---
title: "Latest reads"
description: "The engineering internet, summarised so you can actually read it."
---

# Latest reads
> The engineering internet, summarised so you can actually read it.

## Articles

### [Holotron-12B - High Throughput Computer Use Agent](https://yomu.fyi/post/holotron-12b-high-throughput-computer-use-agent.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Pierre-Louis Cedoz, Hamza Benchekroun, Aurélien Lac, delfosse, Tony Wu, Antoine Bonnet, Kai Yuan, Aleix Cambray (H-AI), Alexandra, Axel Moyal
- Published: Mar 17, 2026
- Type: Announcement

Holotron-12B is a multimodal computer-use model released by H Company and available on Hugging Face. Post-trained from NVIDIA’s Nemotron-Nano-12B-v2-VL-BF16 on proprietary data, it targets agents that must perceive, decide, and act in interactive environments while handling long contexts with multiple images. Its hybrid State-Space Model and attention architecture stores a constant state per layer per generated sequence instead of attention’s per-token key and value activations, reducing memory demands for long-context inference. In a WebVoyager evaluation on one H100 with vLLM, it reached 8.9k tokens/s at concurrency 100 versus 5.1k for Holo2-8B, and its benchmark score rose from 35.1% to 80.5% after training on localization and navigation data.


### [How Socialpruf built a faster, more reliable data stack by replacing Neon with Postgres managed by ClickHouse](https://yomu.fyi/post/how-socialpruf-built-a-faster-more-reliable-data-stack-by-replacing-ne.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: ClickHouse
- Published: Mar 17, 2026
- Type: Problem & solution

Socialpruf, a social media performance tracking platform ingesting hundreds of posts per second, experienced analytics latency, connection dropouts, and high network transfer costs while running Postgres on Neon alongside ClickHouse Cloud. To resolve these issues, the team migrated their transactional system of record to NVMe-backed Postgres managed by ClickHouse, colocating transactional and analytical workloads. Although standard Postgres logical replication failed on their 0.5 TB dataset, they achieved continuous synchronization using PeerDB before conducting a production cutover. The migration eliminated connection dropouts, reduced network transfer fees, and improved overall Postgres query performance by approximately 30 percent. To manage thousands of concurrent connections generated by ingestion workers, Socialpruf implemented ClickHouse's bundled, horizontally scalable PgBouncer instances.


### [Training Composer for longer horizons](https://yomu.fyi/post/training-composer-for-longer-horizons.md)
- Company: [Cursor](https://yomu.fyi/company/cursor.md)
- Author: Federico Cassano & Sasha Rush
- Published: Mar 17, 2026
- Type: Explainer

Composer is trained for long-horizon coding tasks using self-summarization, allowing reinforcement-learning trajectories to exceed the model’s context window. When a fixed token trigger is reached, Composer receives a synthetic summarization query, uses scratch space to produce a condensed context, and resumes with the summary plus conversation state; training rewards every token in the chained rollout with the final outcome. Against a tuned prompt-based compaction baseline, self-summary reduced compaction-related error by 50% on CursorBench at both 80k and 40k triggers, while summaries averaged about 1,000 tokens instead of more than 5,000 and reused the KV cache. An early Composer checkpoint solved Terminal-Bench 2.0’s make-doom-for-mips after 170 turns, summarizing more than 100,000 tokens into roughly 1,000 tokens of retained information.


### [Making AI Write Android Code Our Way: A Practical Guide to Agent Skills](https://yomu.fyi/post/making-ai-write-android-code-our-way-a-practical-guide-to-agent-skills.md)
- Company: [Medium](https://yomu.fyi/company/medium.md)
- Author: Pierrick CAEN
- Published: Mar 17, 2026
- Type: Tutorial

An Android team at Medium describes how it encoded its project conventions so AI assistants generate code aligned with Clean Architecture, Kotlin, Jetpack Compose, Hilt, Apollo GraphQL, and feature modules. An AGENTS.md file provides persistent context about module structure, dependency injection, state management, navigation, testing, and commands, while Markdown skills in .agents/skills/ provide ordered runbooks for specific tasks. Examples cover scaffolding screens and feature modules, migrating Material 2 to Material 3, enforcing ViewModel feature-flag handling, and adding deeplinks, analytics, translations, and tests. Detailed patterns include ViewModel-injecting and @VisibleForTesting composables, assisted injection, StateFlow, Result, MockK, Turbine, and Robolectric. The team iterates skills from review feedback, keeps them version-controlled, and plans file-aware rules plus commands for release diffs and reporting.


### [AI is redrawing the database market](https://yomu.fyi/post/ai-is-redrawing-the-database-market.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Tanya Bragin
- Published: Mar 17, 2026
- Type: Explainer

Emerging artificial intelligence workloads are fundamentally transforming data platform requirements across real-time analytics, data warehousing, and observability. Conversational analytics and autonomous agents replace periodic batch queries with rapid bursts of concurrent, low-latency requests, exposing the architectural limits of traditional cloud data warehouses and standalone transactional databases. To resolve these constraints, architectures are converging on engine-level integrations between Postgres for transactions and ClickHouse for analytical scale, while unifying operational observability and business event storage. ClickHouse also incorporates turnkey agent interfaces through its acquisition of LibreChat alongside dedicated execution tracing from tools like Langfuse. These combined capabilities ensure organizations can reliably support interactive agentic workflows without incurring severe latency penalties or compounding infrastructure costs.


### [Designing the new async-native ClickHouse Python client](https://yomu.fyi/post/designing-the-new-async-native-clickhouse-python-client.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Joe Spadola
- Published: Mar 16, 2026
- Type: Problem & solution

The official ClickHouse Python client, clickhouse-connect, originally addressed asynchronous execution by wrapping its synchronous HTTP client in a thread pool executor. However, this executor model experienced thread exhaustion, GIL contention, and severe tail latency variance under heavy concurrent workloads. To resolve these scalability bottlenecks, the developers redesigned the client around a half-sync/half-async pattern utilizing aiohttp for async network I/O. A custom bounded queue bridges streaming socket reads on the event loop with synchronous, CPU-intensive binary parsing running inside background threads. Benchmarks demonstrate that this architecture stabilizes tail latencies, delivering an average P95 latency of 556 milliseconds compared to 869 milliseconds for the legacy executor client.


### [ClickHouse Release 26.2](https://yomu.fyi/post/clickhouse-release-26-2.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: ClickHouse
- Published: Mar 16, 2026
- Type: Announcement

ClickHouse version 26.2 introduces twenty-five new features, forty-three performance optimizations, and one hundred eighty-three bug fixes across the database engine. This release graduates both the text-index and the QBit vector embedding data type to production-ready status for search and retrieval workloads. For streaming ingestion pipelines where row arrival rates are low, the engine adds time-based block flushing configurations to write data parts before hitting default row or byte thresholds. The update also embeds the ClickStack observability user interface directly into the standard binary distribution, accessible through the local HTTP server port. Further enhancements include automatic min-max indexing for temporal columns in MergeTree tables, interactive time-based one-time password authentication in the client, and a high-performance prime number generator function.


### [Unordered mode for GCS ClickPipes is now available](https://yomu.fyi/post/unordered-mode-for-gcs-clickpipes-is-now-available.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Marta Paes
- Published: Mar 16, 2026
- Type: Announcement

ClickHouse Cloud has added unordered mode to its Google Cloud Storage (GCS) ClickPipes connector, enabling continuous data ingestion regardless of file naming or arrival sequence. Standard GCS ClickPipes rely on polling every 30 seconds and require files to arrive in strict lexicographical order, which causes out-of-order data and backfills to be ignored. Unordered mode replaces polling by consuming OBJECT\_FINALIZE notifications from Google Cloud Pub/Sub whenever objects land in a bucket. State tracking is managed through a metadata store, and the connector automatically reprocesses batches upon failure while maintaining exactly-once insertion guarantees. Setup requires configuring a Pub/Sub topic and bucket notifications via tools like gcloud, followed by configuring the ClickPipe via the console or Terraform provider.


### [Securing our codebase with autonomous agents](https://yomu.fyi/post/securing-our-codebase-with-autonomous-agents.md)
- Company: [Cursor](https://yomu.fyi/company/cursor.md)
- Author: Travis McPeak
- Published: Mar 16, 2026
- Type: Announcement

After a fivefold increase in pull-request velocity over nine months, the team found static analysis and rigid code ownership insufficient for security at scale. It built a fleet of Cursor Automations powered by cloud agents, webhook and GitHub integrations, and a security MCP deployed as a just-in-time serverless Lambda function. The MCP provides persistent data, semantic deduplication using Gemini Flash 2.5, and consistently formatted reporting and follow-up actions, while Terraform standardizes review and deployment. Four templates cover PR review, segmented vulnerability hunting, dependency patching with reachability analysis and tests, and daily invariant-drift monitoring. Agentic Security Review has run on thousands of PRs in two months and prevented hundreds of issues from reaching production; the team says more automations are planned.


### [Transformation in action: What it takes to automate 81% of your customer service while improving CX](https://yomu.fyi/post/transformation-in-action-what-it-takes-to-automate-81-of-your-customer.md)
- Company: [Intercom](https://yomu.fyi/company/intercom.md)
- Author: Declan Ivory
- Published: Mar 13, 2026
- Type: Problem & solution

Intercom describes how its Customer Support team adopted an AI-first strategy after a 2022 review exposed rising demand, weak service-level commitments, and plans to broaden support access. The team began with high-volume informational queries, then expanded Fin through deeper documentation, internal procedures, customer-action capabilities, knowledge management, conversation design, skills-based routing, and organizational changes including a dedicated AI Support team. Fin now resolves over 81% of support volume and helped absorb a 300%+ demand increase without proportional headcount growth, while over 90% of customers receive improved first response performance, 24/7 coverage, and outbound phone support. The remaining human work has shifted toward consultative engagements, while Insights, Guidance, Tasks, and Procedures support ongoing optimization. Next, the team plans to use a four-stage Fin Flywheel—Train, Test, Deploy, and Analyze—and continue developing its Agent Blueprint.


### [AI doesn’t always generate perfect ClickHouse schemas (yet)](https://yomu.fyi/post/ai-doesn-t-always-generate-perfect-clickhouse-schemas-yet.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Al Brown
- Published: Mar 13, 2026
- Type: Explainer

LLMs can generate syntactically valid ClickHouse schemas whose advanced choices are poorly matched to a workload. The post examines recurring pitfalls including query-driven partitioning, per-column codecs, premature projections, materialized-view sprawl, JSON configurations, and forced deduplication with OPTIMIZE TABLE ... FINAL. It explains that partitions primarily support data management, incremental materialized views add work to every insert, and ReplacingMergeTree deduplicates during background merges rather than at insertion. The recommended method is to begin with a simple MergeTree or ReplacingMergeTree, a well-chosen ORDER BY, default compression, and no optional structures, then measure production behavior before adding complexity or seeking human expertise for high-scale, interconnected designs.


### [From resolutions to outcomes: Evolving how Fin delivers value](https://yomu.fyi/post/from-resolutions-to-outcomes-evolving-how-fin-delivers-value.md)
- Company: [Intercom](https://yomu.fyi/company/intercom.md)
- Author: Darragh Curran
- Published: Mar 12, 2026
- Type: Announcement

Fin is changing its pricing metric from “resolutions” to “outcomes” as its capabilities extend beyond fully autonomous support conversations. Historically, a resolution meant Fin solved an issue end to end without human involvement; the model gave teams a clear ROI comparison, and Fin’s average customer resolution rate now stands at 76% across more than 7,000 teams. Customers increasingly deploy Fin on complex, multi-step work involving external systems, policy constraints, and human oversight, where Fin may gather context and act before handing off for confirmation. An outcome counts when Fin successfully completes the action it was configured to perform, including a full resolution or a Procedure. The change keeps pricing simple while recognizing collaborative workflows, supporting compliance, and providing a metric that can scale as Fin takes on more roles beyond service.


### [5 ways to parse Dates and DateTimes in ClickHouse](https://yomu.fyi/post/5-ways-to-parse-dates-and-datetimes-in-clickhouse.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Mark Needham
- Published: Mar 12, 2026
- Type: Tutorial

ClickHouse provides several ways to convert dates and datetimes from Unix timestamps, packed numeric values, known-format strings, mixed inputs, and file data. The examples use fromUnixTimestamp and its millisecond and microsecond variants, YYYYMMDDToDate and YYYYMMDDhhmmssToDateTime, plus parseDateTime or parseDateTimeInJodaSyntax when the input format is known. For unknown or mixed formats, parseDateTimeBestEffort and its DateTime64 counterpart accept inputs such as ISO 8601 strings, dates, and timestamp strings; invalid values otherwise raise an exception, while OrNull returns NULL and OrZero returns January 1, 1970 at midnight. The post also shows direct ::DateTime casting, controlled by cast\_string\_to\_date\_time\_mode: basic supports standard forms, whereas best\_effort broadens parsing but still throws for completely invalid dates. An inline or session setting lets the same casting approach parse rows from a CSV file.


### [From Fragmented Logs to Full-Stack Visibility with SolarWinds Papertrail](https://yomu.fyi/post/from-fragmented-logs-to-full-stack-visibility-with-solarwinds-papertra.md)
- Company: [Heroku](https://yomu.fyi/company/heroku.md)
- Author: Rachel Revoy
- Published: Mar 12, 2026
- Type: Announcement

Heroku applications combine dynos, databases, third-party APIs, and user interactions, making fragmented logs, metrics, and dashboards costly to correlate during production incidents. The expanded SolarWinds Papertrail add-on addresses this “swivel-chair” workflow by unifying real-time log management, metrics dashboards, and alerting in one interface. It supports live tailing and contextual search through the web, CLI, or API, while ingesting native Heroku metrics for dyno CPU, memory, load averages, application throughput, response times, error rates, and database performance. Custom thresholds based on minimum, maximum, average, or summary values can filter transient-noise alerts, with notifications sent to Slack, PagerDuty, and Microsoft Teams. The post’s incident example correlates a response-time spike with R14 “Out of Memory” errors from Dyno worker.1, identifying an oversized batch job; Librato and AppOptics were deprecated at the end of January 2026.


### [Observability and OpenTelemetry: Introducing MongoDB Atlas Log Integration](https://yomu.fyi/post/observability-and-opentelemetry-introducing-mongodb-atlas-log-integrat.md)
- Company: [MongoDB](https://yomu.fyi/company/mongodb.md)
- Author: Sabina Bates
- Published: Mar 12, 2026
- Type: Announcement

During database latency spikes or unexpected outages, operations teams frequently waste valuable time downloading compressed archives and manually inspecting log files to isolate root causes. MongoDB Atlas resolves this operational friction by introducing a direct log integration feature for dedicated M10+ clusters. The system routes mongod, mongos, and audit logs directly to third-party tools such as Datadog, Splunk, Amazon S3, Google Cloud Storage, and Azure Blob Storage. Furthermore, built-in support for the OpenTelemetry standard enables teams to pipe database logs into any OTel-compatible logging backend. This setup eliminates operational silos, streamlines security auditing, and correlates database query diagnostics with application traces and infrastructure metrics in real time.


### [What's new in ClickStack. February '26.](https://yomu.fyi/post/what-s-new-in-clickstack-february-26.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: The ClickStack Team
- Published: Mar 12, 2026
- Type: Announcement

ClickStack introduced several platform updates and query performance enhancements across both open-source and Cloud environments for February 2026. ClickHouse Cloud OpenAPI now exposes ClickStack resources including dashboards, alerts, sources, and webhooks to support automated configuration-as-code workflows. In addition, ClickHouse 26.2 embeds the ClickStack UI directly into the database binary with a 4.1 MB footprint for local exploration and diagnostics. Visualizations now support native SQL mode for tables, allowing users to join disparate datasets like logs and traces across injected dynamic time variables. Finally, query execution speeds improve through streaming skip indexes, index pruning on disjunctions, and an expanded 100,000-row lazy materialization limit alongside new threshold alerting on number charts.


### [Enabling R8 optimization at scale with AI-assisted debugging](https://yomu.fyi/post/enabling-r8-optimization-at-scale-with-ai-assisted-debugging.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Nguyen Van Minh
- Published: Mar 12, 2026
- Type: Problem & solution

Grab enabled R8 optimization for its Android app after repeated attempts were blocked by debugging difficulty, organization-wide testing demands, and slow build feedback. The effort targeted a roughly 10-year-old codebase exceeding 9 million lines and combined AI-assisted debugging, a pragmatic testing strategy, and infrastructure investment. Its QA build was configured to match production R8 processing. Beyond existing shrinking, optimization added techniques such as method inlining, class merging, constant folding, dead-code elimination, and devirtualization. After about 10 weeks with one engineer coordinating across teams, the project reported around 25% lower ANR rates, a 16% reduction in reference-device download size, and nearly 27% faster startup; the startup result combined R8 with corrected Baseline and Startup Profiles.


### [Shopify’s journey to faster breadth-first GraphQL execution](https://yomu.fyi/post/shopify-s-journey-to-faster-breadth-first-graphql-execution.md)
- Company: [Shopify](https://yomu.fyi/company/shopify.md)
- Author: 2023-10-18
- Published: Mar 12, 2026
- Type: Explainer

Shopify describes how high-cardinality, deeply nested GraphQL queries exposed execution overhead beyond data loading, with field resolvers assembling responses consuming much of request time. Its GraphQL Cardinal engine replaces conventional depth-first traversal with breadth-first execution, resolving each selected field once across an aggregated set of objects and reducing repeated field-level work. For large list queries, Shopify reports up to 15x faster execution, 90% less memory use, and reductions of many seconds in P50 latency. The migration included a tuned interpreter, cheaper tracer behavior, shadow verification, benchmarks, and ongoing conversion of tens of thousands of legacy resolvers, while future work includes asynchronous patterns and lower-level C bindings.


### [How we compare model quality in Cursor](https://yomu.fyi/post/how-we-compare-model-quality-in-cursor.md)
- Company: [Cursor](https://yomu.fyi/company/cursor.md)
- Author: Naman Jain
- Published: Mar 11, 2026
- Type: Benchmark

Cursor describes a hybrid online-offline process for comparing coding-agent quality as developer tasks become longer, more complex, and multi-file. Its offline CursorBench suite uses tasks derived from real Cursor sessions, paired with ground-truth solutions through Cursor Blame, and evaluates correctness alongside code quality, efficiency, and interaction behavior; the post focuses on correctness. Tasks are intentionally short and underspecified, span larger codebases and multi-workspace environments, and are scored with agentic graders, while correctness is plotted against median completion tokens to show the compute/latency trade-off. Controlled live-traffic experiments supplement the suite, catching regressions that offline grading misses, and the post reports that CursorBench separates frontier models more clearly and its rankings track developer experience more closely than public benchmarks.


### [Announcing chDB 4: write Pandas, run ClickHouse, now on Hex](https://yomu.fyi/post/announcing-chdb-4-write-pandas-run-clickhouse-now-on-hex.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Author: Ryadh Dahimene, Auxten Wang
- Published: Mar 11, 2026
- Type: Announcement

chDB 4 introduces Data Store, a native component that allows users to write Pandas-like code that executes on the ClickHouse OLAP SQL engine. Standard Pandas workflows suffer from eager execution, excessive memory materialization, and lack of automatic query optimizations. To solve this, Data Store implements a four-layer architecture that lazily records operations as a chain of LazyOp objects and compiles them into optimized ClickHouse SQL plans upon evaluation triggers. When encountering operations unsupported by ClickHouse, the QueryPlanner splits the pipeline into segments, routing tasks between ClickHouse and Pandas while using Python's memoryview to minimize data copying overhead. This release also integrates natively into Hex notebooks.


[Newer posts](https://yomu.fyi/posts/page/152.md) · [Older posts](https://yomu.fyi/posts/page/154.md)