GitHub Actions lockfiles and an AWS SBOM inventory cover both ends of supply-chain hygiene
One pins what a build runs, the other records what production runs; both now ship by default to large numbers of teams.
Software news, read and annotated by a working engineer
One pins what a build runs, the other records what production runs; both now ship by default to large numbers of teams.
Two releases on 4 and 5 November give agents a name and an approval step, at the edge of a website and inside a project's lockfile.
Two releases on the same day, one about reading disclosures and one about fixing weaknesses, describe the same pressure from opposite ends.
Amazon's CEO said committed capacity is largely allocated through 2027 with headroom kept for on-demand, EC2 and Bedrock; engineers report tighter ordinary instances already.
Nolan Lawson had a coding agent build the same Pokémon app five times per framework; Lighthouse scores barely moved, while build cost and shipped JavaScript differed widely.
Robert Haas counts back-patched commits up more than 60% on last year and late reverts of three large features; the cause, he argues, is that bugs became much easier to find than to fix.
Docker's account of a code review bot that rejected a routine PostgreSQL bump with a fake CVE shows where agent safety stops: the agent obeyed its limits, and the risk moved to the person reading its output.
Project Zero counts 45 BigSleep bugs against 37 from Fuzzilli with 25–60% overlap; a Red Hat survey finds only 32.9% of EMEA CIOs have a complete view of their open source supply chain.
AWS sells the model on price and agent benchmarks against GPT-6.1 Sol; GitHub sells it on terminal and C# tasks — and both say little about what changes for teams that already pay.
Cloudflare's Auto Router scores each request from the most recent turns, capped at 8 KiB. The sharpest critique of the launch is about what that compact view misses.