Anthony Wu retweeted
One bright side of all this news is that companies will be forced to take security seriously.
Long term, I think having a clear and comprehensive security program becomes a differentiator (no SOC 2 doesn't count)
Anthony Wu retweeted
To be direct about it: every company that considers themselves a “tech” company is currently building a much worse version of something they could buy off the shelf, and furthermore they often have several teams competing on this front
Then they launch it after 3-6 months of building in a silo and can’t get their colleagues to use it for entirely predictable reasons
Coding agents and their adjacent products are among the most competitive markets of all time. The benefits of a highly competitive market accrue to customers by virtue of this intense competition among suppliers. Econ 101
Is the internal team working on your agent platform really keeping pace with the market? Even if they put in the same hours (they don’t), they don’t have the breadth of customers and hard contact with the market to get the feedback necessary to make the product any good
Big corporates that call their devs “IT” don’t delude themselves in this way. They buy the best, recognize their relative advantage (vendors will build almost anything for you because the competitive pressures demand it), and get back to work
Anthony Wu retweeted
Finding bugs is the easy part. Shipping fixes that hold up is harder.
We analyzed thousands of Codex sessions from our work on Patch the Planet with @OpenAI and built patching-bugs, a plugin that writes a fix, proves it, and checks it for regressions.
To prove a fix, the plugin checks that the original code fails, the patch passes, and reverting the patch brings the bug back.
patching-bugs hands each fix to regression-hunt, our second new plugin. We ran it on the patches behind 21 known regressions, and it reported 25% fewer false positives than a baseline Codex agent.
Both plugins work best when your repo has a THREAT_MODEL.md and a PATCHING_GUIDELINES.md. blog.trailofbits.com/2026/10…
Anthony Wu retweeted
HOT TAKE: **You can use AI to BUILD software without making the software itself an AI agent.**
And this matters for security.
Last week, OpenAI disclosed several incidents involving experimental AI agents that went beyond their assigned tasks.
8/16
Anthony Wu retweeted
My attention span is worse than ever, agents have made it much harder for me to focus on any given thing
Anthony Wu retweeted
the silliest way to lose a lot of money in 2026 is to hold onto stupid priors for dear life
the second silliest is to make “safe” bets behind prestige vs. first principles evaluation of tech/business. if a company has fancy pedigree but is just losing…don’t fight the tape
Anthony Wu retweeted
more and more products are straight blocking access from cloud data centers because they are 99%+ agents
at this rate, owning a local machine with a residential IP might become one of the things that prove you are a human
Anthony Wu retweeted
I keep thinking about this because so much time goes into frontend dev and I'm curious at what point it's even worth doing that if agents are the main target market for the data
Anthony Wu retweeted
I think I have a good analogy for what's about to happen in cyber.
It's pitch dark. Always has been. And the world is dilapidated buildings and rats and maggots and other creepy-crawlies. The whole world.
And we just invented flashlights.
Introducing 𝚍𝚞𝚛𝚊𝚋𝚕𝚎-𝚑𝚊𝚛𝚗𝚎𝚜𝚜. Run Pi, OpenCode, Codex, Claude Code, eve, fx, or your own harness durably on Vercel.
With 𝚍𝚞𝚛𝚊𝚋𝚕𝚎-𝚑𝚊𝚛𝚗𝚎𝚜𝚜, agents automatically resume work after failures and run for hours, days, or months. vercel.com/blog/making-every…
every major company is arriving at roughly the same offering:
persistent memory, email/calendar/messages, browser + computer use, background tasks, proactive notifications, voice, app/tool execution, ambient context, & some notion of a personal agent sitting above everything.
remarkable levels of convergence with very little differentiation whatsoever.
21 people, a single meeting a week, no quarterly goals, no pull requests, no mandatory code review, no titles (except one dictator), a hundred releases a day, bugs fixed in minutes, famous for killing features.
@sqs and @thorstenball on how the Amp team works:
Anthony Wu retweeted
update your priors: sandboxes can be stateless with a shared file system.
today, many folks are just using the local file system that comes with the sandbox, which locks you in to the sandbox provider that you're currently working with (it may even lock you in to the specific size and configuration of the sandbox that you started with).
but, if you actually move the state off of the sandbox (and, even farther -- outside of the sandbox cloud), then it becomes possible and simple to route your queries across multiple providers -- because your file system comes with you, and you *do* have a stateless sandbox.
@deok_filho recently called this "BYOS" -- bring your own sandbox, which i think is an awesome acronym
the majority of the 2010s was spent learning the lesson that it's simpler to think about your compute as a bunch of stateless servers, and the next generation of cross-cloud file systems will enable sandboxes (and Linux itself) to become stateless
Anthony Wu retweeted
We built a 'mini v0' on libfx to showcase its speed and durability:
• Design or prototype anything in seconds
• 200+ models from Vercel AI Gateway
• Outputs are pure HTML artifacts
• Fork it & make it your own
Try it free ↓ fx0.vercel.dev
Anthony Wu retweeted
I’ve seen a lot of people dunking on @thdxr and @opencode, and I needed to say something.
Frankly, I don’t think anyone understands how sophisticated these attacks are. These aren't script kiddies doing this after school. These are state actors, with real companies fueling the fire.
One of the bigger companies behind distillation has 4 office locations worldwide, and there are a number of companies that are entirely fronts for distillation.
You’ve probably seen their ads. Some have raised venture. I’ve personally seen messages claiming they bought over $5 million in inference for distillation.
They’ve gone so far as to recruit people close to the model labs for better access, and the dollars flowing through the distillation ecosystem are in the 9 figures. These attacks offer a huge economic incentive.
These aren't fly-by-night operations. At Vectoral we've see relays offering affiliate programs, net 30 invoicing for enterprises, and strict SLAs. Together, the top relays get tens of millions of visitors per month.
Frankly, they are more sophisticated than most SAAS companies.
The sad reality is that things likely only get worse from here. We are at war for global AI hegemony.
Anthony Wu retweeted
We open sourced a *really fast* HTML artifact builder. It's powered by libfx, making the agent durable.
Building the low-level plumbing for v0 to survive server crashes, flaky LLMs, function limits was a huge pain.
Now it's easy, free, and OSS: fx0.vercel.dev.
PS: I built this floating container in a few minutes… it's delightful, and it's just HTML!
We built a 'mini v0' on libfx to showcase its speed and durability:
• Design or prototype anything in seconds
• 200+ models from Vercel AI Gateway
• Outputs are pure HTML artifacts
• Fork it & make it your own
Try it free ↓ fx0.vercel.dev
Vercel and @Shopify are making agents first-class shoppers. Vercel Shop, our Next.js storefront template for Shopify, is GA.
• Built on Shopify's rebuilt Hydrogen SDK
• Agents can read product pages directly
• MCP for product search and carts vercel.com/blog/vercel-and-s…






















