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
Thinking of ACE? We Can Do It with Fewer Tokens
ALTK-Evolve and ACE are agentic-memory systems that turn an LLM agent’s past trajectories into reusable lessons without weight updates or human labels. Both preserve individual lessons and attach evidence counts rather than compressing experience into a short summary, but ACE injects its full evolving playbook at every step while ALTK-Evolve consolidates typed, provenance-linked guidelines and retrieves a model-appropriate subset or the full set. On AppWorld’s 168-task test_normal split with the same ReAct setup, ALTK-Evolve reached 89.3 TGC and 80.4 SGC on DeepSeek-V3.2 at 263K tokens per task, versus ACE’s 80.4 and 73.2 at 634K. With gpt-oss-120b, it reached 56.0 TGC and 37.5 SGC at 116K tokens, narrowly exceeding ACE’s 54.8 and 35.7 at 777K. The comparison uses single pass@1 runs and in-house ACE results on matched models, while the authors attribute the cost difference to delivery rather than lesson quality.
NVIDIA Nemotron 3.5 Lightning Delivers Fast, Accurate Specialized Task Execution for Long-Running Agents
NVIDIA Nemotron 3.5 Lightning is an open 30B mixture-of-experts model with 3B active parameters, designed for the high-volume execution layer of long-running autonomous agents. The post positions it alongside frontier reasoning models: larger systems handle planning and orchestration, while Lightning handles tool calls, result validation, subagent delegation, and other routine work. Its design combines harness-optimized training, multi-token prediction for speculative decoding, DFlash and DSpark draft models, and NVFP4 quantization alongside BF16. The model is customizable through LoRA or full SFT, reinforcement learning, and environment-based evaluations, with weights, training data, and recipes released under OpenMDW-1.1. The post reports up to 4x the output speed of similar-sized models, 86% accuracy on PinchBench, and completion of 10,000 tasks 30% faster than Qwen3.6 35B at similar accuracy, while NeMo Switchyard routes work among models.
Route AI Agent Workloads Across Models with NVIDIA NeMo Switchyard
NVIDIA NeMo Switchyard addresses the problem of routing AI-agent work across models with different capabilities, costs, latency, and verbosity profiles rather than sending every request to one model. At runtime, its router evaluates a request, context, policies, and system signals, then directs work to a configured model pool, potentially routing individual agent steps and carrying session state when needed. The provider-agnostic switchyard-libsy SDK separates semantic model targets from provider endpoints, while a reference server accepts OpenAI, Anthropic, and Responses API requests and records routing decisions and outcomes. Available approaches include tuning-free LLM classifier, stage, and escalation routers, alongside tunable and custom algorithms. In a LangChain evaluation, escalation routing cut cost 74% versus a frontier-only baseline with a measured six-point accuracy trade-off, while Cognition’s staged routing reported near-frontier coding performance at approximately 28% lower mean cost.
Cloudflare DDoS Threat Report H1 2026: 1 Tbps attacks soar as DNS floods and geopolitical tensions drive a new wave
In the first half of 2026, network-layer DDoS attacks exceeding 1 Tbps surged by 519% quarter-over-quarter, with 935 mitigated attacks recorded across the period. Threat actors shifted their focus toward reflection and amplification methods, making DNS-based floods and CLDAP floods the dominant vectors. Despite the rise in hyper-volumetric traffic, 96.62% of network-layer attacks remained under 500 Mbps and 90.60% ended in under 10 minutes. Geopolitical conflicts and international events heavily influenced target selection, pushing the Media, Production & Publishing industry to the top position. Because brief multi-gigabit bursts trigger downstream routing instability and application timeouts before analysts can respond, automated, always-on edge mitigation has become critical.
Modern Risk Demands a Real-Time Foundation: The CRO’s Mandate
The piece argues that modern risk failures stem less from modeling weakness than from slow, fragmented data architecture. It uses Silicon Valley Bank, Archegos, and the UK LDI crisis to show how batch processing, siloed counterparty data, and static stress tests can obscure liquidity, aggregate exposure, and feedback loops until intervention is difficult. Its proposed shift is a unified Databricks Data and AI Platform combining governed data, real-time aggregation, lineage, scenario analysis, and AI-assisted investigation across risk functions. Specific capabilities described include Unity Catalog permissions and lineage, VaR or Expected Shortfall deltas computed in seconds, and SQL-auditable natural-language queries with MLflow-backed model provenance. The conclusion is that real-time, defensible risk visibility supports the CRO’s expanding role in capital allocation and growth, with Raiffeisen Bank International, Morgan Stanley, and State Street cited as production examples.
What's new in Sentry Logs: The summer 2026 roundup
Sentry introduced a collection of updates to its logging platform aimed at improving search scalability, export flexibility, and general viewer usability. The system now supports URL-synced log pinning alongside an overhauled export interface that handles aggregate data, full attribute payloads, and high-volume JSONL formats across Logs, Errors, and Tracing. Performance enhancements to the log viewer include a more constrained layout with reduced initial data loads, inline error representations, individual chips for multi-value search terms, and trace-level filtering. To assist large-scale queries spanning hundreds of gigabytes or terabytes, frontend and backend modifications added longer query timeouts with continuations, clearer rate-limiting messages, and in-place retries.
Electric joins Databricks to bring WASM Postgres to AI agent sandboxes
Databricks says Electric is joining the company to extend its Postgres capabilities from the lakehouse to AI agent sandboxes. The post frames agentic applications as requiring both durable governed state and rapidly changing local context: agents choose data at runtime, operate in sandboxed environments where cloud databases are reached over the network, and collaborate while working in parallel. Electric’s PGlite is a lightweight Postgres database built for WebAssembly that runs inside an agent, application, browser tab, or device, while its real-time sync engine synchronizes distributed data with centralized Lakebase infrastructure. PGlite’s weekly downloads reportedly grew from 1 million to 13 million in twelve months, and the sync architecture is described as keeping agents’ shared context current. Combined, Lakebase, PGlite, and synchronization are presented as a Postgres-based foundation for collaborative agentic applications at production scale.
How Standardizing Product Telemetry Reduced Time to Insight by 97%
Salesforce faced scaling bottlenecks when individual engineering teams maintained fragmented, custom telemetry pipelines requiring manual data transformations and dashboard creation every release. To resolve these silos, the engineering organization established the Product Data Platform (PDP) using a standardized custom schema built atop Monitoring Cloud infrastructure. This framework mandates core telemetry fields for consistent metric generation while offering governed optional attributes for product-specific flexibility. An AI-driven Model Context Protocol tool assists developers by recommending compliant instrumentation patterns directly within their workflows. Today, the unified platform processes 45 billion rows of data daily across 19,000 distinct events and 2,000 product features, reducing dashboard refresh latency by 97 percent from one month to daily updates.
Mastering Edge AI on Raspberry Pi with LiteRT and Gemma
Running edge AI on a Raspberry Pi 5 is presented as a way to build autonomous, low-latency systems that operate offline, with Reachy Mini serving as the demonstration. The setup combines LiteRT, LiteRT-LM, Gemma models, and LiteRT’s WebGPU (Vulkan) backend via ML Drift, splitting workloads between the CPU and VideoCore VII GPU. On CPU, Gemma 4 E2B reportedly reaches 99 tokens/sec prefill, 9 tokens/sec decode, and 1,432 MB peak memory; its tokenizer supports about 27.3 characters/sec in the voice demo. The pipeline runs YOLO object detection on the GPU, Moonshine speech recognition, Gemma reasoning, and text-to-speech on the CPU, while LiteRT CLI covers conversion, quantization, benchmarking, inference, and model execution from the LiteRT Hugging Face Community.
Why Go is an Ideal Language for AI-Assisted Software Engineering
AI-assisted software development shifts the human bottleneck from writing code to reviewing, verifying, and maintaining generated code, while people still define architecture, service boundaries, safety, and reliability. The post presents Go as a language designed for software engineering and argues that its end-to-end platform—built-in formatting, testing, dependency management, security tooling, and a comprehensive standard library—gives teams and AI agents a consistent foundation. Go’s readability, explicitness, standardized syntax through gofmt, static typing, and single-binary portability are described as guardrails that make generated code easier to inspect, validate, and operate. It also identifies gopls, the rebuilt go fix with modernizers, profiling, execution tracing, and profile-guided optimization as tools for maintaining and tuning systems over time. The conclusion is that Go’s consistency and production-oriented tooling can help humans absorb high-volume AI output without sacrificing reliability, maintainability, or system integrity.
bitdrift brings ClickHouse-powered mobile observability to ClickStack
bitdrift has joined ClickHouse’s House Mates program as a mobile observability partner, bringing mobile-native telemetry, tracing, and debugging to ClickStack. ClickStack provides an open-source or managed observability stack built on ClickHouse, with HyperDX and an OpenTelemetry Collector for logs, metrics, traces, and OTLP ingestion. bitdrift’s Capture SDK supports iOS, Android, and React Native, stores unsampled telemetry in a fixed-size on-device ring buffer, and uses control-plane configuration to decide what is collected or uploaded. Shared OpenTelemetry trace context lets engineers follow a mobile request’s trace ID from a session timeline into ClickStack, where backend traces, dependencies, logs, metrics, database operations, and infrastructure behavior are available, creating a continuous debugging path across the device and distributed systems.
How to ground Genie Agents in both structured data and documents without losing governance
Databricks Genie Agents can combine structured tables and unstructured documents, but the central challenge is preserving data governance when one agent can access both. The proposed architecture runs agents with the end user’s credentials, using Unity Catalog identity synchronization, object privileges, ABAC policies, row filters, and column masks to enforce permissions at query time. Automatic Identity Management syncs users and groups from Microsoft Entra ID and Okta, while just-in-time provisioning carries current memberships into Databricks. Structured assets include tables and metric views, and governed tags can attach protection across newly tagged data; documents are governed through Unity Catalog volumes and their grants. In the Brickstore example, APAC and AMER managers ask the same question yet receive different, authorized results, with regional rows, masked email values, and shared metric definitions; the source recommends testing by impersonating users and planning volume boundaries.
Enabling Seamless Kafka Async Queuing with Consumer Proxy
Uber’s Kafka deployment grew from 1 million to 12 million messages per second, but using Kafka’s ordered, partition-bound consumer model for asynchronous queueing created partition-scalability limits and head-of-line blocking. A one-second RPC example would require 1,000 partitions to reach 1,000 events per second, while non-uniform latency and poison-pill messages could stall unrelated work. Consumer Proxy addresses the mismatch by fetching messages through Kafka’s binary protocol, pushing them individually over gRPC, collecting processing statuses, and committing offsets only when safe. It also provides features including out-of-order commits, dead-letter-queue operations, flow control, and circuit breaking while hiding Kafka consumer-group management from services. At Uber, Consumer Proxy became the primary async-queueing solution, with hundreds of services onboarded; adaptive flow control and Kafka cluster federation are planned next steps.
amazon.science
AWS Trainium Frontier competition: Co-design models and kernels on purpose-built AI chips
Modern language model architectures evolved around specific hardware constraints, but AWS Trainium chips alter memory and compute dynamics through on-chip SBUF memory, systolic matrix multiplication, and explicit DMA control. The AWS Trainium Frontier competition invites teams to train language models from scratch under strict wall-clock time constraints to uncover optimal architectures and custom kernels for Trainium silicon. Participants start from a fifty-million-parameter dense baseline and can modify everything from the model architecture and optimizer to custom Neuron Kernel Interface routines. Phase 1 evaluates thirty-minute single-chip training runs by validation bits-per-byte, while Phase 2 grants four-hour multi-chip server access to evaluate both training efficiency and reasoning on the CORE benchmark. Top finalists will present their findings at NeurIPS 2026, compete for cash prizes, and co-publish research with Annapurna Labs.
Using the GitHub Copilot SDK for Java
The GitHub Copilot SDK for Java offers a framework-agnostic client library to orchestrate AI agent sessions and tool execution directly from server-side Java code. Unlike framework-dependent alternatives, the SDK supports direct model providers such as OpenAI, Azure, and Anthropic through custom endpoint configurations without requiring a Copilot subscription. Developers can register tools declaratively using the experimental @CopilotTool annotation processor or dynamically through inline lambda definitions with ToolDefinition.from. When integrated into a Jakarta EE 11 application on Open Liberty, agent workflows run on container-managed virtual threads that propagate CDI and transaction contexts during blocking calls like sendAndWait. Real-time event subscriptions capture model execution steps and tool invocations to stream updates over WebSockets without exhausting platform threads.
Introducing Claude Sonnet 5
Claude Sonnet 5 is introduced as an agentic Sonnet model that plans, uses browser and terminal tools, and operates autonomously. The post says it approaches Opus 4.8 performance at lower prices while improving on Sonnet 4.6 in reasoning, tool use, coding, and knowledge work. Comparisons on BrowseComp and OSWorld-Verified across effort levels show a wider cost-performance range, with higher-effort results matching Opus 4.8 on some tasks. Safety assessments found lower overall undesirable behavior, hallucination, and sycophancy rates than Sonnet 4.6, while default cyber safeguards address dangerous usage; the model is available across all plans and via the Claude API for $2 per million input tokens and $10 per million output tokens.
Everything we launched during Agents Week
Agents Week introduced a series of products and primitives across Cloudflare to support an Agentic Internet where autonomous software and humans interact. Developing capable agents requires addressing technical challenges spanning identity, communication, orchestration, memory, observability, and security. Cloudflare introduced the Agent Development Lifecycle alongside tools like Cloudflare Agents for live tracing, replay, and human-in-the-loop approvals. The releases also included programmable CI/CD pipelines capable of repairing failures, Cloudflare Wallets for agent transactions, and Kitesurf, an agent-first browser running in V8 isolates. New protocols and tools like WebMCP, MCPv2, and Cloudflare AI Search aim to make web data and applications readable, discoverable, and searchable by autonomous software.
Innocent until combined: Blocking the lethal trifecta with Omnigent Contextual Policies
An AI agent can exfiltrate private data when one session combines access to confidential information, attacker-controlled content, and an external communication channel, even though each capability is individually authorized. Omnigent's contextual policy tracks those three legs as session state: private data, untrusted content, and exfiltration, with human-defined tool or argument-based classifications. It allows ordinary actions until both prerequisite legs are lit, then denies the outbound call that would complete the trifecta. In the support assistant example, a user-supplied ticket instructs the agent to read internal Q3 revenue figures and email them externally; without the policy, the confidential number leaves in a reply, while the protected run blocks that email. A password-reset ticket still proceeds because it lights only the untrusted-content leg, and an empty internal lookup does not light the private-data leg.
Zero-Sum by Design: 10 Years of Uber’s Payments Platform
Uber’s Payments Platform, internally called Gulfstream, grew from a small 2016 microservices effort into infrastructure supporting $217 billion in annualized gross bookings, nearly twice that amount in total money movement, and ledger balances for more than 1.2 billion entities. It is built around immutable Money Orders, zero-sum double-entry accounting, strongly consistent ledger balances, and three core models: money orders, ledgers, and entity changelogs. Independent stateless services communicate primarily through Apache Kafka, while generic abstractions for lines of business and payment instruments let new products and methods plug into the platform; synchronous in-session flows use Cadence. As scale and audit requirements increased, Uber introduced Entity Fares, addressed hot ledger entities, and built LedgerStore for auditability and tamper-evident record-keeping, while also extending the ledger as a multi-tenant service.
Whats new in ClickStack - June + July
ClickStack’s June–July releases add a broader set of observability features, with work centered on trace navigation, metrics sources, dashboards, and filter performance. The trace waterfall now assigns stable colors by service, keeps span durations outside bars, preserves clickable proportions for very short spans, and adds a minimap plus OpenTelemetry span-link navigation. Metrics support includes quantiles for exponential histograms, a leaner default schema, and an experimental UI connection to external Prometheus-compatible endpoints, while PromQL is proxied to the configured endpoint. Dashboard filters can narrow one another, event patterns can appear as tiles, kiosk mode provides a read-only display, and filter-value discovery prefers ClickHouse text indexes when available. Additional changes cover alert annotations, browser RUM dashboard views, external links from table tiles, grid snapping, and service-map coloring by latency, error rate, or throughput.
Top topics
- Architecture834
- Performance816
- LLMs660
- Machine Learning622
- Open Source563
- AI Agents553
- AI470
- Developer Experience448
Weekly digest
The best engineering reads, once a week.