Loading…
Browse by type
Explainer reads
Explainer posts on Yomu, summarised newest first.
- 1,318 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
Steam, Steel, and Infinite Minds
The post frames AI as a new “miracle material” that could reshape knowledge work as steel and semiconductors reshaped earlier eras. It argues that coding agents already let one programmer orchestrate several AI coding agents, but broader knowledge work remains constrained by fragmented context across tools and weak verifiability. At the organisational level, it compares AI with steel and steam, suggesting agents could preserve workflow context, surface decisions, and reduce dependence on meetings and hierarchical approvals; Notion is described as using more than 700 agents alongside 1,000 employees for repetitive tasks. At economy scale, it predicts organisations spanning humans and thousands of agents, with continuous workflows and new rhythms, while acknowledging that this transition may be disorienting and requires redesign rather than simply adding chatbots to existing tools.
Closing the Books Without the Spreadsheet Shuffle: My Fall Internship at Ramp
During a fall internship at Ramp, the author worked on reducing repetitive accounting work around month-end close, focusing on amortization and ERP mapping workflows. One project uses receipt data, merchant category codes, customer history, and embedding similarity to detect likely prepaid expenses and suggest service periods and accounts or templates, with recommendations reviewable before application. A second built Ramp-Native Amortization to create schedules, generate periodic debit and credit journal entries, handle catch-ups and reclassifications, and sync entries to ERPs through a dashboard; transaction and reimbursement support was nearing alpha production release, while bill pay was next. The third generates high-confidence mapping-rule suggestions from customer coding patterns on a nightly schedule. Across the work, backtesting, accuracy metrics, and human workflow research support automation intended to earn trust rather than remove accountant judgment.
Tokenization in Transformers v5: Simpler, Clearer, and More Modular
Transformers v5 overhauls its tokenization framework by separating tokenizer architecture from trained vocabularies. In contrast to v4's dual slow Python and fast Rust files, v5 consolidates each model tokenizer into a single file defaulting to the Rust-backed TokenizersBackend. The pipeline stages—normalizer, pre-tokenizer, model algorithm such as BPE or Unigram, post-processor, and decoder—are now directly exposed and configurable rather than buried in serialized files. Practitioners can instantiate blank tokenizer architectures and train custom vocabularies directly from iterators using native methods like train_new_from_iterator while retaining model-specific formatting rules. The wrapper layer continues to bridge raw tokenization and model requirements by managing chat templates, context limits, and special token insertion.
2026 customer service planning series: Vol. 05
AI-first customer service makes capacity planning less predictable because automation changes volume, work mix, human responsibilities, and performance over time. The proposed planning model puts automation rate—AI Agent involvement rate multiplied by AI Agent resolution rate—alongside inbound volume, human output for complex remaining work, and occupancy. It recommends ambitious automation assumptions only when backed by named ownership, content and systems investment, work-type targets, and staged improvement, while modeling lower cases closed per person and more time for off-queue work. Finance should treat the plan as a revisable set of assumptions, review it at least quarterly, and avoid cutting headcount before automation is proven. Surplus capacity should be redirected to AI improvement, product feedback, new channels, or proactive customer experience work rather than automatically reduced.
How we made Notion available offline
Notion's Offline Mode required turning a best-effort SQLite cache into storage that could guarantee a page and all its dependent records were available without a network connection. The system uses a forest of offline page trees backed by offline_page and offline_action tables, allowing independent reasons such as explicit toggles, favorites, automatic downloads, and inheritance to coexist while preserving hierarchy. To keep content current, clients subscribe to push channels emitted by the page version snapshot system, then use lastDownloadedTimestamp and server lastUpdatedTime to fetch only newer pages after reconnecting. When pages or databases change, the client reconciles fresh page and database snapshots with offline_action, adding or removing rows rather than rebuilding entire trees. This makes offline availability precise and eventually consistent, while supporting conflict resolution through the CRDT data model.
Litestream VFS
Litestream VFS lets SQLite query a Litestream-backed database directly from object storage, including historical states selected with SQL pragmas, without first downloading the complete database. It is a loadable SQLite VFS plugin: Litestream v0.5 uses the LTX data-shipping format, whose compaction removes superseded page versions, while index trailers map database pages to remote file offsets. The VFS fetches only requested page ranges through S3, uses an LRU cache, and can poll second-level L0 backups to maintain a near-realtime replica. Writes remain handled by the regular Litestream Unix program, and the existing SQLite library is unchanged. The result is fast querying and instantaneous point-in-time recovery from arbitrary relative or absolute timestamps, while the optional VFS preserves Litestream’s conventional backup and restore workflow.
2026 customer service planning series: Vol. 04
After defining team roles, an operating model is needed to make progress part of customer-support work and keep an AI Agent improving over time. The stated goal is that “the first time you answer a question should be the last,” with each resolution improving the system so fewer issues repeat, quality compounds, and support scales. The post emphasizes clear ownership, risk-aware guardrails, feedback loops, and a culture that values improvement work; it identifies ambiguity as a common cause of performance plateaus. An AI ops lead, often located in support operations or developed from an existing role, owns performance and drives change, illustrated by Dotdigital’s dedicated specialist after resolution volume held at about 2,800 conversations per month for three months; lightweight governance is presented as a way to keep iterating without excessive bureaucracy.
Setting the standard for support: What Spain’s new customer service law signals
Spain’s proposed customer service law would impose stricter requirements on regulated providers and sufficiently large companies, covering response times, human access, complaint handling, accessibility, and auditability. It would require 95% of support calls to be answered within three minutes, free lines, 24/7 support for continuity issues in essential services, and complaints resolved within 15 days or five days for undue charges. The source also describes protections against spam calls, centralized complaint records, multilingual support obligations for some large companies, and simpler subscription cancellation. It presents AI Agents such as Fin as a way to reduce avoidable volume, answer calls, preserve human escalation, structure complaints, communicate outages, and support reporting. The stated conclusion is that early modernization may ease pressure on human teams while helping organizations adapt to rising regulatory and customer expectations.
Engineering the right opportunities for Thumbtack Pros.
Ashmann Syngle, a backend software engineer on Thumbtack's Pricing team, focuses on systems that connect local service professionals with homeowners. To improve matching and revenue operations, the team recently deployed new pricing features, launched experiments, and partnered closely with Data Science and Monetization Experience groups. Current engineering efforts center on infrastructure enhancements, specifically upgrading alerting and monitoring across owned services to raise system reliability and operational efficiency. Because the team maintains complex monetization systems within a codebase that has evolved over many years, engineers conduct deep system analyses to evaluate edge cases before implementing foundational platform modifications. Thumbtack supports these initiatives through a virtual-first operational model complemented by regular in-person offsites.
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.
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.
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.
SpellVault’s evolution: Beyond LLM apps, towards the agentic future
SpellVault began as Grab’s no-code platform for building LLM apps, using Retrieval-Augmented Generation (RAG) and plugins to connect applications with knowledge sources and external systems. Users could ground responses in Wikis, Google Docs, text, and PDFs, invoke custom HTTP endpoints, and access apps through the web, Slack, or APIs. The platform later added citations, Python execution for mathematical problems, scheduled actions, and drag-and-drop Workflows combining LLM calls, code, and Knowledge Vault lookups, while its executor moved toward graphs with nodes, edges, and states. SpellVault now exposes apps and RAG through MCP, allowing external agents and compatible clients to invoke apps and search, retrieve, or add Knowledge Vault information; TinyMCP adds MCP capabilities to FastAPI as a router.
How we prepare Shopify for BFCM
Shopify’s BFCM readiness program addresses the challenge of operating critical commerce infrastructure at traffic levels informed by 2024 records, while preparing APIs with no prior holiday-traffic history. Starting in March, teams combined capacity planning on Google Cloud, an infrastructure roadmap, risk assessments, chaos-engineering Game Days, and a Resiliency Matrix covering failure modes, recovery objectives, runbooks, and on-call coverage. Genghis load tests and Toxiproxy network faults ran across three GCP regions, while five April–October scale tests exercised browsing, checkout, analytics, webhooks, regional failover, and cascading failures; testing reached 146 million requests per minute and a p99 target of 200 million. The work exposed and addressed Kafka partition, memory, timeout, routing, queueing, and failover issues, while the operational plan adds monitoring, 24/7 incident coverage, merchant communications, live tuning, and a post-BFCM review.
First Block with Jesse Zhang, Co-founder and CEO of Decagon
The episode profiles Jesse Zhang, co-founder and CEO of Decagon, whose AI agents support customer service and reportedly achieve 70–80% deflection rates and 3x improvements in customer satisfaction. In its first 18 months, Decagon prioritized short-term execution and customer deployment, then shifted toward longer-term thinking; the source says this helped it move from stealth to unicorn status in a year. Zhang also describes intentional customer discovery, Agent Operating Procedures (AOPs), enterprise trust, and the need to pair technical excellence with commercial thinking. He credits senior generalists from his network with establishing an early team foundation that supported growth from 12 to 200 people in 18 months, while cautioning against applying generic advice instead of judging personal strengths and timing.
2026 customer service planning series: Vol. 01
Part one of a five-part 2026 customer service planning series argues that support organizations must be redesigned around AI rather than adding AI to existing tools, processes, and org charts. It frames AI Agents as end-to-end resolution engines and infrastructure that can serve as the first and last touchpoint, manage workflows, orchestrate handoffs, and take actions. The proposed mindset shift includes planning for changing work distribution, treating support as a product function, measuring resolution, impact, and system reliability, and assigning clear ownership for agent performance and continuous feedback. The post also says teams should give AI increasingly complex workflows while people improve the system, creating compounding returns. It concludes that adaptability will be essential as AI evolves, and previews a future edition on designing the resulting support organization.
Hollywood demos vs. real
AI demonstrations fall into two categories: polished “Hollywood demos” and raw “real-world demos,” with the latter showing how customer-service AI agents perform under messy, complex interactions rather than edited presentation. Voice provides an especially demanding test because timing, tone, turn-taking, latency, interruptions, and real-time adaptation all shape the experience. At Pioneer, Paul Adams called Fin Voice live; within 90 seconds, it verified identity, retrieved account data, handled an interruption, offered options, completed the workflow, and sent a follow-up email. The demonstration exposed brief pauses for backend work, natural conversational flow, tone changes, and unscripted behavior, while later updates added voice customization, escalation guidance, deployment controls, integrations, support for 28 languages, and roughly 30–40% lower latency since launch.
Top topics
- Architecture830
- Performance812
- LLMs648
- Machine Learning617
- Open Source561
- AI Agents542
- AI454
- Developer Experience446
Weekly digest
The best engineering reads, once a week.