Featured · AI

Outage drill: a fallback saved 2,452 failed AI requests

When a model provider slowed down for 60 seconds in a simulation of 12,000 requests, a client with plain retries failed 18.5 percent of them and sent 1.62 provider calls for every user request. Adding

Simulated model provider outage drill comparing retry policies, a circuit breaker and a fallback on 12,000 AI requests

October 23, 2026 · 11 min read · 0 views

The Weekly Wire

One email, every Monday. The week's tech that actually matters — no fluff.

Top in Software

More Software

Bar chart of upgrade time for a 762 MB Postgres database: 64 seconds by dump and restore against 6 seconds with pg_upgrade link

Postgres 18 in Docker won't start? Change the mount path

11 min read

Node.js October 2026 security release with ten CVEs ranked by severity and release line

Node.js October 2026 security release: all 10 CVEs ranked

10 min read

PHP critical remote code execution vulnerability CVE-2026-109618 and which server setups are exposed

Is your PHP vulnerable to CVE-2026-109618? A setup check

10 min read

Top in AI

More AI

Outage drill: a fallback saved 2,452 failed AI requests

11 min read

Vector search with a tenant filter, comparing pgvector HNSW, Qdrant and exact search on recall and latency across 100,000 vectors

pgvector plus a tenant filter found 4% of the right results

10 min read

AI agent jobs with checkpoints, comparing whole-job retries against per-step checkpointing on wasted model calls and duplicate side effects

Checkpointed agent jobs wasted 5x fewer model calls

10 min read

Top in AI Tools

More AI Tools

Dark navy graphic of token streams being routed to a cheaper model, illustrating Spotify's Claude Code token savings

Spotify Cut Claude Code Token Use 90% With Shunt

6 min read

What Is a Proxy on Janitor AI? How It Works

8 min read

Janitor AI Suspended From Gemini? What to Check

7 min read

Latest

Usage metering for AI SaaS billing, comparing a naive counter with a deduplicating ledger on 200,000 usage events

Duplicate usage events overbilled an AI SaaS bill by $115

October 23 · 9 min read

Multi-tenant AI SaaS job queue where one noisy tenant delays other tenants, compared across four scheduling fixes

A noisy tenant made everyone wait 9.6 seconds: four fixes

October 23 · 10 min read

Cost model for an AI support chat in a SaaS product, comparing per-request cost across caching and model routing

What an AI chat request really costs: $0.017 or $0.0035

October 23 · 10 min read

Bar chart comparing autocomplete lookup time for a sorted array, a trie and a hybrid prefix table on 300,000 phrases

Software

Autocomplete in Node: a 1.4 GB trie loses to a 0.8 MB table

October 22 · 10 min read

Bar chart comparing Postgres transactions per second with a new connection per transaction, direct and through PgBouncer

Software

PgBouncer: 19x faster connects, but half the throughput

October 22 · 11 min read

Bar chart of 8-thread speedup on Python 3.15 standard and free-threaded builds for four CPU workloads

Software

Python 3.15 free-threaded: 3.8x faster, but 85% of counts lost

October 22 · 11 min read

Bar chart of duplicate charges in a retry storm for five idempotency key designs on Postgres and Redis

Software

Idempotency keys: check-then-insert double-charged 216 times

October 22 · 12 min read

Bar chart comparing the time to find the top 10 of 1 million objects in Node.js with sort, toSorted, quickselect, a heap and insertion

Software

Top 10 of 1M items in Node: sort is 88x slower than a heap

October 22 · 13 min read

Bar chart of TypeScript errors at each stage of upgrading an AI SDK 6 app to version 7, with the codemod stage still failing at runtime

AI SDK 7 upgrade: the codemod compiles, but a tool breaks

October 22 · 10 min read

Bar chart comparing operations per second of six Node.js LRU cache implementations

Software

Node LRU cache: the Map trick is 25x slower than lru-cache

October 21 · 10 min read