the 12.8T in the nvidia + cisco USFP announcement is the future number. the first planned modules are 16 lanes at 400G, so 6.4T in an OSFP-sized footprint. 12.8T needs the 32-lane version. 6.4T is still 4x today's 8 x 200G pluggable, at roughly 1.3x the power efficiency per bit. the stated goal is connecting more 1.6T endpoints without adding a network tier. every tier you skip is a layer of switches and optics you don't buy or power. nvidia has its own co-packaged optics switches and still co-designed a new pluggable. cisco's post spells out why: field-replaceable modules, multiple suppliers, the service model ops teams already run. each USFP module can dissipate up to 100 W with no dedicated cooling circuit inside it. newsroom.cisco.com/c/r/newsr…

devin automations can run a script before the agent starts. it showed up as one line in the Sep 30 release notes ("Preflight checks") and it decides whether a session starts at all, which makes it the first thing i'd add to any automation fed by alerts or webhooks. here's how it works 🧵

gotchas worth knowing before you ship one: 1. the Run now button skips preflight entirely. use Test against in the script editor 2. test runs don't start sessions, but any external call your script makes is real 3. default cap is 10 new items per run (max 50). extras aren't queued, so don't advance your cursor past them 4. attaching a secret doesn't open network access. the VM follows the automation's network rules

Big security day for anyone running a website or shipping software. Here's what happened and what I'd do now. 🧵