---
title: "Latest reads"
description: "The engineering internet, summarised so you can actually read it."
---

# Latest reads
> The engineering internet, summarised so you can actually read it.

## Articles

### [Over 30 new plugins join the Cursor Marketplace](https://yomu.fyi/post/over-30-new-plugins-join-the-cursor-marketplace.md)
- Company: [Cursor](https://yomu.fyi/company/cursor.md)
- Author: Kevin Niparko
- Published: Mar 11, 2026
- Type: Announcement

Cursor is adding more than 30 plugins to its Marketplace, including integrations from Atlassian, Datadog, GitLab, Glean, Hugging Face, monday.com, and PlanetScale. The plugins extend Cursor across infrastructure and productivity workflows: users can query observability data, manage GitLab repositories and pipelines, work with PlanetScale schemas and queries, handle Atlassian issues and backlogs, search company knowledge in Glean, and manage monday.com boards and tasks. Each plugin bundles capabilities such as MCPs with skills that instruct an agent how to use them, and those tools can support manually started or automatically invoked cloud agents. Cursor also supports community-built plugins, while Teams and Enterprise admins can create private team marketplaces. Automations can run agents on schedules or events with custom instructions.


### [Fin Product Updates: February recap](https://yomu.fyi/post/fin-product-updates-february-recap.md)
- Company: [Intercom](https://yomu.fyi/company/intercom.md)
- Author: Brian Donohue
- Published: Mar 10, 2026
- Type: Announcement

February updates for Fin address constraints in complex automation, voice quality, Shopify deployment, and helpdesk operations. Procedures can be drafted from existing source material, edited in natural language with hooks for live data, business logic, and deterministic code, then tested step by step in simulations using conversation data and defined criteria. Voice changes add pronunciation rules, 11 new voices, and automatic background-noise reduction, while Shopify setup connects a store, installs Messenger without code, and deploys Fin in under two minutes. Helpdesk additions include 11 call metrics and automatic public-holiday office hours; the post says 19 improvements shipped in 2026, and concludes that all featured updates are live and available now.


### [GitTrends: A Google Trends style view of the GitHub ecosystem](https://yomu.fyi/post/gittrends-a-google-trends-style-view-of-the-github-ecosystem.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Lionel Palacin
- Published: Mar 10, 2026
- Type: Announcement

GitHub produces billions of events across issues, pull requests, and comments, but turning this vast stream into a real-time trends analyzer presents significant technical challenges. GitTrends is an open-source demo application designed to search and compare technology keywords across nearly ten billion GitHub events. Built to showcase the full-text search capabilities recently released in ClickHouse, the application indexes raw text without pre-computed answers or intermediate data transformation. ClickHouse executes full-text searches and aggregations within a single engine in one pass, eliminating cross-system joins and data movement. The demo includes live comparisons toggling between full-text search, bloom filters, and full table scans to illustrate query performance at scale.


### [Announcing General Availability of ClickHouse Full-text Search](https://yomu.fyi/post/announcing-general-availability-of-clickhouse-full-text-search.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Melvyn Peignon
- Published: Mar 10, 2026
- Type: Announcement

ClickHouse has reached general availability for its native Full-text Search feature, designed to accelerate token-based text filtering across large datasets. The capability relies on deterministic inverted indexes rather than probabilistic Bloom filter skip indexes, mapping individual tokens directly to row numbers. It supports plain strings, string arrays, and map keys or values while allowing configurable pre-processing and tokenization using standard SQL expressions. The system does not implement relevance scoring or store positional data for ranking, focusing instead on rapid filtering paired with large-scale analytical aggregations. Testing shows significant reductions in scanned granules alongside substantial latency improvements, though text indexes require more storage than Bloom filters.


### [Sydney becomes Anthropic's fourth APAC office](https://yomu.fyi/post/sydney-becomes-anthropic-s-fourth-apac-office.md)
- Company: [Anthropic](https://yomu.fyi/company/anthropic.md)
- Published: Mar 10, 2026
- Type: Announcement

Anthropic plans to open an office in Sydney in the coming weeks, making it the company’s fourth Asia-Pacific office alongside Tokyo, Bengaluru, and Seoul. The expansion responds to strong demand in Australia and New Zealand and is intended to support their distinct AI ecosystems, with initial attention on enterprise, startup, and research customers. Anthropic says it will hire locally, deepen engagement with Australian institutions, pursue partnerships in national-interest and priority sectors, and explore additional compute capacity through third-party infrastructure already in place, partly to address data-residency requirements. Australia and New Zealand rank fourth and eighth globally in Claude.ai usage relative to population, while reported usage is especially strong for computer and coding tasks, educational instruction, and research.


### [Unfortunately, Sprites Now Speak MCP](https://yomu.fyi/post/unfortunately-sprites-now-speak-mcp.md)
- Company: [Fly.io](https://yomu.fyi/company/fly-io.md)
- Published: Mar 10, 2026
- Type: Announcement

Sprites are disposable cloud computers with durable filesystems, near-zero idle cost, and support for running coding agents and agent-managed workloads. The post announces an MCP interface, allowing Claude Desktop and other MCP-compatible agent tools to authenticate to a Fly.io organization and create or manage Sprites. It argues that MCP is aesthetically and operationally inferior to command-line tools and discoverable APIs because loading tool descriptions consumes context and can overemphasize capabilities an agent does not need. For agents that can run shell commands, the recommended pattern is a short skill description backed by progressive CLI or API discovery; MCP sessions remain a fallback for agents without command execution. Session guardrails can restrict the organization, cap Sprite creation, and apply name prefixes, while the broader design favors real filesystems and networks over stateless sandboxes.


### [Introducing Storage Buckets on the Hugging Face Hub](https://yomu.fyi/post/introducing-storage-buckets-on-the-hugging-face-hub.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Lucain Pouget, Eliott Coyac, Adrien Carreira, Victor Mustar, Julien Chaumond, Quentin Lhoest, Pierric Cistac, Sylvestre Bcht, Hugo Larcher, Rajat Arya, Di Xiao, Assaf Vayner
- Published: Mar 10, 2026
- Type: Announcement

Hugging Face introduces Storage Buckets as a mutable, S3-like object storage layer on the Hub for intermediate machine-learning artifacts such as checkpoints, optimizer states, processed shards, logs, and traces. Unlike versioned Models and Datasets repositories, Buckets support frequent writes, overwrites, directory synchronization, and cleanup while retaining standard Hub permissions and public or private visibility. Built on Xet, they chunk and deduplicate content across files, reducing repeated bytes, bandwidth, transfer time, and, for Enterprise customers, billed storage. The CLI and Python APIs support creation, syncing, dry runs, plans, inspection, individual file operations, and programmatic workflows; HfFileSystem and fsspec also allow compatible libraries to read and write through hf:// paths. Pre-warming can place hot data near AWS or GCP compute, while roadmap work aims to connect Buckets directly with versioned repos.


### [Keep the Tokens Flowing: Lessons from 16 Open-Source RL Libraries](https://yomu.fyi/post/keep-the-tokens-flowing-lessons-from-16-open-source-rl-libraries.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Amine Dirhoussi, Quentin Gallouédec, Kashif Rasul, Lewis Tunstall, Edward Beeching, Albert Villanova del Moral, Nouamane Tazi, Leandro von Werra, Sergio Paniego
- Published: Mar 10, 2026
- Type: Benchmark

At-scale reinforcement-learning training is constrained by long, variable data-generation workloads that leave training GPUs idle, especially for reasoning and agentic models. The surveyed architecture separates inference and training onto distinct GPU pools, connects them through a rollout buffer, and synchronizes weights asynchronously so both stages can proceed concurrently. Sixteen open-source libraries are compared across orchestration, buffering, weight synchronization, staleness, partial rollouts, LoRA, and distributed backends; Ray appears in 8 of 16 orchestration implementations, while NCCL broadcast is the default weight-transfer method. The proposed TRL async trainer favors a lightweight bounded queue with per-token model\_version metadata, packed NCCL transfers, and experimental support for prefix-resume or abort-and-retry handling of rollouts interrupted by weight updates.


### [How Critical Manufacturing uses ClickHouse to bring real-time intelligence to the factory floor](https://yomu.fyi/post/how-critical-manufacturing-uses-clickhouse-to-bring-real-time-intellig.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: ClickHouse
- Published: Mar 10, 2026
- Type: Problem & solution

Critical Manufacturing faced performance and scaling bottlenecks while processing millions of factory floor events using a legacy Microsoft SQL Server analytics stack. Increasing data volumes caused historical queries to take hours, straining real-time dashboards and escalating operational maintenance. The engineering team replaced this architecture by migrating analytics workloads to ClickHouse connected directly to streaming Kafka topics. They implemented an ELT pattern alongside ReplacingMergeTree engines to deduplicate late-arriving sensor records and utilized TTL policies for automated lifecycle management. The resulting platform delivers sub-second analytical queries across historical and real-time operational data, expands into OpenTelemetry observability storage, and maintains embeddings for retrieval-augmented generation workloads.


### [Financial Benchmarks](https://yomu.fyi/post/financial-benchmarks.md)
- Company: [Ramp](https://yomu.fyi/company/ramp.md)
- Author: Kedar Thakkar, Anton Biryukov, Ashwin Kumar, Ryne Carbone
- Published: Mar 9, 2026
- Type: Benchmark

Ramp describes how it benchmarks large language models (LLMs) used in financial products against day-to-day tasks such as invoice extraction, financial-statement OCR, policy review, accounting autocoding, compliance detection, and fund routing. These benchmarks combine task-specific metrics with cost, latency, reasoning effort, human decisions, historical context, and ground-truth datasets where available. For contextual invoice OCR, perfect extraction requires every field to match the user's final bill, while financial-statement OCR uses a 1% relative-error threshold against transcriptions from over 500 real P&L documents. Results show application-specific trade-offs: Gemini 3 Flash is described as a cost-efficient leader for several visual and financial tasks, while Claude models lead some high-accuracy or low-miss-rate settings, and behavior can differ within one provider. The framework emphasizes Pareto trade-offs and continuous testing, with plans to expand from benchmarking current performance to hill climbing on capabilities.


### [What Replo learned optimizing 100+ billion events in ClickHouse](https://yomu.fyi/post/what-replo-learned-optimizing-100-billion-events-in-clickhouse.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: ClickHouse
- Published: Mar 9, 2026
- Type: Problem & solution

Replo built an analytics pipeline on ClickHouse to serve more than 4,000 Shopify merchants processing 3,000 to 5,000 events per second. The initial single-table architecture recalculated session metrics on the fly, leading to poor query efficiency as volume scaled. To speed up dashboard queries, the team implemented a mark-and-unmark pattern with SummingMergeTree and refreshable materialized views, but real-time deduplication and fractional attribution caused runaway query backlog loops. Replo resolved the issue by limiting real-time recomputations to a specialized table holding only the last 40 minutes of purchase-related events. This architecture allowed processing over 100 billion records while keeping query times fast and ingestion lag to approximately one minute.


### [Ulysses Sequence Parallelism: Training with Million-Token Contexts](https://yomu.fyi/post/ulysses-sequence-parallelism-training-with-million-token-contexts.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Kashif Rasul, Stas Bekman
- Published: Mar 9, 2026
- Type: Explainer

Ulysses Sequence Parallelism addresses the memory and compute pressure of training language models on sequences extending beyond tens of thousands of tokens, where attention remains quadratic in sequence length even with FlashAttention. It shards tokens across GPUs, projects local queries, keys, and values, then uses two all-to-all operations to exchange data so each GPU computes a subset of attention heads before restoring sequence-sharded outputs. The post describes integrations with Accelerate, the Transformers Trainer, and TRL's SFTTrainer, including variable-length handling and weighted loss aggregation for uneven valid-token counts. In reported four-GPU tests, SP=4 reduced per-GPU memory sufficiently for 96K-token training on H100 80GB GPUs, reached 13,396 tokens per second at 64K, and matched non-SP canonical token-normalized loss within 0.000005 maximum absolute difference.


### [LeRobot v0.5.0: Scaling Every Dimension](https://yomu.fyi/post/lerobot-v0-5-0-scaling-every-dimension.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Steven Palma, Pepijn Kooijmans, Jade Choghari, Caroline Pascal, Khalil Meftah, Martino Russi, Nicolas Rabault, Michel Aractingi, Virgile BATTO, Thomas Wolf
- Published: Mar 9, 2026
- Type: Announcement

LeRobot v0.5.0 is a major release that expands supported hardware, policies, datasets, simulation, and its software foundation, following more than 200 merged PRs and more than 50 new contributors since v0.4.0. It adds full Unitree G1 humanoid support for locomotion, manipulation, teleoperation, and whole-body control, alongside OpenArm, Earth Rover, CAN-bus motor controllers, and other integrations. The policy additions include Pi0-FAST's autoregressive VLAs with FAST action tokenization, Real-Time Chunking for continuously blended predictions, Wall-X, X-VLA, SARM, and PEFT fine-tuning. For data and simulation workflows, streaming video encoding removes waits between recording episodes, while EnvHub loads environments from the Hugging Face Hub and IsaacLab-Arena supplies GPU-accelerated simulation. The codebase now requires Python 3.12, uses Transformers v5, supports installable third-party policy plugins, and adds remote Rerun visualization plus installation and documentation improvements.


### [Reclaiming Terabytes: Optimizing Android image caching with TLRU](https://yomu.fyi/post/reclaiming-terabytes-optimizing-android-image-caching-with-tlru.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Nguyen Van Minh
- Published: Mar 6, 2026
- Type: Problem & solution

Grab’s Android app cached runtime images with DiskLruCache, but its 100 MB size limit did not remove stale content while the cache remained below capacity, contributing substantially to device storage usage. The team developed Time-Aware Least Recently Used (TLRU), which combines LRU size eviction with time-based expiration using last\_accessed metadata and a TTL, while a minimum cache-size threshold protects a baseline of cached images. TLRU retains the 100 MB maximum and evicts entries when current\_time minus last\_accessed exceeds the TTL; the implementation also supports reading existing LRU journal files by ignoring timestamp suffixes, enabling safe rollback. Configuration was tuned through experiments using cache hit ratio, with a maximum acceptable decline of three percentage points. After production rollout, P95 total app size fell approximately 50 MB, while hit ratio stayed within target and infrastructure costs did not increase.


### [Designing flexible payment flows with Checkout Session](https://yomu.fyi/post/designing-flexible-payment-flows-with-checkout-session.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Allison Farris
- Published: Mar 6, 2026
- Type: Tutorial

The post addresses the architectural challenge of taking a digital product business global while supporting localized tax, currencies, one-time purchases, subscriptions, and a branded embedded checkout. It recommends Checkout Sessions as the orchestration layer, with Stripe creating and managing the underlying Payment Intent while declarative parameters handle features such as Stripe Tax, Adaptive Pricing, subscriptions, and promotional codes. The implementation creates a session with ui\_mode, mode, automatic\_tax, adaptive\_pricing, line items, and optional payment\_intent\_data, then passes the client secret to Stripe Elements for the Payment Element and Currency Selector. This approach supports localized payment methods and currency conversion without a backend rewrite, while direct Payment Intents remain appropriate when an application needs full control over carts, discounts, taxes, or its own checkout abstraction.


### [Working at the intersection of data and AI with Kenza Boulisfane](https://yomu.fyi/post/working-at-the-intersection-of-data-and-ai-with-kenza-boulisfane.md)
- Company: [Thumbtack](https://yomu.fyi/company/thumbtack.md)
- Author: Thumbtack People Team
- Published: Mar 5, 2026
- Type: Explainer

Thumbtack software engineer Kenza Boulisfane develops an artificial intelligence tool called the Marketing Analytics Agent to democratize access to complex cross-channel marketing data. Non-technical staff across Data Science, Growth, Brand, and Product can ask natural language performance queries without requiring SQL expertise or manual reporting. A primary technical challenge involves unifying fragmented data across multiple advertising platforms that define metrics differently while maintaining data reliability. The system interprets natural language queries, analyzes underlying performance metrics, and translates results into actionable, data-backed recommendations for next steps. Boulisfane also notes that virtual work environments enhance collaboration by encouraging clearer documentation, intentional communication, and focused individual productivity.


### [How Chronicle hit #1 on Product Hunt using Notion](https://yomu.fyi/post/how-chronicle-hit-1-on-product-hunt-using-notion.md)
- Company: [Notion](https://yomu.fyi/company/notion.md)
- Author: Mayuresh Patole
- Published: Mar 5, 2026
- Type: Explainer

Chronicle’s Product Hunt launch aimed for a few hundred upvotes but reached number one and generated a waitlist of more than 100,000 signups. The remote team used Notion as a launch “mission control,” expanding an initial planning page into a database with status and due-date properties, assigned owners, collaborator tags, asset links, social copy, and calendar views. The shared hub coordinated feature timelines, asset publication, emails, LinkedIn posts, announcements, pre-launch instructions, and real-time launch-day updates without losing a single source of truth. Templates turned tasks into reusable pages containing creative files, drafts, and strategy notes, while the broader workspace also stored customer interview transcripts and marketing work. The team attributes the result to early preparation, careful copy and assets, coordinated support from its ecosystem, and treating the launch as a company-wide event; it says the process enabled speed, alignment across time zones, and documented repeatability.


### [Transformation in action: When support becomes the blueprint for company-wide AI](https://yomu.fyi/post/transformation-in-action-when-support-becomes-the-blueprint-for-compan.md)
- Company: [Intercom](https://yomu.fyi/company/intercom.md)
- Author: Declan Ivory
- Published: Mar 5, 2026
- Type: Explainer

The post describes how businesses are extending AI beyond customer support across the customer lifecycle, with support teams increasingly positioned to lead that work. Research cited in the 2026 Customer Service Transformation Report says 52% of respondents plan to scale AI to other departments in 2026, while 32% say customer service teams lead their AI transformation strategy. WHOOP deployed Fin on its Join page after prospective buyers faced waits of more than 10 hours; Fin resolved 84% of inbound questions and helped drive a 130% increase in attributable sales. The post warns that isolated departmental agents can fragment experiences, and presents a unified Customer Agent with shared memory and business knowledge as the intended direction. It argues support can provide shared standards, guardrails, operating practices, and implementation expertise as AI expands into sales and other functions.


### [Bringing Robotics AI to Embedded Platforms: Dataset Recording, VLA Fine‑Tuning, and On‑Device Optimizations](https://yomu.fyi/post/bringing-robotics-ai-to-embedded-platforms-dataset-recording-vla-fine.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Gaetan Bahl, Enzo Ruedas, Tess Boivin
- Published: Mar 5, 2026
- Type: Tutorial

Deploying Vision–Language–Action policies on embedded robots is constrained by compute, memory, power, and real-time control requirements; synchronous inference can leave the arm idle and cause oscillatory corrections. NXP presents dataset-recording practices for a tea-bag placement task, including fixed cameras, controlled lighting, a gripper-mounted view, calibration backups, and diverse training and validation splits with recovery episodes. The example uses 120 episodes, three 640×480 cameras at 30 fps, batch size 8, and checkpoint selection based on training and validation success rather than loss alone. For deployment, the VLA is decomposed into vision, LLM backbone, and action-expert stages, with selective quantization and asynchronous scheduling constrained by inference time being shorter than action execution. On the i.MX 95, optimized ACT latency reached 0.32 seconds with 0.89 global accuracy, while ONNX FP32 ACT reached 0.96 global accuracy at 2.86 seconds.


### [Build agents that run automatically](https://yomu.fyi/post/build-agents-that-run-automatically.md)
- Company: [Cursor](https://yomu.fyi/company/cursor.md)
- Author: Jack Pertschuk, Jon Kaplan & Josh Ma
- Published: Mar 5, 2026
- Type: Announcement

Cursor Automations introduces always-on agents that run on schedules or react to events from Slack, Linear, GitHub, PagerDuty, or custom webhooks. Each invocation spins up a cloud sandbox, uses configured models and MCPs, verifies its output, and can draw on memory from past runs. The product supports review and monitoring workflows, including security audits on pushes to main, risk-based codeowner assignment, and PagerDuty-driven incident investigation that proposes fixes. Other automations generate weekly repository digests, add tests after merged code, triage Slack bug reports through Linear, and assemble cross-tool dashboards. Examples from Cursor and Rippling report faster incident response and reduced repetitive work, while Runlayer describes a software factory built with Runlayer MCP and plugins.


[Newer posts](https://yomu.fyi/posts/page/153.md) · [Older posts](https://yomu.fyi/posts/page/155.md)