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
Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains
JetBrains is releasing Mellum2, a 12B-parameter Mixture-of-Experts model trained from scratch on natural language and code. It targets latency-sensitive, high-throughput workloads including routing, retrieval-augmented generation (RAG), summarization, sub-agents, and coding features. The model has 12B total parameters but activates 2.5B per token, with its architecture keeping total capacity high while making inference more efficient for real-time use. Released under the Apache 2.0 license, Mellum2 is described as competitive with similarly sized open models on code generation, reasoning, science, and math benchmarks while delivering more than 2x faster inference. Its text-and-code focus supports high-throughput production workloads and private deployments where proprietary code or internal data may be involved.
Anthropic raises $65B in Series H funding at $965B post-money valuation
Anthropic says it has raised $65 billion in Series H funding at a $965 billion post-money valuation, with the round led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital. The company says its run-rate revenue crossed $47 billion earlier this month and that the financing will support safety and interpretability research, additional compute, and expanded products and partnerships. It reports agreements for up to five gigawatts of new Amazon capacity, five gigawatts of next-generation TPU capacity with Google and Broadcom, and GPU access in SpaceX’s Colossus 1 and Colossus 2; AWS remains its primary cloud provider and training partner. The round includes $15 billion in previously committed hyperscaler investments, including $5 billion from Amazon, alongside strategic infrastructure partners Micron, Samsung, and SK hynix.
Debunking 8 data layout myths: why Liquid Clustering outperforms partitioning
Partitioning has long been the standard physical layout for Hadoop- and Hive-era processing, but the post argues that changing Lakehouse workloads make fixed choices brittle. Liquid Clustering treats keys as guidance for file organization, allowing layouts to evolve, supporting low- and high-cardinality filters, maintaining file sizes, and enabling row-level concurrency. Using transaction-log metadata and per-column statistics, the post says modern Delta and Iceberg engines prune files rather than directories, and reports 35% lower clustering time, 22% faster queries, roughly 90% faster metadata-only deletes, and 23-minute planning for a 10-PB table. Conversion examples include Bolt’s 138% higher write throughput and up to 63% faster reads, while a 1.1-PB workload saw 5.9x faster queries, 86% fewer bytes read, and a 27% smaller table after clustering by date, hour, source, and id.
Beyond LLMs: Why Scalable Enterprise AI Adoption Depends on Agent Logic
Enterprise workflows are dynamic, long-running, API- and data-rich, and constrained by policies or regulations, creating challenges that an LLM alone may not handle efficiently. The article defines agent logic as software primitives—including knowledge graphs, algorithms, and program-analysis libraries—within an agent harness that steer models toward relevant workflow context. Examples span legacy COBOL and PL/1 understanding, automated test generation, incident investigation and code remediation, and compliance modernization; reported results include ~30× lower token consumption for application understanding, 20–45% higher test coverage, and up to 4.0× better incident-investigation performance than a ReAct baseline. A Condition Insights example reports 57% fewer unsupported claims and 77% lower average token usage, supporting the conclusion that structured guidance can improve quality, cost, and rule adherence.
Improvements to Teams Pricing
Teams plan changes increase included usage, add a Premium seat for heavy agent users, and give administrators more visibility and control over spending. Standard seats now separate included usage into a First-party models pool for Auto and Composer 2.5 and a Third-Party API pool, with no cost change: $32 per seat monthly on annual plans or $40 on monthly plans. Premium seats provide five times Standard usage at three times the price—$96 annually or $120 monthly—and can be mixed with Standard seats; the post says the first-party pool should cover a full month of heavy agent usage for 99% of users. The dashboard reports usage by pool and can recommend seat types, while rebuilt alerts support dollar thresholds delivered through Slack or email. Changes apply immediately to new customers and to renewing customers whose billing cycles start July 1, 2026.
Reinforcement learning is an infrastructure problem
Reinforcement learning for post-training large language models has become infrastructure-constrained as teams move to multi-node configurations for larger models like Qwen 3 30B and Kimi K2.6. The RL loop divides into three independent infrastructure challenges: training with reliable forward and backward passes, high-performance inference rollouts, and thousands of concurrent sandbox environments for policy execution. Weight synchronization between trainer and rollout engine becomes the critical bottleneck in multi-node setups, with RDMA offering up to 100x speedup over TCP and significant cost savings per step. Teams commonly get stuck maintaining glue code, queueing for cluster time, and under-utilizing GPUs when sandbox buffers are not correctly sized. Modal addresses these with RDMA-connected clusters via a single rdma=True keyword, fast container boot, and the new open-source Modal Training Gym library that abstracts common scaffolding into under 100 lines of code.
Scaling up your microservice testing with Apache Spark
Microservices whose behavior depends on long-tail production inputs—such as payment cost estimation—are difficult to validate with unit and integration tests alone. The described pattern separates deterministic service logic from transport and side effects, then invokes that logic in Apache Spark across historical records enriched with the required dependency state. The resulting datasets can compare current and candidate implementations, quantify changed outputs, model future rule changes, and identify affected transaction or merchant segments. Replay inputs and artifacts are minimized and protected through tokenized identifiers, redaction, access controls, and retention policies, while smaller privacy-safe golden datasets run in pull-request CI. The approach complements, rather than replaces, unit and integration tests: golden datasets cover known service behavior, and full replays expose changes across the real production distribution before deployment.
Executable UDFs are now in public beta on ClickHouse Cloud
Executable UDFs are available in public beta on ClickHouse Cloud, allowing Python functions packaged as zip files to be called from SQL like built-in functions. ClickHouse runs them in a pool of long-lived sandboxed processes, keeping model code near the data and supporting use in ad-hoc queries, joins, and materialized views. The demonstration wraps a roughly 270K-parameter PyTorch autoencoder in an embed_trade UDF that consumes 14 features, produces a 32-dimensional embedding and reconstruction error, and scores trades as they enter through a materialized view. Training uses streamed Parquet chunks, incremental StandardScaler fitting, early stopping, and saved model artifacts; the deployed pool keeps the 2MB model hot, so inference occurs once per trade and downstream queries read the resulting table.
Enabling Evolutionary Database Development: database branching with Lakebase
Evolutionary database development treats migrations as first-class CI/CD artifacts, but shared development databases still limit per-developer isolation and fast, realistic feedback. The post follows Jen as she adds location, batch, and serial fields to inventory, coordinating application changes, production-data preservation, schema migration, and tests for storing, reading, and displaying the new values. It contrasts shared databases and simplified local substitutes with Databricks Lakebase copy-on-write branches, created in one second with zero storage at creation, including for a terabyte-scale production database. Jen pairs code and database branches, while CI creates a temporary branch, validates an idempotent and reversible migration, runs application and database tests, and posts a schema diff, enabling combined review and shifting DBA attention toward design and maintainability.
AI Doesn't Scale Until You Stop Calling It Innovation
Enterprises often stall between AI proofs of concept and production, and Philippe Rambach argues the remedy is to manage AI as a product rather than innovation. At Schneider Electric, Rambach built a 400-person AI organization split evenly between customer-facing products and internal operations, with business cases owned by lines of business and cross-functional scrum teams responsible through production and support. It standardizes a single core technology set, with Databricks managing infrastructure, data, and data flows, while gate reviews and quarterly portfolio decisions test technical readiness, commercial viability, and the business plan. Models are combined with context, guardrails, interfaces, forecasting, optimization, and real-time decisions; Microgrid Advisor reports up to a 20 percent reduction in energy costs, while Genie’s internal rollout remains early and accuracy is still being addressed.
Databricks at SIGMOD 2026
Databricks says its work on Spark Declarative Pipelines will be featured at SIGMOD 2026, where it received an honorable mention award, and that it will attend June 1–5 in Bangalore as a Platinum Sponsor. The announcement describes two mix-and-match incremental programming paths in SDP: materialized views maintained by the Enzyme engine and streaming APIs with stateful operators and watermarks. Enzyme applies incremental view maintenance to ETL, supporting production materialized views with joins, window functions, aggregations, combinations of these patterns, non-deterministic functions such as current_date(), and AI-specific functions. It also supports Python as well as SQL, detects changes in materialized-view definitions, chooses partition- or row-level updates, selectively caches intermediate results, and uses a cost model based on plan information and prior executions; the post says it outperforms an anonymized competing solution.
Enterprise Business Software and the Mixed-Up Chameleon Problem
Traditional ERP practice prioritizes a standardized, stable core, but the post argues that forcing every process into it can erode an organization’s identity and differentiated ways of creating value. It proposes keeping the ERP as the system of record while placing an interoperability layer above it. This layer uses a shared ontology to reconcile concepts across systems and supports workflows, automation, user experiences, agentic AI, and human engineering under shared security and governance. Because these capabilities remain outside the core, organizations can preserve distinctive processes during ERP upgrades and treat migration as a continuous software- and ontology-driven activity rather than a one-time manual exercise. The intended model is structured adaptability: standardized ERP functions retain discipline, auditability, and financial integrity while operational software adapts to the business.
Winning under CMS TEAM: Building the learning health system to realize success in VBC today and tomorrow
CMS’s Transforming Episode Accountability Model (TEAM), beginning January 1, 2026, requires selected organizations to manage total cost and quality across five surgical episode categories and a 30-day post-discharge window. The post argues that retrospective dashboards and delayed claims data cannot support intervention or prepare hospitals for downside risk, especially amid complex integration across EHR, claims, post-acute, social-determinants, and provider data. It proposes a modern, cloud-native lakehouse foundation with unified data, operationalized AI/ML, embedded clinical workflows, and scalable architecture. Recommended capabilities include risk stratification, next-best-action decision support, post-acute optimization, provider benchmarking, and predictive performance alerts. In a composite 500-bed medical-center example, the approach is associated with 15% lower SNF costs, 8% better risk-capture completeness, and 12% fewer readmissions.
From Rigid Rules to Empathetic Agents: Rebuilding Swiggy’s Order Cancellation Experience with LLMs
Swiggy rebuilt its “I Want To Cancel” (IWTC) support journey because a roughly 150K-line deterministic decision tree treated cancellation keywords alike, repeated canned deterrence, and could leave users stuck without handoff. The replacement is a stateful, multi-turn framework orchestrated with LangGraph: an intent layer identifies needs such as delay help, coupon support, direct cancellation, or human access, while live order signals and deterministic policy checks determine the next action. Full graph state is checkpointed in Redis by order- and session-level thread IDs, allowing conversations to resume across turns. Swiggy evaluated intent reliability using more than 100k historical chats and a pre-deployment suite of about 500 representative queries, then operated the system with reporting and alerts for behavioral and service metrics. The conclusion favors a hybrid design in which LLMs handle language while deterministic systems anchor truth and business decisions.
Explaining Every Rupee: How We Built Reliable LLM Support Bots for Delivery Partners
Swiggy faced close to 9 lakh monthly payout- and incentive-related tickets, where delivery executives needed accurate explanations of earnings, deductions, missed incentives, and credited amounts. The system separates financial truth from language generation: deterministic preprocessing and internal APIs resolve incentive eligibility, payout breakdowns, deductions, and conditions before an LLM explains the result. Two multilingual bots—the Daily Incentive Bot and Payout Bot—use a LangGraph StateGraph with LLM, tool, and validator nodes, Redis checkpointing for multi-turn context, and post-processing translation into the executive’s preferred language. Azure Translator preserve-list controls help keep business-critical terminology consistent, while structured escalation sends out-of-scope cases and context to human agents. Moving retrieval from Databricks SQL queries to an aggregated internal API reduced latency from 8–10 seconds to milliseconds; the bots now resolve about 90% end to end, cut resolution from roughly 20 minutes to seconds, and improved CSAT by 40%.
Profiling in PyTorch (Part 1): A Beginner's Guide to torch.profiler
The post introduces torch.profiler through a beginner-oriented walkthrough of profiling a matrix multiplication and addition, explaining how to configure the profiler and interpret its outputs. It uses a small function with torch.matmul and torch.add, annotates it with record_function, captures CPU and CUDA activities after warm-up iterations, and exports a profiler table plus a Chrome trace. The table summarizes event counts and CPU or GPU time, while the trace shows execution order, call hierarchy, CPU-GPU gaps, kernel launches, and overlap. Comparing 64x64 and 4096x4096 matrices demonstrates a shift from overhead-bound behavior to compute-bound behavior; the guide also covers dispatch chains, hidden device-to-device copies, profiler buffer activity, and torch.compile's runtime and caching behavior.
Compliance versus conversion: Tax ID validation at checkout
Tax ID collection at checkout supports compliant tax invoices and tax determination, including reverse-charge treatment for qualifying cross-border business transactions. Validation can be performed synchronously before purchase completion or asynchronously after checkout: synchronous checks reduce correction and audit risk but may increase abandonment and expose conversion to government-service downtime, while asynchronous checks preserve completion but require follow-up on failures. The implementation discussion covers direct national-registry integrations, third-party validation services, and Stripe’s CheckoutSessions and PaymentIntents integrations, where the Tax ID Element uses verification modes of if_supported or none. Stripe first applies a country-specific format check, routes validly formatted IDs to government APIs, may fall back to asynchronous processing after four seconds, and exposes status through customer.tax_id.updated and the customer object.
From decentralized Docs-as-Code to a centralized repository: Evolving Grab's documentation strategy
Grab adopted Docs-as-Code in early 2021 to bring technical documentation into engineers’ code workflows, addressing scattered sources, version confusion, discoverability, and quality gaps. Documentation was written in Markdown, stored in version-controlled repositories, reviewed through merge requests, checked in CI/CD, and surfaced through an internal developer portal. As engineering scaled, decentralized repositories produced inconsistent structures and terminology, unreliable organization-wide quality automation, aging content, and tracking problems when source locations changed. Grab therefore moved to a centralized repository while retaining Docs-as-Code, adding shared templates and linters, unified search through fuzzy title lookup and Glean, automated synchronization during migration, and training and feedback programs. The post concludes that decentralized, centralized, or hybrid models fit different contexts, with centralization emphasizing discoverability, governance, consistency, ownership metadata, and analytics.
How enterprise leaders are scaling AI agents across their organization
Executives from Danone, Capital One, Warner Bros. Discovery, Ford Credit, Gilead Sciences, and Databricks discuss scaling AI agents for measurable business value while maintaining governance, trust, and cost control. They describe deployment as only the first step, followed by monitoring, observability, performance assessment, and continuous learning, with risk reviews and ongoing evaluation integrated into the agent lifecycle. Organizations are shifting from single-task automation toward specialized multi-agent orchestration for complex workflows, while shadow capabilities and controlled sandboxes test accuracy without exposing live operations. Leaders recommend building momentum through low-risk wins such as Capital One’s Chat Concierge, training employees to prompt, and providing natural-language interfaces, certified data, consistent guardrails, and a secure architecture for deployment and monitoring.
Advancing Apache Iceberg on Databricks: Iceberg v3 GA, Open Sharing, and Unified Governance
Databricks announces a broad set of Apache Iceberg capabilities in Unity Catalog, spanning General Availability, previews, and beta releases. Managed Iceberg is GA, supporting table creation, reads, writes, optimization, governance, and sharing, while Iceberg v3 adds deletion vectors, row tracking, and VARIANT across managed, foreign, and UniForm-enabled tables. Unity Catalog also federates external catalogs, vends credentials, shares live data with Iceberg REST-compatible clients through Delta Sharing, and applies attribute-based access control during server-side scan planning for supported external engines. These capabilities are presented as a unified approach to open APIs, cross-engine governance, zero-copy sharing, and production performance without copying data. The post also outlines Iceberg v4 and a proposal for Delta 5.0 to adopt an adaptive metadata tree structure.
Top topics
- Architecture834
- Performance816
- LLMs660
- Machine Learning622
- Open Source563
- AI Agents553
- AI470
- Developer Experience448
Weekly digest
The best engineering reads, once a week.