangGraph Platform was renamed LangSmith Deployment as of October 2025.

Replit built its Replit Agent on LangGraph and integrated LangSmith, pushing the platform to handle very large agent traces. LangChain and Replit collaborated on new LangSmith features: improved ingestion and frontend rendering for large traces, search and filtering within traces, and a thread view for human-in-the-loop workflows.

Goodfire, a San Francisco AI lab founded in 2024, has made its Silico mechanistic interpretability platform generally available to the public and announced a grant program offering US $1 million in free Silico usage for academic and nonprofit interpretability researchers.

MIT researchers developed CrysVCD, a framework that applies valence-constrained design before material generation to boost chemical stability. In Nature Computational Science, they report nearly 70 percent lattice-dynamics stability, 68 percent mechanical stability, and 85 percent metastability when fine-tuned, and generated candidates with high thermal conductivity or high dielectric constant.

Sentence Transformers v6.0 introduces a fourth model type, MultiVectorEncoder, for ColBERT-style late interaction retrieval, along with a full training approach. A Hugging Face blog post documents finetuning multi-vector models, including a medical retrieval model trained in 14.5 hours on one RTX 3090.

IBM's Granite Team published a technical walkthrough of Granite 4.2, its first family of dense, decoder-only reasoning LLMs, released in 3B, 8B, and 30B sizes under the Apache 2.0 license. The models are post-trained from Granite-4.1 base models via SFT and a multi-stage RL pipeline.

Anthropic has added remote control updates to Claude Code, letting users start a new session from their phone, with dropped laptop-to-phone connections recovering automatically and faster session loading on iOS. The newsletter also notes Claude Academy, free Anthropic courses and guides for using Claude.

Multiverse Computing researchers introduced Quantization-Aware Healing (QAH), a recipe that distills from the original pre-compression model rather than the recovered checkpoint. Applied to a GPT-OSS 120B compressed to 60B and quantized to MXFP4, the resulting model beat its bfloat16 source on 7 of 9 benchmarks, but an author acknowledged the headline table compares checkpoints with unequal training and lacks a control.

A sponsored article from Ollobot published by IEEE Spectrum describes the OlloNi SS1, an AI family companion robot, and outlines the company's "gentle intelligence" approach to companion robotics. It cites loneliness statistics and market projections while detailing the robot's sensors, mobility, on-device processing, and recording features.

Gradio introduced gr.Workflow, a tool that turns AI pipelines into a drag-and-drop graph of typed nodes where each step is runnable and every intermediate result is visible. Each workflow also serves as a REST API and can be deployed to Hugging Face Spaces with one command.

Ollama announced that Claude Desktop can now be configured to work with Ollama as a third-party gateway provider, allowing developers to run open models inside Claude. The integration supports both local and cloud models, with telemetry disabled by default and a Zero Data Retention policy.

MIT engineers developed Extreme Event Aware, or η-learning, a machine-learning method that generates plausible extreme events and worst-case scenarios without needing past extreme-event data. The work by Kai Chang and Themis Sapsis appeared Aug. 20 in Nature Communications.

Researchers at TU Delft developed a system that uses an LLM to translate natural-language passenger requests, such as "I am running late, go fast," into adjustments of a self-driving car's motion-planning parameters. Tested in the nuPlan simulator, it changed speed and smoothness in line with prompts while keeping the human in the loop for confirmation.

IEEE Spectrum reports on how AI is reshaping engineering work and what adaptability means in practice, drawing on interviews with Arizona State University's Samantha Brunhaver, Microsoft's Jenna Butler, and The Pragmatic Programmer coauthor Andy Hunt.

Anthropic has begun watermarking text generated by its Claude models, making it possible for the company to identify its own output, and Sebastian Raschka has published a 48-minute video lecture and slide deck explaining the mechanism, its detection, and how it can fail or be removed.