Loading…
Latest reads
Every post, summarised
Grounded summaries of engineering blogs, newest first. Each links to the original.
- 4,671 posts
- 20 companies
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Latest posts
Preparing for the Customer Agent future
At Pioneer 2025, Eoghan McCabe presented a plan for Fin to become a unified Customer Agent spanning the entire customer experience, from first contact through purchase, support, success, and upsell. The proposal responds to the risk that separate AI Agents for sales, marketing, success, and service will use conflicting priorities, disconnected information, and inconsistent interactions. Fin is intended to evolve from a task-based system into an agentic system with fluid roles, user-defined goals, persistent customer memory, comprehensive business knowledge, and interoperability across tools, systems, and channels. The post argues that customer service leaders are positioned to guide this wider AI rollout because they have experience, data, and operational insight from early AI adoption. Early activity includes WHOOP using Fin for sales conversations, where it is resolving 84%, while new capabilities are expected to roll out over the coming months.
Cloud Agents
Cursor’s Cloud Agents let developers run multiple agents remotely without keeping a laptop connected to the internet. Agents can be launched and managed from the Cursor editor, cursor.com/agents, Slack, Linear, and GitHub, supporting background bug fixes, small todos, and larger feature work. For difficult bugs, the team commonly has multiple models attempt the same problem and selects the best result; for complex features, a locally developed plan can be sent to a cloud agent for implementation, allowing work to continue on another task. Cursor also revamped its GPT-5 Codex agent harness for longer time horizons in the cloud, while plan mode supports sending implementation plans remotely. The workflow is available today in Cursor editor and Cursor Web, as well as the listed integrations.
Aligning to What? Rethinking Agent Generalization in MiniMax M2
MiniMax M2 addresses the performance gap between agent benchmark scores and real-world deployment across diverse scaffolding frameworks and environments. While initial alignment efforts relied on scaling tool variety, models continued to fail when subjected to slight variations in prompts, scaffolding, and tool responses. To resolve this instability, the team introduced interleaved thinking, which enables models to execute internal reasoning at arbitrary steps throughout long-horizon tasks to adapt dynamically to external tool outputs. Additionally, developers constructed a data pipeline designed for full-trajectory generalization by introducing perturbations across system prompts, user objectives, environments, and tool feedback. In internal evaluations, MiniMax M2 maintained robust tool-calling and instruction-following performance when tested against unfamiliar, cold-start agent scaffolding frameworks.
Machine-learning predictive autoscaling for Flink
Grab’s Flink platform saw application count grow 2.5 times, making cost-effective, self-service CPU provisioning difficult for users who often relied on trial and error and over-provisioned resources. Reactive autoscaling with Adaptive Scheduler, Kubernetes HPA, and metrics such as CPU usage or consumer latency caused restart spikes, scaling spirals, fluctuations, and additional constraints when connectors such as Kafka reached fixed partition parallelism. Predictive Resource Advisor instead uses vertical scaling and predictive, deterministic CPU decisions, automatically collecting data, training models, and applying configuration changes. The source says applications using it experienced fewer autoscaling-related disruptions, adapted to workload growth and surges, and averaged more than 35% cloud infrastructure cost savings. It also reports simpler default-configuration deployment, while planned work includes memory tuning, improved burst handling, and more efficient model training.
Improving performance by prefetching product pages from Etsy Search
Etsy’s search and web performance teams tested the Speculation Rules API (SRA) to reduce loading time for product listing pages reached from desktop organic search. They added a speculationrules configuration that prefetches eligible listing HTML after a 200-millisecond hover, using URL and data-attribute selectors with moderate eagerness; unlike link prefetch, SRA uses both memory and HTTP caches. The experiment required attention to a two-page memory limit, five-minute retention, non-cacheable listing HTML, video hover behavior, cookies, new-tab targets, and analytics events triggered during prefetches. Results showed 23.6% faster 75th-percentile TTFB, with improvements of 20.7% in FCP, 21.1% in LCP, 20.4% in DOMContentLoaded, and 10.6% in Page Load, while roughly one in 14 prefetches led to navigation.
First Block with Fabian Hedin, Co-founder and CTO of Lovable
First Block features Fabian Hedin, co-founder and CTO of Lovable, an AI-powered software development platform that lets non-technical people build production-ready applications through natural language. The conversation covers Lovable’s launch in November 2024 and its reported growth from $1M to $100M ARR in eight months, alongside Hedin’s path from running Minecraft servers to founding the company. Hedin argues that teams should build around what foundational models can do now instead of designing elaborate agent systems for anticipated capabilities. He also says Lovable launched later than it should have because of its waitlist and uses its own product extensively for internal tools, including office CRM and operations software. The episode frames early user feedback and obsessive dogfooding as practical ways to improve an AI product.
On the Shifting Global Compute Landscape
United States export controls on advanced artificial intelligence hardware have catalyzed rapid expansion across China's domestic silicon and open-weight model ecosystem. Faced with restricted access to high-end NVIDIA GPUs, Chinese developers accelerated deployments on domestic accelerators, including Huawei Ascend, Cambricon, and Baidu Kunlun. Hardware scarcity spurred architectural and algorithmic innovations in compute efficiency, notably DeepSeek's Multi-head Latent Attention and Group Relative Policy Optimization, alongside substantial post-training cost reductions. Organizations such as Baidu and Ant Group now train foundation models directly on domestic hardware, fostering non-CUDA software stacks and lowering inference costs globally. Consequently, the global artificial intelligence infrastructure is shifting from an exclusively American-focused paradigm toward a dual-ecosystem landscape powered by domestic chips and open-weight architectures.
Building our next R&D hub in Berlin
Intercom will open its next R&D hub in Berlin on January 19, 2026, to support investment in its AI customer service platform and Fin, its AI Agent. It has already welcomed its first Berlin hires and plans to hire 100 people over the following year across engineering, AI, data science, product, and design. Berlin was chosen for its technical talent, creative culture, AI-focused startups and research labs, and proximity to the existing R&D hubs in Dublin and London. Fin is described as the primary contributor to this growth, with more than 7,000 businesses having transformed customer service with it and Fin months away from passing $100 million in ARR. The new office at Zimmerstraße 78 will host an R&D recruitment event on February 26, 2026, while an AI builders day is planned for February.
Composer: Building a fast frontier model with RL
Composer is a mixture-of-experts agent model for software engineering that targets interactive use, with reported frontier coding results and generation speed four times faster than similar models on the authors’ benchmarks. It is trained with reinforcement learning on real-world software engineering challenges in large codebases, using production search and editing tools, terminal commands, and codebase-wide semantic search. Training responses may be code edits, plans, or informative answers, while the process rewards efficient tool use, parallelism, minimal unnecessary responses, and evidence-based claims. Evaluation uses Cursor Bench, built from real agent requests and curated optimal solutions, measuring correctness alongside adherence to existing abstractions and software engineering practices. The training stack combines asynchronous reinforcement learning with PyTorch and Ray, MXFP8 MoE kernels, expert parallelism, hybrid sharded data parallelism, and sandboxed cloud environments scaled to hundreds of thousands concurrently.
Introducing Cursor 2.0 and Composer
Cursor 2.0 introduces Composer, described as the product’s first coding model, alongside an interface designed for running multiple agents in parallel. Composer is built for low-latency agentic coding, completes most turns in under 30 seconds, and was trained with tools including codebase-wide semantic search to improve work in large codebases. The new interface centers on agents and outcomes rather than files, while still allowing users to open files in a redesigned layout or return to the classic IDE; parallel runs are isolated through git worktrees or remote machines. The release also targets review and testing bottlenecks by simplifying change inspection and adding a native browser tool that lets Cursor test its work and iterate, and Cursor 2.0 is now available.
Building a Healthcare Robot from Simulation to Deployment with NVIDIA Isaac
Simulation in healthcare robotics has historically been slow, siloed, and difficult to translate into real-world physical systems. The NVIDIA Isaac for Healthcare v0.4 release introduces an end-to-end SO-ARM starter workflow to streamline simulation, training, and deployment for embodied surgical assistants. The architecture relies on a Sim2Real mixed training approach that fine-tunes the GR00T N1.5 foundation model on combined real-world and synthetic datasets using LeRobot. Synthetic data generated in simulation constitutes over 93 percent of the policy training data to bridge physical data collection constraints. Real-time inference on the SO-ARM101 precision manipulator utilizes RTI DDS communication, dual-camera vision, and TensorRT model conversion.
How to Build a Healthcare Robot from Simulation to Deployment with NVIDIA Isaac for Healthcare
Simulation in healthcare robotics has historically been slow, siloed, and difficult to translate to physical hardware. NVIDIA introduced the Isaac for Healthcare v0.4 developer framework to streamline this lifecycle through an end-to-end SO-ARM starter workflow. Developers collect mixed real-world and synthetic datasets using LeRobot and SO-101 teleoperation, with synthetic generation accounting for over 93 percent of training data. The pipeline post-trains GR00T N1.5 models using dual-camera vision inputs, evaluates policies in Isaac Lab, and deploys them to physical hardware with RTI DDS communication. This integrated approach allows surgical assistant robots to interpret natural language instructions and execute physical manipulation tasks reliably.
Granite 4.0 Nano: Just how small can you go?
IBM has released Granite 4.0 Nano, a collection of sub-billion and ~1.5B parameter language models tailored for edge and on-device deployments. Trained on more than 15 trillion tokens under an Apache 2.0 license, the suite includes four instruct models alongside their base counterparts. The lineup introduces hybrid-SSM architectures—specifically the ~1.5B parameter Granite 4.0 H 1B and the ~350M parameter Granite 4.0 H 350M—as well as standard transformer variants suited for runtimes like llama.cpp. According to benchmark results across general knowledge, math, code, and safety domains, these compact models demonstrated strong capability improvements relative to their parameter footprint. Furthermore, evaluation on IFEval and Berkeley's Function Calling Leaderboard v3 indicated competitive performance on instruction following and tool calling tasks.
Voice Cloning with Consent
Realistic voice generation presents significant deepfake risks alongside legitimate communication benefits. To prevent unauthorized voice duplication, a voice consent gate integrates verification directly into the voice cloning workflow. The system relies on a language model to generate novel sentence pairs containing explicit consent and phonetically diverse phrases on random everyday topics. Users must read these generated lines aloud through a direct microphone stream rather than uploading pre-recorded files, and an automatic speech recognition system verifies the utterance against the generated text. Once confirmed, the consenting audio snippet is passed directly to the voice-cloning text-to-speech model as the reference sample for synthesis.
Streaming datasets: 100x More Efficient
Training machine learning models on multi-terabyte datasets often requires long download times or suffers from network request storms when multiple DataLoader workers initialize independently. Hugging Face revamped the backend of its datasets library to make dataset streaming viable for large-scale distributed training without modifying the standard API. Startup latency was reduced by caching resolved data file lists across workers and bundling API calls, cutting initial request volume up to one hundred times. The team also implemented background prefetching for Parquet files and added configurable buffering parameters to keep GPUs fully saturated during training runs. Benchmark tests demonstrated up to twice the sample processing throughput and zero worker crashes across 256 concurrent workers on 64 H100 GPUs.
huggingface_hub v1.0: Five Years of Building the Foundation of Open Machine Learning
The Hugging Face Hub team released huggingface_hub v1.0, marking the library's transition into a mature foundation for open machine learning after five years of development. The major update migrates the underlying network backend to httpx and replaces the legacy hf_transfer tool with the hf_xet binary package to enable chunk-level file deduplication across repositories. In addition, the release introduces a redesigned Typer-based CLI using the streamlined hf command alongside explicit authentication functions and expanded inference primitives. Legacy components, including the Git-based Repository class and the older InferenceApi interface, have been removed to establish a cleaner and more maintainable architecture. Most dependent machine learning libraries remain compatible across versions, though transformers requires v0.x for version 4 and v1.x for its upcoming version 5 release.
Compare 8 Instagram Analytics Tools for 2026
The guide compares eight Instagram analytics tools for 2026, arguing that native Instagram Insights is useful for a quick overview but offers limited history, exporting, and competitor comparison. It frames third-party platforms as a way to consolidate metrics such as reach, engagement, follower growth, Story performance, demographics, and hashtag results while reducing manual reporting. The listed tools serve different needs: Brandwatch and Sprout Social emphasize cross-platform management and reporting, Iconosquare offers 100+ Instagram metrics, Pallyy focuses on creators and posting-time recommendations, and Sendible targets agencies managing multiple profiles. Selection should follow goals and reporting requirements, with engagement rate, reach, follower growth, click-throughs, and conversions identified as core measures; competitor and influencer tracking is available in many tools.
LeRobot v0.4.0: Supercharging OSS Robot Learning
LeRobot v0.4.0 expands the open-source robot-learning stack with scalable datasets, simulation benchmarks, processing tools, policies, and hardware integrations. LeRobotDataset v3.0 introduces chunked episodes, video streaming, unified Parquet metadata, improved loading and memory efficiency, plus utilities for editing and converting datasets. A modular Processor system separates policy and robot pipelines, while direct Accelerate integration simplifies multi-GPU training; the release also adds LIBERO, Meta-World, PI0, PI0.5, and GR00T N1.5 support. A plugin architecture enables third-party robots, cameras, and teleoperators to be installed without changing the core library, with Reachy 2 and phone teleoperation included, alongside a self-paced Robot Learning Course and hands-on tutorial.
Platform Thinking: Three Views from Two Sigma Leaders
Two Sigma structures its quantitative investment operations around foundational platforms that balance operational speed with scientific rigor. In data engineering, foundational teams provide both raw and curated datasets using BigQuery, CI/CD-managed transformation pipelines, and reusable data contracts termed ice cubes to serve ninety percent of use cases. Quantitative modeling incorporates open-source large language models trained strictly on point-in-time data to rapidly generate features while preventing temporal leakage. Long-term platform innovation follows an epsilon and omega strategy that couples overarching vision with iterative, learning-focused steps. Platform leadership categorizes artificial intelligence applications into advisory, oracle, operational, and agentic functions designed to enhance human productivity rather than replace researchers.
Using a web-based POS with a mobile Terminal reader
Web-based POS systems can reuse ecommerce infrastructure for showroom-floor selling, but mobile readers such as the Stripe M2 and Tap to Pay cannot be accessed directly by a mobile browser. The proposed wrapper app avoids a complete native POS rebuild by embedding the site in a React Native WebView and connecting a mobile or Tap to Pay reader through the Stripe Terminal React Native SDK. A bridge message carries cart information, such as a cart ID and ready-to-pay status, to a ReaderManager that retrieves the amount and calls createPaymentIntent(), collectPaymentMethod(), and confirmPaymentIntent(). Payment results can return to the POS through a backend webhook and WebSocket or frontend polling, while security requires origin-checked messages, nonces, and server-side validation. The approach retains WebView performance, offline, native-access, debugging, and security limitations.
Top topics
- Architecture834
- Performance816
- LLMs660
- Machine Learning622
- Open Source563
- AI Agents553
- AI470
- Developer Experience448
Weekly digest
The best engineering reads, once a week.