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
Expanding our long-running agents research preview
Cursor has expanded its long-running agents research preview to all Ultra, Teams, and Enterprise users. The preview targets autonomous work on ambitious projects, using a custom harness designed to address frontier models’ failures on long-horizon tasks. Long-running agents propose a plan for approval before execution and use multiple agents to check work and maintain follow-through. Reported examples include runtimes of 36, 30, and 25 hours, while initial results showed substantially larger PRs with merge rates comparable to other agents. Internal examples include a Rust migration and custom kernels for video rendering, a 10,000-line network-policy PR, and secure sudo prompting for Cursor CLI; the company says agents can now deliver working solutions after hours or days, though safer production deployment and collaboration across agents remain active areas.
OpenEnv in Practice: Evaluating Tool-Using Agents in Real-World Environments
OpenEnv is an open-source framework from Meta and Hugging Face for evaluating AI agents against real systems rather than simulations, with stateful environments that connect to tools and APIs. Turing's production-grade Calendar Gym applies this model to calendar operations, combining access control lists, partial visibility, temporal reasoning, and multi-step workflows in isolated sessions. Agents often handled individual actions but became less reliable on longer or ambiguous tasks: success was close to 90% with explicit calendar identifiers and roughly 40% when requests used natural-language descriptions. More than half of failures involved malformed arguments or incorrect action ordering despite selecting the correct tool, while structured validation, permission, and datetime errors provided concrete recovery signals.
Increased usage for agents
Cursor has raised limits for Auto and Composer 1.5 for all individual plans. The update introduces separate first-party models and API usage pools: first-party usage is increased for Auto and Composer 1.5, while API pricing and plan allowances remain unchanged. Composer 1.5 has a standard limit of three times Composer 1’s, with a temporary increase to six times through February 16. Cursor says this supports developers moving from autocomplete toward daily agentic coding while accommodating different preferences for speed, intelligence, and affordability. Composer 1.5 is described as scoring above Sonnet 4.5 on Terminal-Bench 2.0 but below the best frontier models, and both pools reset with the monthly usage cycle; the limits are live for Pro, Pro Plus, and Ultra plans.
Make Secure App-to-App Connections Using Cross App Access
Enterprise app-to-app integrations can create governance, security, and consent problems when API keys lack user context and standard OAuth leaves connection approval to individuals. The tutorial adds Cross App Access (XAA), an OAuth extension based on the Identity Assertion Authorization Grant, to a NestJS note-taking app and connects it to a to-do app through xaa.dev. It uses TypeScript, openid-client, and the testing site’s IdP, with two token exchanges: an ID token becomes an ID-JAG token, then a JWT-bearer grant exchanges that assertion at the resource authorization server for an access token. After authentication, the completed flow calls the to-do API and exposes each handshake step and response in console logs.
Introducing Composer 1.5
Composer 1.5 is the successor to the first agentic coding model, Composer 1, and is positioned for fast, interactive daily use. It was built by scaling reinforcement learning 20x further on the same pretrained model, with post-training compute exceeding the base model’s pretraining compute. In an internal benchmark of real-world coding problems, performance quickly surpasses Composer 1 and continues rising, with the largest gains on challenging tasks. Composer 1.5 generates thinking tokens to reason about codebases and plan next steps, using minimal thinking on easy problems and continuing until it finds a satisfying answer on hard ones. It also recursively self-summarizes when context runs out, a training behavior intended to preserve original accuracy as context length varies; the release is described as significantly stronger and suitable for interactive use.
Transformers.js v4: Now Available on NPM!
Transformers.js v4 is now available on NPM, marking a major release after a year of development. It introduces a C++-rewritten WebGPU runtime tested across about 200 supported model architectures, enabling the same JavaScript code to run WebGPU-accelerated models in browsers, Node, Bun, Deno, and desktop applications. Specialized ONNX Runtime Contrib Operators support new export strategies for large language models, and the post reports a roughly fourfold speedup for BERT-based embedding models using MultiHeadAttention. The release also restructures the repository as a pnpm-workspaces monorepo, modularizes model definitions, moves examples to a separate repository, and switches from Webpack to esbuild, reducing build times from two seconds to 200 milliseconds and shrinking transformers.web.js by 53%. New APIs cover model asset inspection, caching, environment fetch controls, logging, and standalone tokenization, while added architectures include Mamba, MLA, and MoE models compatible with WebGPU.
Minions: Stripe’s one-shot, end-to-end coding agents
Stripe’s Minions are homegrown, fully unattended coding agents designed to complete tasks in a single run, beginning with an invocation such as a Slack thread and ending with a pull request ready for human review. More than 1,000 pull requests merged weekly at Stripe are minion-produced and contain no human-written code, although engineers review them and can provide follow-up instructions. Each run uses an isolated, pre-warmed devbox and a customized fork of Block’s coding agent goose, combining agent loops with deterministic operations for version control, linting, and testing. Minions use curated MCP tools, including Stripe’s Toolshed with more than 400 tools, to retrieve internal context before and during work. Local checks and selective CI provide feedback, with no more than two CI rounds, balancing correction against token, compute, and time costs.
NVIDIA commits 3x more code across 30,000 developers with Cursor
NVIDIA has deployed Cursor across its software development lifecycle, with more than 30,000 developers using it daily and committing over three times as much code as before. NVIDIA’s long-lived, interdependent codebases and varied technology stacks made it difficult to understand downstream effects and keep pace with rapid change. Cursor maps and semantically reasons over large codebases, retrieving relevant context, while custom rules and MCP servers extend it from code generation into reviews, testing, debugging, issue-driven fixes, CI troubleshooting, and git-flow automation. The company also uses it to help new hires learn unfamiliar systems and experienced engineers work across programming languages and stack areas. NVIDIA reports that bug rates have remained flat despite higher coding velocity, while code-style consistency has improved.
Engineering at scale, with purpose.
Senior software engineer Brett Shouse outlines site reliability engineering initiatives and infrastructure modernization efforts underway at Thumbtack. The engineering organization is currently leading an operational project to migrate multiple self-hosted observability services onto a single, unified SaaS platform. This architectural transition consolidates application logs, distributed traces, and system metrics into one accessible view tailored for engineers, customer support staff, and company executives. Moving away from legacy self-hosted monitoring systems reduces systems administration overhead, mitigates alert fatigue, and lowers direct operational infrastructure costs. Concurrently, the reliability team tackles technical debt accumulated from rapid organizational growth by establishing structured incident response processes and automating repetitive operational toil.
Global Consumer Insights Ahead of 2026 Ramadan
Ramadan reshapes daily routines, online activity, spending, dining, and hospitality, making it a major consumer and cultural moment across Muslim regions. Drawing on nearly 7 million English- and Arabic-language online conversations from 2025, the analysis tracks shopping, food, home aesthetics, hospitality, and regional posting patterns. Ramadan mentions rose 76% in the three months before the 2026 holiday year over year, while fragrance mentions increased 33%, Iftar-party mentions approached 900,000, recipes rose 144%, and Iftar or Suhoor packages and ticketed events rose 145% in 2025. Joy accounted for 57% of emotion-categorized Ramadan mentions, and conversations centered on togetherness, spirituality, traditions, and generosity. Afternoon activity peaks between 2 pm and 4 pm across the analyzed regions, but brands are advised to use their own regional and platform analytics and meet audiences through authentic participation rather than promotion.
Introducing SyGra Studio
SyGra 2.0.0 introduces Studio, an interactive environment for building, previewing, executing, and debugging synthetic data generation flows on a visual canvas. Users can configure connectors and models, preview source rows, reference automatically created state variables in prompts, define structured outputs, and inspect the generated YAML or JSON graph configuration. Studio streams node-level execution status, token usage, latency, cost, logs, and guardrail outcomes, while storing run history in .executions/runs/*.json and autosaving drafts. It also runs existing workflows, including a Glaive Code Assistant example with conditional revision loops, tunable dataset and batching inputs, and observable intermediate critiques; generated outputs can support training, evaluation, or annotation workflows.
Towards self-driving codebases
The post describes research into scaling long-running autonomous coding agents, using a web browser as a complex benchmark and running thousands of agents continuously for one week. An initial single-agent plan failed because the model lost track, stopped prematurely, and stalled on complex details; shared-state coordination then introduced locking contention, while more structured planner, executor, worker, and judge roles improved ownership and throughput. The team built a Rust-based harness on one large Linux VM, logged agent messages and tool activity for replay and analysis, and iterated through designs that eventually favored a more dynamic system with freshness mechanisms and fewer overloaded roles. The research concludes that architecture, prompts, observability, explicit throughput trade-offs, and anti-fragile recovery matter, reporting linear scaling of token throughput with minimal harness overhead while humans retained taste, judgment, and direction.
Defining customer support in 2025: Why it's key to your long
Customer support has shifted from a perceived cost to a differentiator as expectations rise and companies focus on retention, advocacy, and product feedback. Modern support aims to resolve issues, troubleshoot problems, and help customers get more value through personal, contextual, knowledgeable, and proactive interactions. The post argues that human+AI partnerships can improve speed and efficiency: AI agent Fin handles common questions in more than 45 languages, while human teammates address complex queries empathetically. It also explains why ticketing became necessary for tracking asynchronous work, internal ownership, and customer updates within an all-in-one support platform. Examples from Hospitable and Rebag report faster responses, reduced response times, and higher NPS, while the broader guidance emphasizes clear expectations, robust processes, escalation paths, and a clear route to human help.
Community Evals: Because we're done trusting black-box leaderboards over the community
Hugging Face is introducing Community Evals, a system for decentralized and transparent reporting of model evaluation results on the Hub. It responds to saturated benchmarks and inconsistent scores across model cards, papers, and evaluation platforms, while acknowledging that benchmark performance does not necessarily reflect real-world capabilities. Benchmark dataset repositories can register as benchmarks, define reproducible eval specifications in eval.yaml, aggregate scores from model repositories and community submissions, and display leaderboards in dataset cards. Model results are stored in .eval_results/*.yaml, while users can submit sourced scores through pull requests that remain visible as community results, with Git history and APIs supporting tracking and further aggregation. The beta feature aims to expose what was evaluated, how, when, and by whom, but it does not claim to solve benchmark saturation, test-set training, or the benchmark-reality gap.
Edge AI Made Easy: MongoDB and ObjectBox Data Synchronization
Modern edge AI deployments require immediate real-time decision-making, local privacy, and offline operation while retaining the analytical scale of central cloud infrastructure. To address these hybrid demands, MongoDB and ObjectBox partnered to release the ObjectBox MongoDB Sync Connector as part of ObjectBox 5.0. ObjectBox operates as a lightweight on-device database with local vector search, multi-language bindings, and high-throughput I/O suited for resource-constrained hardware. In environments such as industrial factory floors and restaurant point-of-sale terminals, devices capture and evaluate operational data onsite without depending on active network connections. The native connector subsequently synchronizes critical operational datasets to MongoDB Atlas for long-term storage, fleet-wide monitoring, and machine learning model retraining.
The Future of the Global Open-Source AI Ecosystem: From DeepSeek to AI+
This third and final article in a series examines how Chinese AI organizations are shaping an open-source ecosystem after DeepSeek R1's January 2025 release. It argues that open sharing of models, papers, deployment infrastructure, and techniques has become the dominant near-term approach, with organizations ranging from Alibaba, Tencent, ByteDance, and Baidu to startups and application companies contributing at different layers. Qwen illustrates the ecosystem strategy: by mid-2025, more than 113,000 Hugging Face models used Qwen as a base, while organizations also connected models with chips, cloud platforms, toolchains, evaluation, and deployment systems. The article places this shift on infrastructure built since 2017, including the East Data, West Compute plan, and says R1 catalyzed wider use of existing compute, energy, and data capacity. It concludes that open-source models are becoming reusable, composable components in industrial workflows and localized AI systems as China moves from DeepSeek toward AI+.
Training Design for Text-to-Image Models: Lessons from Ablations
The second part of a series on training efficient text-to-image models from scratch documents training-design ablations for the PRX model. The authors establish a Flow Matching baseline (PRX-1.2B, 100K steps, 256×256, AdamW, GemmaT5, RoPE, 1M MidJourneyV6 synthetic images) and benchmark interventions against FID, CMMD, DINO-MMD, and throughput across four buckets: Representation Alignment, Training Objectives, Token Routing and Sparsification, and Data. REPA representation alignment boosts early convergence when used as a burn-in, while the Muon optimizer lowered FID from 18.20 to 15.55 and improved CMMD and DINO-MMD. Storing denoiser weights in BF16 instead of using BF16 autocast raised FID to 21.87, and token routing methods such as TREAD/SPRINT were minor at 256² but became major wins at higher resolutions. Long captions proved critical, synthetic versus real data shifted texture versus structure, and upcoming work releases the full training code alongside a 24-hour end-to-end speedrun.
Building AI Search on Heroku
The reference architecture addresses a common RAG failure: vector search can return semantically similar documentation that does not answer the user's question. Its query pipeline retrieves the top 20 chunks with cosine similarity in Heroku Postgres and pgvector, then sends those candidates to Cohere Rerank 3.5, a cross-encoder reranker, before Claude 3.5 Sonnet generates an answer. The indexing pipeline crawls URLs, filters likely boilerplate, splits documents into 1,000-character chunks with 200-character overlap at natural boundaries, embeds them with Cohere Embed Multilingual, and batch-inserts them. Server-Sent Events stream embedding, search, reranking, sources, and generated text to avoid a blank screen during the multi-step request. The reranking stage can be made optional when lower latency matters more than precision.
Cursor at Grab: Adoption and impact
Grab introduced Cursor in late 2024 as one of several AI coding assistants and reports rapid adoption across engineering and non-technical teams. Monthly active usage reached 98% of Tech Grabbers, weekly usage about 75%, and Cursor suggestions were accepted around 50% of the time, compared with a cited industry average of 30%. About half of users engage on at least 10 consecutive days monthly, while more than a third of merge requests incorporate Cursor; common uses include test generation, refactoring, code navigation, bug fixing, and routine automation. Grab built monorepo indexing and distributed preconfigured rules for environment-specific suggestions, while a fixed-effects regression effort is examining productivity; early findings report a statistically robust dose-response relationship, with gains increasing alongside usage intensity.
Litestream Writable VFS
Litestream’s VFS is being extended to support Sprites, whose storage stack needs SQLite-backed block maps to serve writes immediately during cold starts, before a full database or block map is available. The VFS indexes each database page by file, offset, and size in LTX files, serves reads from object storage, and now optionally buffers writes locally before syncing them roughly every second or on clean shutdown. Write mode assumes a single writer, disables polling, and offers only the same eventual durability already provided by Sprite storage. A separate hydration mode serves remote reads while reconstructing a complete database file in the background, then switches reads to that file; the temporary file is discarded on exit. These features target Sprite-like workloads, while ordinary Litestream sidecar deployments do not require them.
Top topics
- Architecture834
- Performance816
- LLMs660
- Machine Learning622
- Open Source563
- AI Agents553
- AI470
- Developer Experience448
Weekly digest
The best engineering reads, once a week.