The Week Somebody Fixed the Pipes (While Everyone Watched the Fireworks)
Week 26, 2026 — Sal Ferro covers AI advancements, cloud services, and data analytics, amidst the doublespeak of corporate optimization.
In the voice of Orwell - Plain English as moral act, clear-eyed against doublespeak, politics of language
The Week Somebody Fixed the Pipes (While Everyone Watched the Fireworks)
Week 26, 2026 — by Sal Ferro
Every week, tech throws a fireworks show. This week was a good one. Google lit up the sky with a new Gemini that "thinks deeply" and reset half the leaderboards before lunch. SpaceX bought the company behind everyone's favorite code editor for the price of a small country. OpenAI quietly pocketed the little Python toolmaker that half of you have running in your build pipeline right now.
Gorgeous. Loud. Worth watching.
But I spent my week in the basement, where the real week happened — and I want to tell you about it, because the basement is where the building actually stands up.
Start with the bug nobody wants to find. The folks at Redpanda went public with something genuinely unnerving: Apache Kafka's log compaction can quietly corrupt your data. Not crash. Not alarm. Quietly. That's the worst kind — the slow leak behind the drywall you don't notice until the wall's already soft. They reproduced it, explained it, and shipped a fix. No keynote, no confetti. Just somebody finding the rot and patching it before it reached your foundation. That's the whole job, right there.
Then there's the least sexy software on Earth, made lean. A team rewrote a Postgres backup tool in Rust — "WAL-RUS," because engineers cannot resist a pun — and got it using 70% less virtual memory than the thing it replaces. Backups! Nobody dreams about backups. Backups are the sump pump in the corner you forget exists until the night it saves your house. Somebody spent real weeks making that pump cheaper to run. Bless them. While we're down here: ClickHouse shipped a new fiber runtime called Silk, and Weaviate cut a clean 1.38 release. Joists and floorboards. The stuff you stand on.
Now the part that actually moved me. Microsoft Research put out something called Talos — a system for rare-disease diagnosis. The number that stopped me: it recovered 90% of the in-scope diagnoses while surfacing just 1.3 candidate variants per case. Translate that out of the spec sheet. A human geneticist drowning in possibilities, handed a short list of barely more than one. Families who wait years for an answer, getting one sooner. That's not a feature. That's a kid getting a name for the thing that's wrong with them. And here's what gets me: it went out as a research blog post, not a billboard. The week's best story, buried under its own modesty.
The chore-doers. This was a strong week for AI quietly doing the paperwork instead of pretending to be magic. Visa wired up a conversational analytics agent and turned multi-day reports into sub-second answers — and, more to the point, handed 8 to 10 hours a week back to each person who used to babysit those reports. Huntington built a pipeline to find and redact personal data across 400 million-plus documents. AWS showed voice agents that just… book the appointment. None of it is a robot uprising. All of it is the dishwasher of office work finally getting loaded. I'll take a dishwasher over a fireworks show any day I actually have to cook.
The foundations and the power plant. The semiannual supercomputer list landed, and NVIDIA is now under 81% of the world's 500 fastest machines, with 26 of them running its Grace chips. OpenAI and Broadcom showed off a custom inference chip with the excellent name "Jalapeño." OpenAI also teased its next model, GPT-5.6 "Sol." And Meta — credit where it's due — engineered a battery narrow enough to live inside the arm of a pair of glasses while still powering a camera, speakers, and a display. That's not glamour. That's somebody solving the physics of fitting a power plant into a temple tip. Respect.
So here's the thing I keep noticing. The loudest companies this week didn't spend their money on fireworks. They spent it on pipe. SpaceX bought a code editor. OpenAI bought a Python toolchain. Even the giants, with all that sky to light up, looked at the week and decided the smart buy was the plumbing.
They're right. The fireworks are beautiful and they're gone by morning. The pipes are still there when you wake up, still doing the one thing nobody will ever thank them for: keeping the water running.
This was a basement week. And the basement, between you and me, is exactly where I like it.
— Sal
Sources
25 events · 25 source articles
- Build interactive PDF text extraction from Amazon S3In this post, you’ll build a server that extracts text from PDF files in Amazon S3 in real time. This protocol-based approach provides programmatic document access. You’ll walk through the...🏢 Companies:AWS AI/ML
- How Cara pioneers domain-specific AI for enterprise insurance brokerages with AWSIn this post, we explore how Cara, built in cooperation with AWS, addresses these challenges. We walk through the technical design decisions and the AWS services that support the solution. We also...🏢 Companies:AWS AI/ML
- Production-grade AI agents for financial compliance: Lessons from StripeIn this post, you learn how Stripe built a production-grade AI agent system for financial compliance. We cover the technical architecture of Stripe’s ReAct agent framework and the infrastructure...🏢 Companies:AWS AI/ML
- Previewing GPT-5.6 Sol: a next-generation modelOpenAI previews GPT-5.6 Sol, a next-generation model with stronger capabilities in coding, science, and cybersecurity, paired with its most advanced safety stack.🏢 Companies:OpenAI
- From test bench to lakehouse: how AVL modernizes measurement data analytics with Impulse1. Introduction - Impulse: time-series analytics for measurement dataA single automotive...🏢 Companies:Databricks
- Optimize model training on Amazon SageMaker AI with NVIDIA BlackwellThis post shows you how to configure training jobs on Amazon SageMaker AI to get the most out of Blackwell’s architecture on AWS. You learn how to select batch sizes and sequence lengths that take...🏢 Companies:AWS AI/ML
- Understanding the brain with AI-driven explanations and experimentsResearchers introduce generative causal testing, which translates black box models into clear hypotheses and verifies them in the scanner, revealing what specific brain regions respond to in...🏢 Companies:Microsoft Research
- Unlocking smaller, smarter RF front-ends with SLATE™ advanced packaging technology on 9SWAs consumer demand accelerates for next-generation smart mobile devices and RF and battery performance requirements continue to rise, every square millimeter of board space matters. Front-end modules...🏢 Companies:GlobalFoundries
- AI SDK 7AI SDK, with over 16 million weekly downloads, is the TypeScript SDK for building AI applications, features, frameworks, and agents across any model provider. It's the same layer eve, Vercel's...🏢 Companies:Vercel AI
- Weaviate 1.38 ReleaseThis release brings the HFresh disk-based vector index and the built-in MCP Server to general availability, rebuilds cluster-wide async replication to run from a single scheduler (on by default), and...🏢 Companies:Weaviate
- Announcing Silk: a silky smooth fiber runtime for ClickHouseSilk is a new open-source C++ fiber runtime built for ClickHouse, combining a NUMA-aware work-stealing scheduler, io_uring I/O, and zero heap allocation in the steady state to deliver...🏢 Companies:ClickHouse
- ClickHouse is now a native connection in Notion Custom AgentsClickHouse is now a native connection in Notion Custom Agents. Connect with OAuth and let agents run read-only queries on your data without leaving Notion.🏢 Companies:ClickHouse
- Why we rewrote WAL-G for Postgres backups in Rust: Meet WAL-RUSHow we built WAL-RUS, an open-source Rust-based Postgres backup tool that reduces virtual memory usage by over 70% compared to WAL-G while maintaining full compatibility.🏢 Companies:ClickHouse
- Kafka's log compaction corrupts data. Here's how we fixed itThere's a problem with Apache Kafka's log compaction. Here's what we found, how to reproduce it, and how we solved it in Redpanda.🏢 Companies:Redpanda
- Huntington Bank: Redacting sensitive data from 400M+ documents with AWSIn this post, we walk through how Huntington built a scalable AWS solution to detect and redact Personally Identifiable Information (PII) and Payment Card Industry (PCI) data from over 400 million...🏢 Companies:AWS AI/ML
- Build a healthcare appointment agent with Amazon Nova 2 SonicIn this post, you will learn how to build a voice agent that handles appointment reminder conversations using Amazon Nova 2 Sonic and Amazon Bedrock AgentCore. The agent authenticates patients by...🏢 Companies:AWS AI/ML
- How Loka Built a Natural, Low-Latency Voice Agent with Amazon Nova 2 SonicIn this post, we demonstrate the architecture and approach Loka used to solve a common frustration: robotic, slow voice assistants that cause customers to hang up, damaging brand reputation and...🏢 Companies:AWS AI/ML
- Talos: Scaling rare disease diagnosis with automated, iterative genomic reanalysisTalos was built to help resolve a major bottleneck in genomic medicine: human review time. The open-source system recovered 90% of in-scope diagnoses while surfacing just 1.3 candidate variants per...🏢 Companies:Microsoft Research
- OpenAI and Broadcom unveil LLM-optimized inference chipOpenAI and Broadcom introduce Jalapeño, a custom AI chip built for LLM inference to improve performance, efficiency, and scale across AI systems.🏢 Companies:OpenAI
- Announcing the Managed ClickStack MCP ServerAnnouncing the Managed ClickStack MCP Server: purpose-built observability tools, full-fidelity telemetry, and dedicated compute for AI-driven investigations at scale.🏢 Companies:ClickHouse
- How Visa went from multi-day reporting to conversational analytics agents with ClickHouse Cloud and LibreChatVisa paired LibreChat with ClickHouse Cloud to build conversational BI agents on Authorize.net payments data, cutting multi-day reports to sub-second queries and reclaiming 8-10 hours per user each...🏢 Companies:ClickHouse
- Genesis Workbench: A blueprint for industry AI in life sciences, powered by Databricks and NVIDIABringing GPU-accelerated drug discovery to your dataLife sciences leaders need domain-specific,...🏢 Companies:Databricks
- How Meta Engineered Ultra-Narrow Batteries for AI GlassesSmart glasses like the Ray-Ban Meta and Oakley Meta Vanguards need to pack enough energy to power features like cameras, speakers, AI workloads, and even a display. But it all has to fit into the...🏢 Companies:Meta AI
- NVIDIA Powers Over 400 of the World’s 500 Fastest SupercomputersNews Highlights: NVIDIA technology runs 81% of the TOP500 and 90% of the systems new to the list. 26 systems on the TOP500 adopted the NVIDIA Grace CPU, up eight from the previous list. The top eight...🏢 Companies:NVIDIA
- NVIDIA Brings Trusted, 24/7 AI Agents to Telecom OperationsTelecom operators have seen remarkable returns from using generative AI to automate network management, customer care and back-office operations. Most of that impact has been task‑based: automation...🏢 Companies:NVIDIA