Loading…
Latest reads
Every post, summarised
Grounded summaries of engineering blogs, newest first. Each links to the original.
- 4,668 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
Behind the build: 2025's BFCM globe in a pinball machine 🌎🕹️
The 2025 BFCM live globe was implemented inside a browser-based pinball machine to represent real-time merchant sales and celebrate milestones. Built with three.js and react-three-fiber, the experience runs at 120fps, uses Server-Sent Events for order arcs and live metrics, includes a 128x32 shader-driven dot-matrix display, and supports AR through react-three/xr. For performance, the team simulated pinball physics in 2D with Rapier2D and used raycasts for ramps and elevation, reducing the physics step from 2ms to 0.2ms. The scene also includes interactive mini games, infrastructure statistics, Easter eggs, and milestone-triggered reactions, while Apache FlinkSQL pipelines process event streams behind the commerce visualizations. The globe was also displayed on the Exosphere in Las Vegas for the third consecutive year.
2026 customer service planning series: Vol. 03
AI performance in customer support can drift without clear ownership, even when the underlying model is capable. This part of a 2026 planning series presents four roles: AI operations lead, knowledge manager, conversation designer, and support automation specialist. The roles cover monitoring and guardrails, structured and compliant knowledge, conversational tone and flow, and reliable action through backend systems, with the AI operations lead coordinating reporting and improvement. The post describes an operating loop in which performance patterns drive knowledge fixes, interaction changes, and expanded automation, while phased ownership moves from part-time assignments to formal roles and hiring. It says dedicated specialization becomes essential when AI handles 50–70% of incoming volume, and frames these roles as the backbone of an AI-first support organization.
How Ramp built an AI operating system for scalable work
Ramp spent the last year building what operations and AI product leader Ben Levik calls an “AI operating system” for the company, with the goal of becoming more productive. Rather than waiting for tools to mature, Ramp promotes a “Builder” mindset and organizes adoption around Prompt, Knowledge, and Workflow. Employees refine vague requests by having AI ask clarifying questions, while a Notion knowledge base connected to other tools uses feedback loops in which AI drafts corrections and humans approve them. AI agents and connectors then automate tasks such as summarizing project updates from five sources and researching prospects for review-ready sales emails. About 90% of 1,200 employees use Notion AI monthly, and Ramp reports shipping 270 features in the first half of 2025, more than during all of 2024, shifting the bottleneck toward choosing what to automate next.
Streamlining Security Investigations with Agents
Slack's Security Engineering team handles billions of daily security events and needed a reliable way to streamline on-call alert triage. An initial prototype relying on a single 300-word prompt produced inconsistent results and frequently reached spurious conclusions without properly challenging assumptions. To gain precise control, the team decomposed the workflow into chained model invocations with structured JSON outputs organized across three agent personas: a Director, four domain experts, and a Critic. Domain experts gather raw evidence through tool calls, the Critic evaluates finding quality and synthesizes a timeline, and the Director steers investigation phases using tiered model costs. The multi-agent system enables engineers to supervise investigations via a real-time dashboard while uncovering emergent issues like credential exposures across process ancestry chains.
A Decade of Defense: Celebrating Grab's 10th Year Bug Bounty Program
Grab marks the tenth anniversary of its bug bounty program, launched with HackerOne and initially involving 23 researchers. Over the decade, it grew into a global community, partnering with more than 850 active HackerOne researchers to help protect over 187 million users across Southeast Asia. The program broadened its scope in 2023–2024, adding assets that included Indonesian financial services and AI systems, while introducing a mobile-specific bounty table and experimenting with campaigns and reward methods. Grab says it now ranks among HackerOne’s top 20 programs worldwide and top three in Asia, and credits live hacking events and regional club campaigns with expanding testing coverage and researcher relationships. Anniversary activities included invite-only campaigns in Germany, Morocco, and India, plus a broader campaign that received 461 submissions, with xchopath receiving the Best Hacker Bonus.
Transformers v5: Simple model definitions powering the AI ecosystem
Transformers v5.0.0rc-0 introduces major architectural updates focused on simplicity, training, inference, and ecosystem interoperability across modern AI workflows. The release adopts a modular modeling approach and centralizes attention implementations into a unified AttentionInterface abstraction to reduce contribution and code review overhead. Support for Flax and TensorFlow is officially sunset in favor of focusing on PyTorch as the primary backend, while tokenization is standardized around the tokenizers library. For execution workloads, v5 adds native continuous batching, paged attention mechanisms, and a dedicated transformers serve OpenAI-compatible serving system. Finally, weight loading is refactored to make low-precision quantization a first-class citizen alongside broad interoperability with formats such as GGUF, MLX, and TorchAO.
Social Media Posting Schedule: 9-Step Guide
Managing multiple social accounts makes consistency, timing, content mix, and collaboration difficult, so the guide presents a nine-step scheduling process. It recommends auditing active and dormant channels, defining goals and KPIs, studying audience behavior and platform differences, selecting a tool, and building a calendar around content pillars, campaigns, roles, and a two-to-four-week planning horizon. The process then covers batch creation, reusable assets, platform-specific captions, approval workflows, analytics, time zones, and flexibility for timely content; Brandwatch Publish is presented as an example supporting TikTok, Instagram, Facebook, X, LinkedIn, and Pinterest. It concludes that scheduling should create capacity for engagement and strategy without sacrificing authenticity, and advises regular audits, experimentation, and readiness to pause or adapt queued posts. The guidance also warns against excessive promotion, ignoring engagement, identical cross-platform posts, weak visuals, and over-reliance on automation.
More context, more confidence: The new CX Score explained
CX Score was created to give support teams a broader view of customer experience than survey-based measures such as CSAT and NPS, which capture only a subset of interactions. Its latest iteration evaluates more of each conversation, combining answer quality for Fin and human teammates with customer effort, strong emotion, product or service feedback, and policy feedback. The update also expands scoring to shorter and more transactional threads, adds reasons behind individual ratings, and enriches summaries with the issue, handling, and drivers of the score. The stated goal is to help teams identify operational problems, route conversations for review or follow-up, and move from measurement toward continuous improvement. The new version is rolling out gradually to all customers and is scheduled to be in workspaces by December 3rd.
Real-time data quality monitoring: Kafka stream contracts with syntactic and semantic test
Kafka stream data can fail syntactically when actual records do not match expected schemas, or semantically when fields violate agreed patterns, ranges, or constants. The described platform addresses this gap with user-defined data contracts covering schema agreements, field-level semantic rules, stream configuration, and ownership metadata. An LLM recommends semantic rules from schemas and anonymized samples, while a transformation engine converts them into Test Runner configurations and inverse SQL queries. A long-running FlinkSQL Test Runner reads Kafka with its own consumer group, publishes issue events and stores them in AWS S3; Genchi sends Slack alerts and Coban shows violating fields and sample records. Deployed across 100+ critical topics, the solution provides real-time detection, alerting, and visibility, with future work targeting producer tracing, cross-field tests, completeness, freshness, and DTO change detection.
Reducing experiment duration with predicted control variates
Etsy describes how its experimentation platform evolved from CUPED to CUPAC to shorten the time needed to reach statistically powered results. CUPED uses pre-experiment visitor data as covariates in linear regression, while CUPAC first uses a nonlinear model to predict the outcome and then uses that prediction in the variance-reduction correction. Etsy trained and tuned LightGBM models on more than 100 pre-experiment features in Vertex AI, generating predictions through parallel Dataflow jobs and applying the correction with Spark within an Airflow-orchestrated pipeline. CUPED reduced variance by 7% on average, whereas CUPAC achieved 27%, shortening average experiment duration by almost three additional days, although results ranged from 2% to 77% depending on the metric and experiment; Etsy is balancing CUPAC’s impact against its training and maintenance costs.
2026 customer service planning series: Vol. 02
When AI Agents resolve most customer conversations, support organizations need to move beyond tiered models centered on queues, handoffs, and volume-based productivity. The proposed AI-first operating model distributes four foundational responsibilities: an AI operations lead monitors and improves performance, a knowledge manager maintains structured content, a conversation designer shapes communication and handoffs, and a support automation specialist builds executable workflows. Existing enablement, QA, workforce management, and leadership roles also evolve, with player-coach leaders managing both AI performance and people. Examples from Dotdigital, Clay, Lightspeed, and Intercom illustrate dedicated roles and an Intercom structure organized around Human Support, AI Support, and Support Operations and Optimization.
200+ Social Media Holidays 2026: Monthly Calendar
The calendar assembles more than 200 social media holidays, events, and suggested hashtags for 2026, giving content teams prompts to use throughout the year. It organizes occasions month by month, from New Year’s Day, Science Fiction Day, and Data Privacy Day in January to Halloween, Cyber Monday, and New Year’s Eve later in the year. Entries span awareness observances, cultural and religious dates, seasonal milestones, and lighter themes such as National Pizza Day, World Emoji Day, and National Cookie Day. The list is intended to add variety to social media strategies, and the post points readers to separate monthly roundups with examples of brand posts and additional ideas. Its conclusion presents the calendar as a source of content inspiration for increasing marketing activity in 2026.
Unwrap Innovation: How Sellers and Buyers Use AI to Brighten the Holidays
eBay describes how its AI tools aim to reduce holiday-season friction for sellers and improve discovery and trust for buyers. More than 10M unique sellers have used its AI features, while sellers create more than half a million new listings daily using Generative AI. The simplified listing experience can start with an image, background enhancement adds festive professional-grade backgrounds, and AI Assistant for messaging drafts replies to common questions about item specifics or shipping. For shoppers, an AI shopping assistant pilot, LLM-powered fashion recommendations, and access to eBay’s inventory through third-party agents including OpenAI partnerships support personalized discovery. Trust features include Authenticity Guarantee expert verification for handbags, jewelry, and trading cards, plus Certilogo’s digital IDs and authentication for apparel, while eBay says these measures support safer participation in the circular economy.
Run for your money: Engineering a treadmill that prints store credit
For the 2025 NYC marathon weekend, Shopify and Endorphins Running built a gamified treadmill that awarded in-store credit for maintaining a self-selected pace. Participants had 30 seconds to reach their target, then earned credit for each second spent within a defined pace zone, with a personalized barcode receipt redeemable through a Shopify POS UI extension. The implementation used a curved, motorless TrueForm Trainer, FTMS messages, a Python relay, Socket.IO, a local Remix application, receipt printing, and a production-hosted POS integration. Interval timestamps were normalized because treadmill messages arrived irregularly, while a state machine coordinated warm-up, gameplay, operator controls, database updates, discounts, and printing. The resulting experience attracted new customers, with 63% of participants identified as new customers, and its patterns could support other sensor-driven in-store incentives.
Stablecoin payments for Stripe developers: zero crypto knowledge required
The guide explains how Stripe developers can add USDC stablecoin payments without managing private keys, gas estimation, network selection, or wallet connection flows. Stripe supports USDC on Ethereum, Solana, Polygon, and Base, settles transactions as fiat in the merchant’s Stripe balance, and charges a 1.5% fee; the described feature is currently limited to US businesses. Integration can use payment_method_types: ['card', 'crypto'], although Stripe recommends automatic_payment_methods for Dashboard-managed configuration, filtering, ordering, and testing. Stripe Checkout and Elements expose crypto without frontend changes, while confirmation, webhook handling, monitoring, and familiar failure processing remain aligned with existing payment flows. The guide also describes Sandbox testing without testnets or fake cryptocurrency and presents crypto payments as a way to offer global settlement and potentially lower processing costs.
Diffusers welcomes FLUX-2
Diffusers introduces support for FLUX-2 models through the Flux2Pipeline and Flux2Transformer2DModel classes. Running the 4-bit quantized checkpoint diffusers/FLUX.2-dev-bnb-4bit requires loading Mistral3ForConditionalGeneration as the text encoder and the transformer in bfloat16 precision with CPU offloading enabled. During image generation with a prompt and 50 inference steps, the pipeline encodes text embeddings through the Mistral 3 model before executing diffusion operations. However, executing the pipeline on a GPU with 14.56 GiB capacity triggers a CUDA OutOfMemoryError during 4-bit dequantization inside bitsandbytes matrix multiplication operations. Additionally, the execution logs warn that Flax classes are deprecated in Diffusers and will be removed in version 1.0.0.
Continuous batching from first principles
Large language model serving requires running expensive next-token generation across multiple concurrent user requests. Traditional batching approaches introduce severe padding inefficiencies when mixing variable-length prompts and different generation phases, especially under static shape constraints like CUDA graphs. Continuous batching resolves these inefficiencies by combining key-value caching, chunked prefill, ragged batching, and dynamic request scheduling. Ragged batching eliminates the traditional batch axis by concatenating token sequences into a single tensor and using boolean attention masks to isolate independent sequences. By dynamically removing completed prompts and packing decoding tokens alongside chunked prefill tokens up to a hardware memory budget, serving systems maintain high hardware utilization and throughput.
Building Deep Research: How we Achieved State of the Art
Building production AI research agents presents challenges around context window pollution, escalating token costs, and architectural brittleness across model updates. To resolve these issues, Tavily rebuilt its deep research system around simplified orchestration, compact tooling, and active context curation. The agent mimics human research workflows by distilling tool outputs into concise reflections for ongoing reasoning, withholding raw retrieved web data until the final deliverable stage. Compared to traditional ReAct propagation architectures where token consumption scales quadratically, this reflection-based approach achieves linear token growth. As a result, the system reduced token consumption by 66% compared to Open Deep Research while reaching state-of-the-art performance on DeepResearch Bench.
OVHcloud on Hugging Face Inference Providers 🔥
OVHcloud is now integrated as a supported Inference Provider on the Hugging Face Hub, expanding serverless inference options across model pages and client SDKs. The integration enables access to open-weight models, including gpt-oss, Qwen3, DeepSeek R1, and Llama, using European infrastructure with pay-per-token pricing starting at €0.04 per million tokens. Developers can connect via Python and JavaScript SDKs using either direct custom API keys or automatic routing through Hugging Face tokens. The service supports structured outputs, function calling, multimodal workflows, and embedding models while providing sub-200ms first-token response times. Calls routed through Hugging Face pass through standard provider pricing without markups, and PRO tier subscribers receive monthly inference credits.
Unlocking Peak Performance on Qualcomm NPU with LiteRT
Google introduces the LiteRT Qualcomm AI Engine Direct (QNN) Accelerator, developed with Qualcomm to replace the previous TFLite QNN delegate and simplify NPU deployment across supported Qualcomm SoCs. It integrates vendor compilers and runtimes behind a unified workflow, supports ahead-of-time and on-device compilation, and can fall back to GPU or CPU when needed. Across 72 canonical vision, audio, and NLP models, the accelerator supports 90 LiteRT ops, enables full NPU delegation for 64, and reports up to 100x CPU and 10x GPU speedups. On Snapdragon 8 Elite Gen 5, FastVLM-0.5B uses int8 weights, int16 activations, and specialized transformer kernels to reach 0.12 second time-to-first-token, over 11,000 prefill tokens per second, and over 100 decode tokens per second.
Top topics
- Architecture834
- Performance815
- LLMs659
- Machine Learning621
- Open Source563
- AI Agents553
- AI469
- Developer Experience448
Weekly digest
The best engineering reads, once a week.