latent space astronaut, builder of https://xitter.cf/t.co/4ms0PmjFyq
Los Angeles, CA
Joined March 2009
- Tweets14.2K
- Following688
- Followers1.4K
- Likes14.1K
Pinned Tweet
agents are misunderstood and that is the heart of the problem with adoption.
companies expect to replace current employee workflows with agents but ai will always fail at this because the accountability structures within the company won’t allow it.
what do you do when a person messes up? you reprimand them. you have an organization that is designed to correct people issues so this is easy.
but what do you do when you drop an agent into this type of system and the agent makes a mistake? you don’t have the structure in place to resolve the issue so you end up in a failed state that you can’t recover from.
this is what we are seeing is happening in companies.
coding is different because the workflow for code (design, code, review, merge, test) is resilient to mistakes. you don’t generally trust any code from a dev without review/test so the dev portion of the process can be more fungible.
but something like customer service doesn’t really work like this. there is no review step…it’s all live fire.
then there are things like accounting workflows —something i’ve been working on. there is no generally agreed on tooling to handle things like a quarterly close that makes the human fungible. the workflow is usually filled with tribal knowledge and caveats that the agent designer probably didn’t know about.
so really the tools / workflows need to be rebuilt to be ai first then people can adopt agents within the context of the tool. for many companies this will be very hard to do so we will continue to see failed pilots.
the companies that will succeed will have one or more of the following;
1) no existing workflows thus they can tool up and agentify from ground up.
2) the understanding that you can’t replace existing workflows with agents but you can replace new flows with agents.
3) a very strong sponsor who can cram new workflows / tooling down the throat of the organization.
the g in agi means general.
which means that it will be able to do anything specific.
Vertical AI startups are all going to fail.
It shows a fundamental misunderstanding of the technology and the pace of progress.
Remember Jasper AI? When they were founded, they were on the cutting edge, so they were valuable. But 3 years later and what WAS the cutting edge is no longer even a full feature of an otherwise sellable product.
The way to build a company in 2026 is to orient around the cutting edge. Build a machine to stay on the cutting edge -- or better yet, to push it further.
Every decision we make at @tenex_labs is to optimize toward that.
🤯
almost seems like every able body eng needs to drop everything and help @openai and @AnthropicAI to figure this out.
i’m reminded of the episode in battlestar galactica where after the homeworld is attacked the surviving humans use ftl to jump away.
after 30 min the cylons find the humans so the humans need to jump again. at which point the cylons find them again after 30 min and they jump again. on and on for a week.
Jimi Smoot retweeted
If your company refers to engineers as the IT department, you have nothing to worry about. You were already breached last week.
time to just ship binaries i suppose.
Node.js Security Advisory: a high-severity RCE flaw in Node's Brotli decoder affects Node.js 22, 24, and 26.
Vercel WAF rules are live for every project and block known exploit patterns.
vercel.com/kb/bulletin/node-…
"software engineering" today is really just dev ops. i spend most of my day wrangling servers now.
and a small part on architecture.
Jimi Smoot retweeted
notice..
> Delta is blocking AI agents from booking flights
> JPMorgan Chase and Charles Schwab block agents from accessing accounts
> Yelp requires paid data licensing for agents
> Amazon has blocked Grok bot and other shopping agents.
the internet is starting to block AI agents that were supposed to do everything for us
Build in secure places.
❗️ The CEO of a South African coffee chain vibe-coded an internal dashboard for all of his 132 stores, only it had one big flaw: all of the sensitive internal data was stored in the source code of the login screen.
The dashboard URL was discovered through employees posting photos of their laptop screens on Instagram.
The CEO said he had discovered Claude and was "almost certain" that software itself will soon "become free," they said.
Jimi Smoot retweeted
All this talk about speed as a moat gets it wrong.
The whole point of having a moat is that if you slow down execution you still win because others can’t break your moat.
One difference of Twitter and Facebook is both had network effects and won. But FB never slowed down.
Sadly EVERYONE will have this moment sometime in the next 5 years.
I had this moment 3.5 years ago, and it took me a while (two years) to recover.
I think experiencing it sooner rather than later is probably better, but one way or another, that moment is inevitable.
🇫🇷🤖🧑🔬 FLASH | « Ils ont annihilé mon domaine de recherche » : le mathématicien français Hugo Duminil-Copin, médaillé Fields, révèle avoir PLEURÉ après qu’OpenAI a publié des centaines de travaux prétendant résoudre des problèmes mathématiques encore ouverts, dont 16 sur lesquels il travaillait.
Il s’inquiète particulièrement de l’avenir des jeunes chercheurs et évoque un « démantèlement » de sa discipline. (Le Monde)
