Loading…
Browse by type
Explainer reads
Explainer posts on Yomu, summarised newest first.
- 1,364 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.
Explainer posts
What is an AI Assistant?
AI assistants are software systems that use language models, data retrieval, and reasoning to understand natural-language requests and return information or perform tasks. For enterprise data teams, they can generate SQL, build dashboards, troubleshoot pipelines, and automate repetitive work, but their usefulness depends on integration with data catalogs and governance rather than generic text generation alone. The described pipeline captures input, parses intent and context, retrieves knowledge, generates an output, and either delivers it for review or executes an action; advanced systems incorporate tool and API use plus feedback from corrections and approvals. The article categorizes assistants as voice, conversational, general-purpose, domain-specific, data and developer, or autonomous agents, while noting risks including hallucinations, privacy exposure, integration complexity, over-reliance, and bias. It presents assistants as augmenting human work through faster, more accessible, and more scalable decision support, not replacing human oversight.
Agentic payments for marketplaces: How a smart fridge can shop safely
GreenMart’s smart-fridge scenario addresses how an agent can turn a detected grocery need into an authorized order without receiving unrestricted access to household card or wallet credentials. GreenMart uses Stripe Agentic Commerce Suite to expose its catalog and checkout through a hosted Agentic Commerce Protocol (ACP) endpoint, while retaining responsibility for inventory, pricing, fulfillment, and order operations. A fridge agent discovers products and creates a checkout, and a companion app evaluates the final checkout against deterministic household rules covering items, totals, delivery, substitutions, and budgets. After approval, Link can provide bounded payment authority, while direct Shared Payment Tokens (SPTs) support lower-level Stripe integrations; virtual cards are described for eligible non-Stripe merchants. The design emphasizes auditable policy decisions, explicit escalation or denial, idempotent retries, webhook reconciliation, and stopping when changes exceed the household’s authority.
What are Agentic Workflows?
Agentic workflows are AI-driven processes that let one or more agents plan, execute, and refine multi-step tasks toward a defined goal with minimal human intervention. Unlike chatbots and fixed automation, they evaluate context and intermediate results at runtime, gather data, select tools dynamically, and adapt when outcomes are incomplete, incorrect, ambiguous, or unexpected. The described pattern combines agents and LLMs with APIs, databases, code executors, prompts, feedback mechanisms, memory and state management, and, for complex work, coordinated specialized agents. Production reliability depends on clear success criteria, action guardrails, human review points for high-stakes actions, audit trails, governance, and production-grade infrastructure. Examples include customer-service resolution, fraud investigation, supply-chain optimization, and event-driven document processing across systems.
What is Tool Calling?
Tool calling lets an AI model interact with external tools, APIs, and systems to retrieve information or perform actions beyond its training data, turning a conversational model into an agent that can act. The process is an application-mediated loop: the model recognizes when outside help is needed, selects a tool from schema-described options, emits a structured request—typically JSON—and receives the result as context for its response or next step. Available tools can support information retrieval, code execution, process automation, and smart-device or IoT monitoring, while complex tasks may chain multiple calls into a workflow. The piece also distinguishes tool calling from function calling, describing the latter as a subset, and presents MCP as a standardization layer for describing, discovering, and invoking tools, with Databricks Agent Bricks offering governed integrations and agent-development support.
How Cursor Router chooses the right model for the task
Cursor Router selects models for developer turns using production behavior rather than benchmark scores, balancing user satisfaction and cost. Its routing system first uses Compass, a continuous 0-to-1 complexity predictor trained on follow-up behavior, to keep simple turns on a price-efficient model or upgrade complex ones. Then a taxonomy of domains, tasks, and modifiers compares frontier-model strengths, while an optimizer applies a one-sided 75% uplift threshold and each mode's cost budget. The dataset contains hundreds of thousands of live Cursor turns, preserving conversation signals, model-switching effects, inferred performance from subsequent user actions, and token-based costs including cache misses. Current results report Auto Intelligence at above Fable-level satisfaction with 68% lower cost, and Auto Balance outperforming Opus 4.8 at 41% lower cost while raising satisfaction by 3%.
From User Sequences to Scaling Laws: A Multi-Stage Architecture for Meta’s Ads Ranking
Meta’s ads ranking systems process millions of candidates per second under millisecond latency budgets, creating tension between richer sequence models and efficient serving. The post presents a multi-stage architecture that moves long-history user modeling into asynchronous offline transformers, caches user-level embeddings, and combines them with fresh user and ad signals in an online ranking model. Dense tokenization places sparse features and behavioral sequences in one vocabulary, while target-aware multi-head attention learns higher-order interactions between historical behavior and the candidate ad. On real-world ads traffic, performance follows a log-linear relationship with compute across depth, enrichment, width, and sequence length, with balanced growth and diverse sequences identified as scaling levers. Together with broader model innovations, the approach contributed to a cumulative 6% conversion lift on Instagram, 3% on Facebook, and 3.5% lift in Facebook ad clicks, and serves as a core component of GEM.
What is WAL backpressure, and why does ClickHouse Managed Postgres need it?
Postgres writes every change to the write-ahead log (WAL), and Managed Postgres uploads completed segments to object storage for point-in-time recovery; if uploads lag, retained WAL can fill the disk and cause a PANIC shutdown. ClickHouse Managed Postgres counters this with a data-plane systemd timer that checks pending segments every 15 seconds and uses cgroup v2 I/O controls to throttle client write bandwidth at 80%, 50%, or 20% of provisioned throughput as backlog thresholds rise. The drain path—including the archiver, checkpointer, and related processes—is placed in an immune cgroup, while client backends are throttled and reads remain uncapped. In a test with a 500 MB/s gp3 volume and an archive limited to 4 MB/s, the backlog reached 3,433 segments and 53 GiB before recovery. Once the limit was removed, the archiver cleared the queue, the throttle released automatically, and the disk stayed below 33% full.
Explorers, exploiters, and the myth of the 100x engineer
Engineering leadership often responds to rapid AI productivity gains by attempting to isolate and replicate the traits of individual outlier engineers. Vivek Raghunathan, SVP of engineering at Snowflake, frames this dynamic using reinforcement learning's explore-versus-exploit split, where roughly five percent of an organization fearlessly tests new AI tools while ninety-five percent prioritize execution on paved paths. Instead of treating these categories as fixed personality types or relying on external hiring, organizations succeed by viewing them as a continuum. Effective leadership establishes structured learning time, AI communities of practice, and direct mentorship to capture explorer discoveries and elevate the median engineer. Measuring progress requires tracking organizational movement along this capability scale rather than celebrating isolated high-performing individuals.
Scaling AI Agent Infrastructure with the MCP Stateless updates
The 2026-07-28 Model Context Protocol specification release candidate addresses production bottlenecks caused by the earlier session-oriented HTTP model, including session pinning, sticky routing, lost state, and Redis-backed infrastructure. It removes the initialize/initialized handshake and Mcp-Session-Id, carrying protocol version, client information, and capabilities in _meta on each request while exposing method and name through HTTP headers. The update also introduces ttlMs/cacheScope caching, Multi Round-Trip Requests for elicitations and task progress, issuer and resource-indicator verification, and full JSON Schema 2020-12 support. The post says the stateless core enables ordinary load balancing, serverless deployment, and transparent failover, while SDK beta releases for TypeScript, Python, Go, and C# support migration.
Sidekick's continual learning loop
Frontier models can accelerate an AI product launch, but their serving cost and latency can become impractical as usage grows, while their weights do not learn from production failures. Shopify’s proposed flywheel captures low-quality conversations, defines quality with a rubric and calibrated judges, improves the surrounding harness through autoresearch, and then converts repaired trajectories into supervised fine-tuning and GRPO training signal. A panel of frontier reasoning models critiques failures, an arbiter produces repair instructions, and unresolved cases go to expert annotators; gist compression then shortens the agent’s static prompt without measured judge-quality loss. In production, the GraphQL agent serves up to 2,000 requests per minute and reportedly achieves a 96% serving-cost reduction versus an estimated frontier-model cost, with lower latency and higher throughput. The approach is presented as a continual-learning system that turns production experience into model-weight updates.
Beyond VLAs: How World Action Models Reshape Robot Manipulation
Robot policies built on vision-language-action models often struggle with physical generalization because vision-language backbones describe scenes instead of predicting physical dynamics. World action models address this limitation by building policies on video world models that jointly predict actions and future world states. NVIDIA Cosmos 3 provides an open foundation using a Mixture-of-Transformers architecture where an autoregressive transformer guides a diffusion transformer to synthesize continuous modalities including video and robot actions. Specializing Cosmos 3 on the DROID platform produced policies capable of simultaneously outputting action trajectories and predicted video observations while boosting RoboLab success from 28.1% to 36.8%. These models support flexible deployment tiers, from workstation serving on RTX PRO 6000 GPUs to real-time 15 Hz on-device execution on NVIDIA Jetson Thor embedded hardware.
The Software Supply Chain Is Under Siege. Devs Are Still the First Line of Defense
A research report by Omdia examines escalating software supply chain security risks driven by growing dependencies on third-party code, open-source software, and artificial intelligence. Over three-fourths of surveyed organizations experienced a software supply chain incident in the preceding year, with known vulnerabilities in third-party software representing the most common exploit vector. Surveyed respondents ranked AI technology as their top supply chain concern, alongside broader issues involving vulnerability remediation and code composition identification. While shifting security left is a high priority for 98% of organizations, 45% still lack confidence in the robustness of their supply chain defenses. Although secure container libraries and Software Bills of Materials provide measurable mitigation benefits, many organizations only generate inventories on a case-by-case basis.
The 9 best tools for your early-stage startup tech stack in 2026
Early-stage startups face a crowded software market and limited budgets, making technology selection a question of leverage rather than completeness. The guide proposes three selection principles: choose tools that scale with usage, integrate with the existing stack, and solve a clearly defined problem. It then recommends Stripe for payments, Slack for collaboration, Granola for meeting notes, Fin and Intercom for customer service, Linear for planning, Datadog for observability, MongoDB for data, Rippling for workforce administration, and Apollo for outbound sales and CRM. The examples emphasize operational coverage, from Stripe’s subscriptions and MongoDB’s flexible schema to Fin’s stated ability to resolve up to 76% of support conversations. The Fin Startup Pack is presented as a way to reduce adoption costs through partner credits and discounts, including more than $500,000 in credits and specified offers for each recommended tool.
GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model
Meta’s Generative Ads Recommendation Model (GEM), a hybrid foundation model for ads recommendations on Instagram and Facebook, trains across several thousand latest-generation GPUs and combines trillions of sparse embedding parameters with billions of dense parameters. The post frames its engineering problem around recommendation-specific jagged and asymmetric sequences, memory-bound operations, numerical sensitivity, communication overhead, activation pressure, and load skew that make conventional LLM training techniques insufficient. Its co-design approach combines custom kernels such as Jagged Flash Attention (JFA), BlockAttention, and Generalized Dot-Product Attention (GDPA), mixed ultra-low-precision training including MXFP8 attention and MLP, topology-aware 5D parallelism, SM-free collectives, and memory and load-balancing techniques. The reported result is a doubling of end-to-end training efficiency to 20–25% Model FLOPs Utilization (MFU) while total training FLOPs increased fourfold over 12 months; Base Batch Shuffling (BBS) alone delivered a reported 4% efficiency gain, with 4% QPS improvement and 4% peak memory reduction.
The New Monday Morning Report: How Generative AI can deliver the insights your executives need.
Retail and CPG joint planning meetings often spend time reconciling fragmented, stale, and inconsistent data instead of deciding how to adjust spend, inventory, or forecasts. The proposed Monday Morning Brief replaces the manually stitched weekly deck with an overnight agent-generated brief built from continuously refreshed point-of-sale, shipment, and inventory data alongside trade, promotion, forecast, and external signals. It uses Delta Sharing for a shared view, Genie Ontology for business context, Unity AI Gateway for permissions, guardrails, logging, rate limits, and human approval, and supports models across AWS, Azure, and Google Cloud. The agent scans item-store combinations, ranks material watchouts, answers cited natural-language follow-ups, and can draft actions such as purchase-order changes, while humans retain approval for consequential decisions. The proposed progression from report to ritual to intelligent decision system includes a 90-minute workshop, a day-30 pilot scope, and a first live Monday targeted for day 90.
Scaling real-time AI agents with session-aware load balancing
Real-time AI agents require infrastructure that manages continuous, bidirectional conversations rather than isolated requests, including interruptions that halt speech generation and update context without dropping the connection. The post argues that QPS and instantaneous CPU utilization are insufficient because they miss long-lived committed work and silent sessions that can become active simultaneously. It proposes application-level active-session tracking around the stream lifecycle, with cleanup in finally blocks and consistent snapshots, then combines session pressure with CPU and memory utilization in a hybrid routing model. The model normalizes session counts into a rate and uses target utilization, smoothed utilization, cost per session, and a safety scaler to estimate effective capacity and regulate new assignments. Validation should use long-lived, varied workloads and metrics such as session distribution, overloaded assignments, startup latency, dropped sessions, and counter behavior; high-concurrency trackers may require sharded or LongAdder-style aggregation.
How AI is transforming analytics at Grab
Grab describes analytics as central to business decisions and presents AI agents as a way to democratise access to trustworthy answers. Over two years, increased model capability has enabled agents to prepare data, write and run queries, analyse opportunities, design experiments, interpret results, and draft commentary in minutes rather than a week. The proposed ladder ranges from AI-assisted work through human-planned and agent-planned workflows to end-to-end autonomy, while people retain accountability for framing, definitions, causal interpretation, decisions, and stakeholder relationships. Grab describes Spartan, which routes natural-language Slack requests across more than 50 skills and 120 analysis frameworks, plus Scarlet for pipeline failure triage and recovery. The programme also maintains agent context through ContextIQ and reports lower analyst intervention, a roughly 33% reduction in cycle times, and greater self-serve resolution in its support channel.
Backstage with Lakebase, part 3
This third part describes joining Backstage’s live operational Postgres catalog with analytical billing data so teams can query resource ownership and cloud spend together. Lakebase isolates compute per workload, allowing portal traffic and FinOps queries to use the same underlying storage; benchmarked catalog queries ran at 55–65 ms end-to-end and searches at two to four ms. Because Lakehouse Federation’s Postgres connector accepts static credentials while Lakebase applications use OAuth JWTs, the proof-of-concept creates a separate SCRAM-SHA-256 Postgres role for federation. This enables a single query to join Backstage resource names with Unity Catalog’s system.billing.usage without data movement, while branch billing remains visible per branch and endpoint. The operational considerations include password rotation, read-only grants, branch TTLs, and scale-to-zero endpoints that stop billing when idle.
Co-Designing AI Model Attention for Fast, Interactive Long-Context Inference
As context lengths increase in long-context and agentic workloads, attention increasingly dominates GPU inference runtime. Dense attention performance depends on architectural parameters like group size, head dimension, and sequence length, alongside GPU parallelization strategies. Prefill remains compute-bound and scales quadratically with input sequence length, showing minimal runtime sensitivity to group size variations. Conversely, decode generation is memory-bound and scales linearly with KV cache length, where doubling group size roughly halves decode execution time. To maximize throughput and interactivity on NVIDIA hardware, model architects should push group sizes higher, select head dimensions of 128 or 256, reduce effective KV state, and keep tensor parallelism less than or equal to the number of KV heads.
Using Triggers On Schemaless, Uber Engineering's Datastore Using MySQL
Schemaless triggers provides a scalable, fault-tolerant, and lossless way to listen for changes in Schemaless, Uber Engineering’s datastore using MySQL, and drives workflows including trip billing, analytics-warehouse ingestion, and cross-datacenter replication. Schemaless stores immutable cells and all their versions, allowing each shard to function as a partitioned change log with monotonically increasing added IDs while retaining random-access APIs. The framework tails shards through get_cells_for_shard, assigns them to worker processes under a leader, persists offsets in shared storage, and invokes column-specific callbacks such as bill_rider; handlers must be idempotent because retries can occur. Failed workers’ shards are reassigned, repeatedly failing cells can move to a separate queue, processing continues during leader or shared-storage outages although redistribution pauses, and replicas insulate triggers from database-node failures.
Top topics
- Architecture832
- Performance814
- LLMs658
- Machine Learning620
- Open Source563
- AI Agents549
- AI465
- Developer Experience448
Weekly digest
The best engineering reads, once a week.