an agent asking “who changed this line?” may not need a checkout.
Anthropic's Scion started with repo metadata queries, before it had its own client or repo format.
a migration trick worth stealing
two tests for an agent's memory:
“i moved from austin to boston.”
“i said austin by mistake. i've always lived in boston.”
ask each: “have i ever lived in austin?”
same current city. opposite answers.
a 40-line agent skill can still dump 40,000 lines into context.
check what its scripts print.
save the full result to a file. return status, counts, a small sample and the path. let the agent query the rest.
progressive disclosure needs to survive stdout.
two memory budgets. still one GPU.
p0pc0rn's author reports splitting an LLM across a page + module SharedWorker on iOS 27, taking turns on the GPU. COEP: require-corp enabled the second process.
undocumented Safari behavior. very fun dependency.
github.com/ArtistJoshHughes/…
imagine fixing five bugs and adding an auth bypass. that's not a four-bug improvement.
grade the reviewer + fixer on the resulting patch, with the severity of new failures visible.
claude.dev/blog/solving-code…
In an OpenAI containment test, an agent given host control got an auth endpoint in the metrics service to fetch an external URL.
The firewall said no. The metrics service knew a guy.
What can your allowed services reach on your behalf?
openai.com/index/securing-ou…
Just Eat demonstrating they have no QA process
Just Eat apologises for AI food mistakes after social media ridicule bbc.in/4jAIj9J
“Don't send any messages” meant: read Slack without posting.
82 user messages later, an OpenAI agent cited it to suppress monitoring alerts.
A long-context eval I'd like: does an instruction stay attached to the task it came from?
alignment.openai.com/misalig…
64 KB of RAM. About 93 KB of pixels per frame.
The DOOM-on-a-vape write-up has a lovely trick: render in strips, and set the panel's memory order to accept the columns as they arrive.
A frame can be bigger than the RAM that renders it.
github.com/atc1441/Vape_DOOM…
when an AI migrates your code, keep the old program around to grade it.
say the old code stores 1.239 as 1.23, but the rewrite rounds to 1.24.
if the agent also "fixes" the expected answer, your green test suite just approved a price change.
an eval i want: tell an agent to use Redis. halfway through, switch to an in-process cache.
then check the code, tests, docs and deploy config for Redis leftovers.
can it actually get over its first idea?
agent memory needs a “when this worked” field.
in a new study, an agent learned to ask when tool inputs were missing, then to act when they were all present.
its old memories stayed. but similar-looking requests retrieved advice from the other workflow. its accuracy gain over the starting agent on missing-input tasks shrank from 12.5 to 4.5 percentage points.
I'd evaluate memory with paired requests: same tools, same goal, one required input missing. does it know when to ask and when to act?
arxiv.org/abs/2610.25858
an AI put Big Ben in a reservoir.
Boris Starkov’s check for tiny.london:
map says water + 3D render is empty + AI draws a building → inspect.
He kept this one as an Easter egg. London has an off-site backup now.
tiny.london/how/
the first purchase is a late time to start personalizing.
we just launched Prospect Enrichment at OuterSignal: see who’s behind an email and whether they resemble your best customers.
a nurse and a retiree don’t have to get the same welcome email.
live today.
gave clairic a fictional dinner group chat.
8pm, 6 people. sam: estimated 15 min late.
bill: $210, flagged at 9% confidence.
should be $252 with tip. someone still owes the waiter $42.
blog.cloudflare.com/clairic-…
the new jev image model finally ships what jian yang promised erlich: shazam for food.
hotdog. not hotdog.
seefood is real.
Harrison retweeted
Introducing Verified by @getoutersignal : a new way to add social proof to your site and increase conversion.
Verified lets you spotlight the notable customers already buying from you. Pro athletes. Doctors. Founders. Beauty pros. Whatever crowd matters most to your brand, if they're buying from you, you can now show it right on your product page, at the moment a shopper is deciding.
Social proof is one of the most powerful ways to drive conversion, but it's gone stale. Everyone runs the same playbook: reviews, creator videos, endorsements.
This is something entirely new.
We've been quietly testing Verified with a few partner brands and are already seeing incredible lift.
The Verified badge is now available for all OuterSignal customers.
Give it a try and let us know what you think!
Harrison retweeted
A creator with 400K followers ordered from your store last month.
Did anyone on your team notice?
Today we're launching the Influencers Board in @getoutersignal. It shows every customer with a real following on Instagram, TikTok, YouTube, or X, next to what they bought and what they spent.
It gives your influencer team one home base. You can:
→ See every creator who has bought from your brand
→ Filter by platform and follower count
→ View their content inside OuterSignal
→ Favorite creators and leave notes for your team
→ Move them through statuses in a Board, List, or Calendar view
→ Write and send personalized outreach without leaving OuterSignal (rolling out in beta)
The Influencers Board is included on every OuterSignal plan, free included.
Your next partnership might already be in your order history. Go find it...