hebel. today · 10 November 202621 / 74

Science · Data Science Applications

Teams pivot from shipping volume to outcome-based AI benchmarks

PostHog engineers now use automated agents to identify and resolve product bugs, leading to a high enough fix rate that the system has moved out of beta. The team at Juno, an AI health…

PostHog engineers now use automated agents to identify and resolve product bugs, leading to a high enough fix rate that the system has moved out of beta (PostHog Blog). The team at Juno, an AI health assistant, merged 26 of the last 30 PRs generated by these agents to fix bugs like hanging chat boxes and add medication tracking features. SoundBoost, an AI mastering tool, used automated reports to detect that a domain migration caused daily Google traffic to drop to zero, allowing them to roll back the change immediately. PostHog's own self-driving team discovered through automated queries that their agents were skipping thousands of potential PRs, prompting the creation of a dashboard to visualize and resolve these failures (PostHog Blog).

Cambri saw a 44% increase in monthly ticket completion and a doubling of merged PRs after switching to Claude in 2026, even as their QA team shrank from 2.5 to 2 people (Roman Nikolaev from High-Impact Engineering). Meta's DEmate assistant uses task-specific instructions and automated checks to guide code changes, replacing general-purpose coding tools that struggled with their custom infrastructure (Ananth Packkildurai from Data Engineering Weekly). PostHog engineers improved ranking quality by 25% on 20 test queries by using a decision model to rerank session recordings based on how closely they matched user search terms (build mode).

  • Higgsfield merged over 120 self-driving PRs directly from GitHub to address issues like broken Stripe tracking and noisy exceptions (PostHog Blog).
  • Teams can use Monte Carlo simulations on historical throughput data to generate probabilistic delivery forecasts rather than relying on single fixed dates (TLDR Product).
  • Mercury uses a Slack bot with shared business definitions to answer employee data questions, identifying gaps in underlying data through the questions asked (Ananth Packkildurai from Data Engineering Weekly).

Try today · 30 min

Build a tracking report that ranks your top three recurring technical issues by frequency to identify where to focus your next fix.

Start by querying your error tracking or ticketing system for the most common failure patterns from the last 30 days.

One email instead of your newsletter pile.

Forward the newsletters you already get (a Gmail filter does it) and tell Hebel your goals. Every morning it reads them with 20,000 others and sends one short email, with one small thing to do today. Fewer emails, not more. Free during beta.

Start my paper →
LinkedIn X

Sources

  1. PostHog Blog on hebel. →Details how teams use automated reports and PR generation to fix bugs and monitor traffic drops.

  2. TLDR Product on hebel. →Outlines using Monte Carlo simulations on historical data for probabilistic delivery forecasting.

  3. Ananth Packkildurai, Data Engineering Weekly on hebel. → ✎ Written by a personExplains how Mercury and Klaviyo use shared business definitions to make AI data tools trustworthy.

  4. build mode on hebel. →Shows how reranking search results with a decision model improved ranking quality by 25%.

  5. Roman Nikolaev, High-Impact Engineering on hebel. → ✎ Written by a personReports on the impact of moving to Claude on ticket volume and PR throughput at Cambri.

More in Science

Translating science through personal relevance

NASA expands lunar infrastructure with $10 billion contract pool

Global emissions reach record high despite oil consumption dip

Data Science Applications yesterday

Establishing baselines before automating technical workflows

Sign in to Hebel first

Likes, your collection and done Hebel need an account. We email you a six digit code; no password.