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
From 2D Drawings to 3D Models: Designing an AI Reconstruction Pipeline on MongoDB
BIRD is a research framework that converts five 2D perspective drawings into a proportionate, dimensioned 3D building envelope. It uses a dual-branch CNN to separate architectural line segments from noise, a text recognition model to scale them to real-world inches, and a wrapping algorithm to fold the elevation views around the floor plan. The framework achieved a median accuracy of roughly 93% across complete building plan sets from small residences to a multi-story apartment complex. The proposed deployment on MongoDB Atlas stores each project as a single JSON document containing all drawings’ segments, corrections, and final geometry, with S3 for image storage and Atlas Stream Processing for event-driven workflow coordination. MongoDB Search and Vector Search with Voyage AI embeddings enable full-text and semantic search over past projects, allowing drafters to start from corrected models of similar builds.
Canary rollouts: upgrade models in production without downtime
Deploying a new model checkpoint in production typically requires a tradeoff: a hard swap exposes all users to regressions at once, while a DIY staged cutover relies on manual traffic shifting. Together AI’s rollout system automates this by moving traffic between source and target deployments through configurable canary steps, such as 5%, 25%, 50%, and 100%. Each step runs health checks before shifting traffic, enforces a propagation wait for routing caches, and drains the source only after traffic has moved. Optional metric gates, like p95 latency, run after each step; if a gate trips, the rollout pauses automatically and waits for human action. In a real canary from Qwen2.5-7B to Qwen3.5-9B, a 137% p95 regression was caught at 10% traffic, allowing reversal with no failed requests.
The 2026 CDP Shift: Why AI Agents Are Only as Smart as Your Data Infrastructure
The 2026 Customer Data Platform Report identifies a critical enterprise challenge: while AI tools grow individually smarter, fragmented customer data causes a worsening, not better, experience. The shift is from computational capacity to contextual continuity, as disconnected pipelines force AI agents to start every interaction from scratch. Three failures arise from fragmented systems: agent amnesia, escalating operational costs from wasted tokens, and trapped intelligence that never reaches active workflows. Modern CDPs have evolved from passive audience builders into API-first context engines that support three pillars: pre-execution context with working memory, unified behavioral and conversational data streams, and infrastructure-level governance enforcing compliance before data reaches AI models. Organizations like Twilio, Trustpilot, and Quartz demonstrate that embedding real-time customer context into the conversation layer outperforms deploying the smartest standalone models.
How to design an AI-to-human handoff that preserves context
AI-to-human handoffs often fail because context is lost during transfer, forcing customers to repeat themselves. 76% of consumers report that human agents have little to no context about their issue, and 91% of brands have seen customer satisfaction drop due to AI missteps, with 40% attributed to context loss between channels. A cold transfer drops the customer into a new queue with nothing carried over, while a warm transfer briefs the receiving agent beforehand. Five steps to preserve context include defining escalation triggers early, specifying a payload with conversation history and customer profile, generating a three-line summary, maintaining context across channel changes, and writing back learned information to customer records. Measuring repeat-explanation rate and post-handoff resolution rate helps identify whether handoffs are working or if the AI is over-holding conversations.
How Stripe is designing Checkout for AI agents
Stripe implemented WebMCP on its Checkout UIs, which processes around 0.45% of global GDP, to optimize the buying experience for AI agents. Previously, agents relied on slow browser automation, consuming an average of 1.8 million tokens and 38.83 tool calls per transaction, taking over 2.5 minutes. WebMCP exposes MCP tools from the client side, allowing agents to execute actions like get_order_summary or fill_payment_form instead of interpreting the DOM. The implementation uses progressive tool disclosure, revealing only relevant tools; in evaluations across six models, agents using WebMCP consumed 42% fewer tokens, made 38% fewer tool calls, and completed checkout 39% faster. Stripe reused existing UI state and forms via imperative and declarative APIs, avoiding a separate agent-only code path and keeping human and agent experiences aligned.
How UK AISI and EvalEval Are Making Benchmark Results Reproducible
The UK AI Security Institute (AISI) has begun using EvalEval’s Every Eval Ever (EEE) schema and Evaluation Cards platform to openly share LLM evaluation results. This collaboration, stemming from a joint workshop at NeurIPS 2025, aims to make benchmark results more reproducible by providing standardized configuration and context alongside verified scores. AISI released results for five benchmarks—HealthBench, FrontierMath, Humanity’s Last Exam, SWE-Bench Pro, and Terminal-Bench 2.0—across six frontier models, plus two cyber evaluations using a partially overlapping model set. The release accompanies a paper studying how inference-time compute and evaluation protocol affect performance, revealing that models continued solving tasks when given correctness feedback. By adopting a shared reporting schema, evaluators can compare findings across studies and understand how setup choices influence reported performance, while EvalEval invites model developers, evaluation creators, and researchers to contribute to the shared infrastructure.
Confused Deputy: The Old Bug That AI Agents Keep Reintroducing
The Confused Deputy problem, first documented in 1988, describes a program with legitimate authority being tricked into misusing that authority on behalf of a less-privileged actor. AI agents are particularly vulnerable because they process instructions and untrusted content through the same natural-language channel, making prompt injection straightforward. The post traces this pattern through CSRF, SSRF, and recent attacks like Clinejection, where an issue-triage bot executed an attacker-supplied command from a crafted issue title. That attack led to a poisoned GitHub Actions cache and a compromised npm package downloaded 4,000 times. The fix involves task-scoped short-lived credentials, capability-scoped permissions, credential isolation from the model, and human-in-the-loop approval for high-risk actions.
Data Sovereignty by Design: Five Practical Evaluation Lenses for IT Leaders
Data sovereignty is a market expression, not a universal legal standard, and IT leaders should evaluate it through architecture and governance questions rather than a single label. The assessment covers where data resides, who can access it, how it moves, and whether control is retained, with considerations varying by jurisdiction, industry, and risk posture. Physical location is only one part of a broader control boundary that includes operations, key custody, and access paths. The article proposes five technical lenses for evaluation: jurisdiction-native data placement, in-use encryption, customer-held keys, network isolation and identity-based access, and forensic-grade audit evidence. These lenses are practical reference points for architecture decisions, not a legal test or certification checklist, and organizations should assess their own requirements with qualified counsel.
The Headless Dashboard Initiative
Resend announced a headless dashboard initiative to make all dashboard capabilities available programmatically via the MCP, CLI, and API. Recent releases include the Email Metrics API for deliverability and engagement data, a Headless Webhook API for listing events and replaying deliveries, and a Cancel Broadcast API. The Duplicate Broadcast API copies any broadcast into a draft, and the Share Email API provides expiring read-only email links. The team aims to make the dashboard best-in-class while committing to full headless operation for every future entity. Developers can start building today by connecting the MCP server or using the CLI.
Transformers now runs llama.cpp quants
Transformers now supports running GGUF quantized models locally, reusing llama.cpp’s ggml kernels for efficient inference on Apple Silicon. The GGUF format packages weights, tokenizer, and chat template in one file, with quantization levels like Q4_K_M that trade precision for memory. Users can load a checkpoint from the Hub with from_pretrained and generate text using the standard Transformers API, or serve it via an OpenAI-compatible endpoint. The initial focus is on the Qwen3.5 architecture, with benchmarks on a MacBook Pro M2 Max showing token-generation rates close to llama.cpp, including prefill time. The integration uses packed quantization kernels and generation-loop optimizations to reduce overhead, though it is currently limited to single conversations on MPS without batching support.
Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community
Jun Kim, creator and maintainer of oMLX, has joined Hugging Face to support the MLX community. MLX is Apple’s framework for local AI, optimized for Apple Silicon, and Hugging Face is the Hub for MLX models. oMLX, staying Apache 2.0, will gain stability and faster development as a funded, maintained project, allowing Jun to guide contributors and build for the long term. The goal is to unblock the community to run local AI in any form, with oMLX serving as a testbed leveraging dependencies like mlx-lm and mlx-vlm, and collaborating with projects such as LMStudio. A key focus is streamlining the transition from transformers model definitions to reference MLX implementations for different engines, so each can focus on unique features.
Simplifying Model Serving Across Multiple GPUs with NVIDIA TensorRT Multi-Device Integration in NVIDIA Dynamo-Triton
NVIDIA TensorRT multi-device inference enables a single TensorRT network to execute across multiple GPUs using NCCL-backed distributed collectives, supported from TensorRT 11.0. Integrated into Dynamo-Triton release 26.07, one model instance can own multiple GPUs and expose a single gRPC endpoint so the client does not coordinate GPU ranks. The post demonstrates this with Cosmos 3 Nano video generation, where the transformer uses Ulysses context parallelism to distribute 44,160 video tokens across up to eight GPUs. End-to-end latency drops from 156.6 seconds on one GPU to 34.2 seconds on eight GPUs, achieving a 6.09x transformer RPC speedup. Teams can trade additional GPU resources for shorter request latency while keeping the application interface and surrounding workflow stable.
How to Evaluate AI Agents From Tool Calls to Task Completion
Standard LLM benchmarks are insufficient for evaluating AI agents because they score static outputs rather than multi-step tool-call sequences against a live environment. The post explains that agent evaluation requires two scoring layers: step-level (process scoring) checks each call’s validity and usefulness, while end-to-end (E2E) scoring checks only the final environment state. A trace logs every step and environment state, and metrics roll up through a hierarchy of benchmark, trial, task, turn, and step. Key metrics include task success rate, consistency, tool-call precision, argument accuracy, steps per success, and cost per success. The post argues that benchmarks are converging on tool use because deployed agents always use tools, and enterprise benchmarks should measure task completion against real APIs and policies.
Amazon launches research initiative with Stanford University to advance AI and science
Amazon and Stanford University have launched the Stanford and Amazon Research Initiative, a formalized framework to advance research in AI, energy, and healthcare. The initiative builds on a deep relationship where currently more than 10 teams across Amazon fund active research and PhD fellowships at Stanford, spanning humanoid robotics, post-quantum cryptography, causal measurement science, and AI-driven radiology. It aims to accelerate the path from breakthrough research to real-world solutions that benefit society at large and broaden participation from diverse scholars. Focus areas will leverage strengths in artificial intelligence, machine learning, automated reasoning, and health, supported by Amazon’s cloud computing and AI services. The collaboration includes joint research projects, PhD fellowships, and interdisciplinary symposia, and was announced at a Stanford event with AWS CEO Matt Garman and Vice Provost David Studdert discussing the impact of university-industry collaborations.
Advancing AI for biology: Teaching models to design and characterize antibodies
Monoclonal antibodies are a mainstay of biopharmaceutical development, with over 100 FDA-approved drugs and established manufacturing and regulatory pathways, yet conventional discovery is costly and slow, typically taking six to twelve months from target to lead candidate. AI promises to make antibody development cheaper, faster, and more flexible by designing and characterizing therapeutic antibodies computationally, but scientific questions remain. Development hinges on three factors: the best binding site on the target, which candidates bind most tightly, and whether candidates can survive manufacturing and the clinic. Predictive models for each factor do well on familiar targets and assays but considerably worse on unfamiliar ones, a gap that in-distribution benchmarks have made difficult to measure. Three papers from Amazon Bio Discovery’s science team address research questions about each factor, with methods integrated into the AI-powered application that connects biological AI models with lab services.
Open-Sourcing Rebalancer: A Generic, High-Performance Library for Solving Assignment Problems
Meta has open-sourced Rebalancer, a generic library for solving assignment problems that has been used internally for over nine years. Rebalancer separates problem specification from solution, using a three-step approach with modeling constructs like dimensions, partitions, scopes, and utilization, and a high-level spec API for common objectives and constraints. Problems are translated into an expression graph, and Rebalancer offers two solving techniques: an optimal solver using mixed integer programming (MIP) with commercial or open-source solvers, and a local search solver that evaluates millions of moves per second for large-scale problems. The local search solver handles problems with over a million objects and 5k bins, with an average solve time of 171 seconds. Rebalancer is used for roughly 40 million assignment problems daily across Meta’s infrastructure, and the open-source release includes a debugging UI called Rebalancer Explorer. The accompanying paper was published at OSDI’24.
Postgres on NVMe: performance and the convergence of transactions and analytics
The post examines how local NVMe storage changes Postgres performance, particularly when working sets exceed memory and disk latency becomes the bottleneck. It argues that while faster storage accelerates transactional workloads, it does not alter the row-store layout, so analytical queries still require different approaches. Benchmark results show 9.2× more throughput on NVMe, with a median of 16,030 TPS versus 1,734 on EBS. The post introduces WalShadow, an open-source replication engine that reads the physical Postgres WAL to enable sub-second replication to ClickHouse without logical slots. It concludes that failover is solvable with tools like Patroni, and that storage latency is only part of the performance problem.
Turn Your Latest Observations Into Timely Weather Decisions With NVIDIA Earth-2
Weather-sensitive industries can now turn real-time observations into timely decisions using the AI data assimilation tools in NVIDIA Earth-2. The tutorial covers two techniques: Score-Based Data Assimilation (SDA) for constraining diffusion models with point observations, and HealDA for assimilating disparate datasets into a consistent global state. SDA works within downscaling models like CorrDiff and forecasting models like StormCast, guiding denoising steps toward observations without retraining. A concrete example showed SDA reducing wind-speed RMSE by 54% at held-out stations when applied to CorrDiff downscaling over Europe. Earth2Studio provides a unified interface across gridded, conventional, and satellite data sources, making it easy to connect proprietary observations to weather pipelines.
Introducing Our New Dropbox API Documentation
Dropbox has launched a modernized API documentation site at https://docs.dropboxapi.com, rebuilt to make integration faster and more intuitive. The new documentation features interactive API testing that lets developers set parameters, issue API calls, and view real responses from the Dropbox API directly in their browser without switching tools. An embedded AI assistant offers instant help with endpoint questions, troubleshooting, and code examples, while the documentation integrates with MCP (Model Context Protocol) so AI tools like Claude Code and Cursor can access up-to-date API information. Through the MCP server at https://docs.dropboxapi.com/_mcp/server, developers can connect their AI assistants to answer questions, explore the API, and assist with integrations using information directly from the docs. The update delivers a clean, responsive, searchable interface that brings endpoints, types, schemas, and related documentation together, with improved navigation, and provides comprehensive information including detailed error responses, type definitions, and full request/response schemas.
Pruning LLMs Like a Physicist: Block Removal as an Ising Optimization Problem
Removing entire transformer blocks is a cheap way to speed up large language models, yielding predictable inference gains and stacking with quantization and other compression methods. The challenge is that block choices interact, making it a combinatorial problem rather than a simple ranking task. The authors reformulate block selection as a constrained binary optimization (CBO) problem, mapping it onto an Ising glass, a disordered spin system with all-to-all interactions and a fixed number of removed blocks. They note that mean-field methods, which treat each block independently, are inadequate for this interacting system. Block removal composes with quantization, low-rank compression, width pruning, and knowledge-distillation-based healing, fitting into a larger pipeline rather than competing with it. The paper is titled ‘LLM Compression by Block Removal with Constrained Binary Optimization.’
Top topics
- Architecture834
- Performance815
- LLMs659
- Machine Learning621
- Open Source563
- AI Agents553
- AI469
- Developer Experience448
Weekly digest
The best engineering reads, once a week.