PM Intelligence Feed

The Thin Orchestrator: Making AI Agents More Reliable with Less Router Work

Sometimes, giving an AI agent all of the knowledge it needs is the wrong approach. At Meta, we’ve seen notable improvements in efficiency and accuracy by keeping the orchestration layer of our data pipeline agents deliberately thin: It routes work and enforces reliability checks, while specialized agents do the domain reasoning. We’re sharing our o…

Oct 19, 2026

Inside Hyperdrive: Solving a Rare MySQL Caching Bug

While adding MySQL support to Hyperdrive, we found a subtle protocol bug when caching query results. Here is how we tracked down a 0.001% rate error and made Hyperdrive MySQL generally available.

Oct 14, 2026

Turning AI Inference into Better Ad Recommendations

At Meta, we’ve built a system that enhances ads recommendations by leveraging idle inference capacity in our fleet. Two elasticity levers in the system allow ads recommendation work to expand across opportunistic capacity, without compromising baseline performance. This post describes how we designed and now operate the system at Meta’s scale and h…

Oct 12, 2026

Part 5: Operating an LLM System: Observability, Cost, Routing, Platform Underneath

Your service can be 100% up and still quietly approving the wrong things, burning its budget, or failing over into untested quality. Level 5 is the infrastructure that lets you see your decisions, bound your spend, route and fail over between models, kill bad behavior in seconds — and the platform that makes all of it possible.

Oct 08, 2026

Agents Refactor 300K Lines in Three Weeks, Practitioners Ask What It Proves

CodeScene has published a case study in which coding agents refactored 300,000 lines of C over three weeks for roughly $4,000 in tokens, verified by a frame-by-frame replay harness. The agents built a playbook of codebase-specific recipes along the way. Practitioners have questioned the scope, the metric, and how much of the result depends on the h…

Sep 30, 2026

Is Your Human-in-the-Loop Actually Slowing You Down?

Sep 28, 2026

Four months of VoidZero at Cloudflare: Making the open-source JavaScript toolchain faster for all

Since joining Cloudflare, VoidZero has delivered more than 80 releases that drastically speed up JavaScript compilation, linting, and testing. From a 10x faster React compiler to Vite+ 1.0, here’s how we are building faster tools for developers and AI agents.

Sep 28, 2026

Presentation: Context Engineering at LinkedIn: Building an Organizational Context Layer for AI_agents_with_MCP

Ajay Prakash discusses how LinkedIn overcomes AI agent limitations in large codebases. He explains Contextual Agent Playbooks and Tools - built on Model Context Protocol (MCP) - which serves procedural memory, code search, and runbooks directly to coding agents. Prakash shares architectural details and operational guardrails that deliver a 20% prod…

Sep 19, 2026

Implementing Chaos Engineering in Financial Payment Systems: Lessons from Enterprise Deployments

Standard chaos engineering assumes experiments stop cleanly, blast radius is knowable in advance, and production is fair game. Payment systems violate all three. Salim Adedeji describes ECS-specific failure modes from enterprise deployments: a 60-second DNS TTL that produced 93-second failover, retry logic amplifying database load 2.4x, and AZ reba…

Sep 08, 2026

How we saved 100 terabytes of memory by optimizing 1.1.1.1's DNS cache

Five Rust-level memory optimizations to the DNS cache layout of Big Pineapple cut per-entry memory by 56%, freeing approximately 100 TB of memory across Cloudflare's fleet.

Aug 27, 2026

Alt text passes automated checks but isn’t necessarily good

We built a plugin for the GitHub Accessibility Scanner to make sure your alt text is actually accessible. Here's how it works. The post Your alt text passes automated checks. That doesn’t mean it’s any good. appeared first on The GitHub Blog .

Aug 24, 2026

Rightsizing Platform Engineering: Building the Platform Your Organization Needs

Shift-left and DevOps have impacted how we flow changes from inception to production, but at the cost of increased cognitive load and duplication of effort across testing, security, and maintenance. This article explores the real-world challenges of rightsizing developer platforms and finding a cultural match for engineering teams who use them to r…

Aug 24, 2026

How We Raised Mobile End-to-End Test Stability to 98%

We rebuilt our mobile end-to-end testing framework with a strict API and computer vision, raising test stability drastically.

Aug 12, 2026

Sidekick's continual learning loop

How we compress production failures into model weights every day, beat frontier-model quality, and cut serving costs 96%.

Aug 05, 2026

How we're rethinking work at Cloudflare with Cloudflare OS

We built Cloudflare OS to equip our teams to safely rethink how they get work done with AI. The platform brings together the best of our technologies, from our Compute primitives to our Zero Trust suite. This post walks through our journey to give our users the best AI tools available.

Aug 05, 2026

Explorers, exploiters, and the myth of the 100x engineer

The “find the special ones and promote their traits” approach isn’t the best or only way to drive AI adoption and productivity on an engineering team. ​​​​‌‍​‍​‍‌‍‌​‍‌‍‍‌‌‍‌‌‍‍‌‌‍‍​‍​‍​‍‍​‍​‍‌​‌‍​‌‌‍‍‌‍‍‌‌‌​‌‍‌​‍‍‌‍‍‌‌‍​‍​‍​‍​​‍​‍‌‍‍​‌​‍‌‍‌‌‌‍‌‍​‍​‍​‍‍​‍​‍‌‍‍​‌‌​‌‌​‌​​‌​​‍‍​‍​‍‌‍​‌‍‌‌​​‍‍‌​‌‌​‌‍​‌‌‍​‌‍‍‌‍‌‌…

Aug 05, 2026

GEM Training: How Meta Doubled Ads Foundation Model Efficiency

Meta’s Generative Ads Recommendation Model (GEM), the foundation model behind ads recommendations across Instagram and Facebook, now trains at LLM scale on several thousand of the latest-generation GPUs. This post goes into the details on how we achieved: doubling end-to-end (E2E) training efficiency to 20–25% Model FLOPs Utilization (MFU) while sc…

Aug 03, 2026

BGP Origin Attribute Manipulation and Its Impact on the Internet

When BGP attributes are manipulated, PMs typically focus on immediate performance metrics — but this shows the long-term implications of trust and reliability in routing decisions can be overlooked. Understanding how subtle changes in network attributes can influence traffic behavior is crucial for making informed trade-offs between flexibility and stability in product design.

Jul 24, 2026

Agent platform (Part 1): How we help Grab build and run AI agents at scale

When product teams prioritize immediate feature development, PMs typically overlook the foundational infrastructure needed for scalability — but this shows that investing in a robust framework from the start can lead to exponential growth and adaptability in the long run. By focusing on solving recurring problems through a flexible architecture, teams can avoid the pitfalls of short-term fixes that hinder future innovation.

Jul 24, 2026

GitHub Increased Instant Navigation from 4% to 22% by Rethinking Client-Side Architecture

When prioritizing immediate responsiveness, PMs typically focus solely on backend performance — but this shows that client-side architecture and caching strategies can significantly enhance user experience without compromising data integrity. This highlights the importance of considering how data retrieval methods impact perceived performance, particularly in applications with frequent user interactions.

Jul 22, 2026

From weeks to a day: speeding up LLM evaluation

When engineering teams focus solely on individual components, PMs typically assume that a well-functioning part means the whole system is robust — but this shows that the integration layer is often the most critical and overlooked aspect of product performance. Prioritizing end-to-end validation over isolated metrics can lead to more meaningful insights and ultimately a more reliable product, challenging the common belief that component-level success guarantees overall effectiveness.

Jul 14, 2026

Better tools made Copilot code review worse; how we improved it

When PMs prioritize tool upgrades for efficiency, they typically underestimate the importance of contextual alignment in workflows — but this shows that even the best tools can falter if they don't match the user's actual processes. The real challenge lies not in the tools themselves, but in ensuring that the instructions and workflows are tailored to how users engage with those tools, highlighting the need for a holistic approach to product design.

Jul 10, 2026

The Data Canary: How Netflix Validates Catalog Metadata

When data integrity is compromised, PMs typically focus on code-level issues — but this shows that product resilience must encompass data validation as rigorously as it does for code deployments. Ignoring the complexities of high-velocity data pipelines can lead to significant user experience failures, emphasizing the need for PMs to integrate data quality checks into their overall product strategy.

Jun 19, 2026

AlloyDB Ships Proxy Models for Local Database Inference

When optimizing for performance, PMs typically prioritize throughput and cost reduction — but this shows the importance of rethinking the architecture of interactions between systems. By leveraging local proxy models, teams can significantly enhance efficiency while minimizing dependency on external resources, a tradeoff that often goes overlooked in favor of immediate performance metrics.

Jul 09, 2026

Adopting AV1 for Real-Time Communication at Scale

When PMs prioritize bandwidth efficiency, they typically overlook the nuanced impact of codec choice on user experience — but this shows that selecting the right codec can significantly enhance video quality, especially in low-bandwidth scenarios. Additionally, the focus on advanced encoding techniques reveals that optimizing for specific content types, like screen sharing, is often neglected in favor of broader performance metrics.

Jun 22, 2026

Coding Is No Longer the Constraint: Scaling Developer Experience at Spotify

When coding ceases to be the primary bottleneck, PMs typically focus on scaling team size and resources — but this shows that investing in developer experience and automation can yield exponential productivity gains without simply adding more engineers. This shift highlights a common oversight where PMs underestimate the transformative impact of tools and processes that enhance efficiency, often favoring traditional scaling methods instead.

Jun 03, 2026

When history fails you, borrow from geography

When faced with unprecedented market shocks, PMs typically rely on historical data to inform their strategies — but this shows that leveraging geographic insights can provide critical, timely forecasts when traditional models fail. This highlights the importance of being adaptable and seeking alternative data sources, as rigid adherence to past patterns can lead to significant miscalculations in rapidly changing environments.

Jun 02, 2026

Background Coding Agents: Supercharging Downstream Consumer Dataset Migrations Part 4

When automation tools are introduced to streamline complex processes, PMs typically focus on immediate efficiency gains — but this shows that understanding the underlying context and dependencies is crucial for successful implementation. Neglecting to consider how different systems interact can lead to oversights that undermine the benefits of automation, resulting in more work in the long run.

Apr 22, 2026

Lessons Learned Running Presto at Meta Scale

When scaling a product rapidly, PMs typically prioritize feature delivery — but this shows that operational reliability and deployment automation are equally critical to maintaining user satisfaction. Overlooking the balance between feature rollout and infrastructure stability can lead to degraded performance and frustrated users, ultimately undermining the product's success.

Jul 16, 2023

PHP Patches Two Critical RCE Flaws That Were Already Under Attack

PHP released 8.5.12, 8.4.27, 8.3.36, and 8.2.35 to fix two critical vulnerabilities allowing unauthenticated remote code execution. Amazon recorded an exploitation attempt four days before the patches shipped, and operators spent the day working out whether they ran PHP, where to get the package, an…

Oct 25, 2026

PyTorch Conference 2026: Google and Meta Open-Source TorchTPU for Native PyTorch on TPUs

Google and Meta open-sourced TorchTPU, a native PyTorch backend for Google's TPUs, at PyTorch Conference North America 2026, and Meta said it plans to move larger LLM workloads onto TPUs with it. AWS released TorchNeuron for Trainium the same week, while developers questioned TPU availability and wh…

Oct 25, 2026

Google ML Drift Accelerates On-Device AI Inference Across Diverse GPUs

Google has released ML Drift, an open-source, cross-platform GPU compute engine designed for on-device AI and machine learning inference. Licensed under Apache 2.0, it abstracts the complexities of different GPU hardware and low-level APIs, supporting OpenGL ES, OpenCL, Metal, and WebGPU. By Sergio…

Oct 25, 2026

PingCAP Breach Exposes Customer Records and Source Code

PingCAP recently confirmed unauthorized access to its internal systems and the theft of source code and customer-related data. The attacker, cyber extortion group FulcrumSec, claimed to have stolen 10.6 TB of data, including TiDB Cloud control-plane records, customer SQL statements, API keys, connec…

Oct 25, 2026

Google Optimizes Agent Development Kit's 200K-line Python Codebase Overnight with Gemini 4 Argon

Google explained how an 11-hour unattended run of Gemini 4 Argon produced two accepted optimizations for Google Agent Development Kit's codebase, reducing in-memory session-preparation overhead by about 40% in long sessions and cold package import time by 21%. By Sergio De Simone

Oct 24, 2026

Beyond API Compatibility: Building a Tool to Measure the Blast Radius of Library Changes

By Roberto Perez Alcolea , Java Platform Team Quasar connects a proposed API change with observed references in consumer repositories, giving the producer evidence for a release decision At Netflix, shared Java libraries support applications developed across a large, independently evolving repositor…

Oct 23, 2026

Origin Range Requests: Serving large files without range-request tuning

Naive byte-range caching can waste gigabytes of origin bandwidth on small requests. We rebuilt our cache architecture in Pingora to coordinate partial fetches automatically, cutting R2 p99 latency in half.

Oct 23, 2026

Bringing post-quantum cryptography to geographically fenced key management

Cloudflare's new post-quantum GeoKey Manager protects geographically fenced decryption keys from harvest now, decrypt later attacks. Learn how we redesigned our cryptography stack to support strict data sovereignty in a post-quantum world.

Oct 23, 2026

Rumors of the SaaS demise are greatly exaggerated

Recorded on the floor of LDX3, Ryan chats with Nivia Henry, an engineering leader most recently at Netflix, about the reality of the supposed SaaS apocalypse, the complex calculus of the modern build-versus-buy decision in the era of AI, and the critical importance of understanding system architectu…

Oct 23, 2026

Single-tenant Titus: VM-level isolation on a multi-tenant container platform

Samartha Chandrashekar & Nick Parker You click play on a Friday night, and in the second that follows, dozens of services answer in milliseconds. Most of them run on Titus, our internal container platform. Titus is, in essence, a Kubernetes-based container hosting platform: we keep both the cont…

Oct 22, 2026

AWS Continuum for code vulnerabilities is now generally available

AWS Continuum for code vulnerabilities is a new security service that uses AI to automatically find, validate, and prioritize code vulnerabilities in your environment, then delivers ready-to-review fixes as pull requests replacing what previously took hours of manual investigation and days of cross-…

Oct 22, 2026

The Regression Tests Nobody Writes: Capturing How Software Actually Behaves

Oct 22, 2026

The Pulse: Time to rethink software security because of AI?

A month’s worth of security breaches in a single day – will things get worse before they improve? Also: Anthropic accused of version control “rip off”, and more

Oct 22, 2026

Localizing compliance at scale: achieving South Korea’s ISMS certification

Cloudflare has achieved South Korea's ISMS certification following an audit across 80 controls. Learn how we localized user consent flows, adapted social sign-up paths at the edge, and validated our global security program.

Oct 22, 2026

We gave AI agents a mobile dev environment

We built Remote Dev so AI agents can build and test apps in parallel, without maxing out developers’ laptops.

Oct 22, 2026

How to castle in chess

Protect your king—here’s how!

Oct 22, 2026

The revised rules of engineering management with Will Larson

Will Larson, CTO of Imprint, explores how AI and shifting industry expectations are changing engineering management, career ladders, and the path from engineer to CTO.

Oct 21, 2026

Building Immersive Virtual Economies: An Introduction to KursarScript (KSPL)

Oct 21, 2026

The bridge to agent-scale Git infrastructure

We moved a copy of 99.99999% of GitHub's repositories to Azure. Here's how per-reference quorums will bring over the rest, and how object storage lets reads scale independently. The post The bridge to agent-scale Git infrastructure appeared first on The GitHub Blog .

Oct 21, 2026

Introducing full transaction detection for Cloudflare WAF

Cloudflare's web application firewall can now correlate an HTTP request with its response. Full transaction detection helps every customer find successful exploits, data exposure, and risky misconfigurations that request-only inspection can miss.

Oct 21, 2026

Survey Says: You'll swap the model, but everything you build around it is staying put

Earlier this month, my team at Stacklok surveyed 520 technical leaders to understand whether they were concerned about being locked into frontier labs. The data says they’re already swapping models in and out, but the trap they’re concerned about is the layer on top of the model.

Oct 21, 2026

Cloudflare deploys protections against php2shell vulnerability

Cloudflare has deployed default web application firewall (WAF) protections for newly disclosed PHP vulnerabilities before public disclosure. The rules use request and response signals to stop exploit attempts and prevent exposed data from reaching attackers.

Oct 21, 2026

Fragments: October 21

Gergely Orosz is frustrated at those that entered his explicitly “No AI” essay competition and used AI. He goes on to say: When I start to suspect some part is AI about someone I don’t know: how can I tell how much of it is? And how much time and effort do I want to invest to figure out? I fear most…

Oct 21, 2026

How Shopify turns open models into task experts

Sidekick’s frontier model hands its most token-heavy work to open models that Shopify post-trains and serves itself. Here’s the loop we use to build them, and where NVIDIA’s models, software, and GPUs make each turn faster and cheaper.

Oct 21, 2026

App stability at scale with a layered defense

Introduction Artificial intelligence (AI) coding assistants have changed how fast we can ship. They have not changed how mobile users experience a crash. That gap is the pressure every mobile stability program now operates under. The Grab superapp, which millions of users across Southeast Asia use t…

Oct 21, 2026

What’s the ROI of software engineers’ spending on AI?

AI spending has greatly increased at most companies this year, but has it actually changed shipping velocity and PR output? We analyse data from 27 companies – with some interesting findings

Oct 20, 2026

Introducing Amazon ECS Gateway: A fully-managed ingress solution for ECS services

Try Amazon ECS Gateway, a new account-level resource that simplifies ingress networking for containerized workloads by dynamically provisioning and managing the necessary infrastructure on your behalf, with all resources created directly in your account.

Oct 20, 2026

Cross-store product clustering across millions of stores

How we group product listings across millions of stores using tiered matching models, graph partitioning, and learned cluster merge decisions.

Oct 20, 2026

How to form simple past verbs in English and when to use them

English past tense verbs have more exceptions than rules—and this guide covers both!

Oct 20, 2026

Don't forget about the UI for AI

Recorded at OpenAI Dev Day 2026, Ryan sits down with Andrew Ambrosino, member of the technical staff at OpenAI, for a conversation on OpenAI's strategy for merging ChatGPT and Codex into a unified experience, the advantages of localized computer access in their successful desktop application, a…

Oct 20, 2026