Engineering Culture & Leadership

How engineering organizations actually run — hiring and onboarding, on-call rotations and incident review, developer experience, and the metrics teams use to argue about productivity. Less code than the rest of the site, and more of what companies learned managing the people writing it.

Loading
Building the Agent Host Protocol for sessions that continue across devices
GitHub describes the Agent Host Protocol (AHP), an open standard that keeps Copilot agent sessions continuous across VS Code, the GitHub Copilot app, mobile apps, and cloud sandboxes. The post explains why per-client integrations failed, how a URI-channel, reducer-based state design handles reconnection and concurrent mutations, and what was learned from building two independent hosts in TypeScript and Rust. It also covers how the protocol stays agnostic to agent behavior and where to connect...
How AI Slashed 600-Test-Failure Triage From Days to 90 Minutes
Salesforce engineers describe building AI-Assisted Test Failure Buddy, a Python and FastMCP tool exposing eight MCP tools to Claude Code and Cursor, to triage over 600 test failures caused by retiring the SOAP API login mechanism. The post explains how deterministic code handles API calls, batching, signature matching, and reassignment while AI interprets unmatched stack traces and ambiguous ownership, reducing estimated two to three days of triage to about 90 minutes.
Agents are users, so your software needs evals
The post argues that software used by AI agents needs evals, meaning repeated, graded runs rather than single tests, and lays out an agentic testing pyramid. It walks through building an end-to-end eval with Harbor on Vercel Sandbox and AI Gateway, including choosing between pass@k and pass^k, designing tasks and verifiers, and analyzing traces to find failure causes.
How the Claude Code team thinks about verification
The post explains how the Claude Code team uses end-to-end, property, and differential tests to let coding agents verify their work with evidence. It also describes the practices used to keep verifiers honest, including a commit gate, a reliability checklist tied to diffs, and evaluations that measure review recall against past incidents.
Maintainers don’t need more AI bug reports. They need fixes.
GitHub and Microsoft ran a pilot using Codename MDASH, an AI scanning harness, to find vulnerabilities in six open source projects. An AI triage agent removed duplicates and researchers verified each finding by hand before reports went out privately with proofs of concept and suggested fixes. The post distills lessons for anyone sending AI-generated security reports: verify findings, report privately, and include a fix.
Your coding agent shouldn't grade its own homework
Momentic's engineering team argues that coding agents should not verify their own work, citing an audit where only 9.6% of 1,000 AI-authored unit tests were useful and a comparison showing agents that wrote tests did no better than those that did not. They describe an independent QA agent that tests live applications from the original requirements, which their benchmark found surfaced 3-5 times as many bugs as agents checking their own output.
Cuttlefish on Mac: Running Android’s Virtual Device Framework Natively on Apple Silicon
Meta's Reality Labs ported Android's Cuttlefish virtual device framework to macOS on Apple silicon, consolidating a split Goldfish and Cuttlefish setup into one high-fidelity platform. The post details the technical changes: making QEMU's vhost-user path portable beyond Linux, increasing memory-region tracking slots in the HVF backend, adding a native Cocoa display, and handling JIT entitlements and code signing.
Walk into every customer meeting knowing what you promised
Atlassian describes how to connect Gmail and Google Calendar to its Teamwork Graph so that Rovo can compare customer commitments in email against Jira delivery dates, prepare meeting briefs, and draft follow-ups from Confluence notes. The post is a product walkthrough with example prompts and setup links, and it offers little technical detail on how the integration works.
Detecting credential-driven attacks on Databricks
The post describes how attackers use stolen credentials such as session cookies, personal access tokens, and service-principal secrets to move through a Databricks workspace in six stages, from authentication through persistence and data exfiltration. It gives concrete audit-log events to monitor at each stage, including discovery call bursts, secret harvesting sequences, and persistence via admin group changes. It also outlines response steps and preventive controls such as context-based...
NVIDIA CUDA-X Libraries Give Agents the Tools to Accelerate Scientific Discovery
NVIDIA announces expanded CUDA-X libraries, including cuPhoton, cuPotential, cuSpa, Daviator, DAQIRI, rapids-singlecell and HMMER, for scientific workloads in astrophysics, fusion, materials science, drug discovery and single-cell biology. The post reports speedups for each, such as an Event Horizon Telescope correlation step cut from over a month to eight minutes, and it positions the libraries for use by AI agents.
Recycle your tests: many shades of green
The author describes how trivago repurposed a subset of its existing end-to-end Cucumber/Selenium scenarios, using Gherkin tags, for three new jobs: synthetic production monitoring, per-release browser performance measurement, and stage stability checks after backend changes. The post highlights practical lessons such as avoiding greenwashing with meaningful assertions, keeping tests side-effect free, and recycling selectively to avoid overloading the stage environment.
How Pinterest Improved Analytics Agent Compliance with Loop Engineering
Pinterest Engineering describes how it improved compliance of its Analytics Agent, which must decide which procedural rules apply to each question. The team built paired required and skip test suites, moved from persistent sessions and memoryless Ralph Wiggum loops to fresh contexts with structured handoffs, and used a repair/break gate, acceptance critic, and regression guard to accept only safe edits. The post also notes that case-level flip rates of about 10% can hide real variation behind...
Test your applications against AWS locally with AWS Emulator
AWS Emulator is a free, open-source local emulator written in Rust that lets developers run AWS CLI, SDK, and coding-agent workloads against a local HTTPS endpoint. The post explains its SigV4 enforcement, embedded mode for Rust test suites, and a fidelity-testing method that checks responses against real AWS. It also notes limits such as the lack of IAM policy evaluation.
Beyond Test Cards: A Fake Payment Provider for Testing at Scale
Lyft engineers describe a Fake Payment Provider that simulates arbitrary processor behavior, such as a capture that fails once and then succeeds, without calling real processors. Scenarios are dictionaries carried inside the card token, so state survives asynchronous retries and follow-on charge operations. The fake is used in staging and in production load tests and canary rollouts, restricted to simulated riders.
Toward a Theory of Evaluability
This Apple Machine Learning Research paper by Kunal Talwar formalizes evaluability for black-box models whose test samples may have leaked into training. It characterizes the distributions where accurate output-only evaluation is feasible, gives tight sample complexity bounds, and shows that limited query access can strengthen evaluators, drawing connections to self-correctors from program testing.
Building strong safety evaluations
The post gives Anthropic's recommendations for designing, running, and reporting AI safety evaluations, drawing on lessons from the company's system cards. Key points include matching eval design to the question, measuring the right construct (such as appropriate response rate in multi-turn conversations rather than refusal rate), calibrating LLM graders against human labels, preregistering analyses, and reporting uncertainty and configuration fully.
GitHub availability report: October 2026
GitHub's October 2026 availability report covers eight incidents, including Actions job-assignment failures caused by a cloud networking disruption, a migration-driven database overload, a write outage from a misrouted database leader, and a Copilot credential failure. It also details ongoing Azure capacity expansion and reliability work, with each incident broken down by root cause, response, and planned mitigations.
Vercel Sandbox Python SDK 1.0 is now available
Vercel released version 1.0 of its Python SDK for Vercel Sandbox, stabilizing the public API for running commands and managing files in isolated Linux microVMs. The release adds context managers for automatic cleanup, AnyIO-based async support for asyncio and Trio, and a sync module, and it switches command input and output to bytes, a breaking change for 0.x users.
Well, That Wasn't Fun: October 16 '26
Fly.io describes its emergency response to CVE-2026-99999, a Linux kernel checksum-offload bug that let a guest write into host IPv4 headers and read out-of-bounds memory. The post explains the exploit mechanics, why XDP rules could not catch it, how the team staged reboots across hosts and Managed Postgres clusters, and the credential and secret-handling lessons they took away.
How H6.ai builds AI-powered financial automation with Lakebase Postgres
H6.ai describes how it uses Lakebase Postgres copy-on-write branches to freeze per-customer financial databases as regression-test snapshots for its invoice-processing AI. The post explains the test harness flow of migrating snapshots, forking run branches, and judging outcomes with deterministic assertions plus a panel of AI judges. It is an engineering write-up with concrete testing and isolation design rather than a product tour.