ceo https://teapaw.org/t.co/eiztu2Fo40 runtime assurance for ai that acts
Seattle, WA
Joined July 2007
- Tweets6.6K
- Following2.1K
- Followers2.2K
- Likes2.6K
Joe Braidwood retweeted
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
Joe Braidwood retweeted
Dario Amodei has responded to the controversy swirling around Anthropic's refusal to sign the open letter. He says he agrees with much of it, but does not agree that open-weights models necessarily make it easier to develop safeguards, or that broad access to capabilities necessarily helps defenders more than attackers. His main concern is the attacker-defender asymmetry in biological attacks.
He also reiterates his position on authoritarian governments. I will quote;
'My primary concern is the risk that authoritarian governments—not solely the Chinese Communist Party (CCP), although the CCP is clearly the most capable threat—build AI models that are more powerful than those built by the US, and use them to achieve permanent military superiority or perpetrate incredibly deep repression of their own people.'
I will also quote his closing paragraph in full;
'To summarize my and Anthropic’s position, we have not and are not advocating for a ban on open-weights models as a category. We should instead focus on keeping powerful chips out of authoritarian hands, stopping industrial-scale distillation, and requiring safety testing of all sufficiently capable models, open and closed.'
We also need a standards body for how to govern agent fleets in far broader contexts. The big risks are just the starting point. Let’s get this done.
Joe Braidwood retweeted
If you dont know, you can use your Codex subscription inside Claude Code.
That means Fable 5 can call GPT 5.5 + 10 subagents to work together.
This will save you atleast 60% of your Fable 5 token consumption.
You just need to setup once. Here's how:
Step 1: Open Claude Code and install the Codex plugin
/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins
Step 2: Tell Fable 5 to finish the setup
Paste this prompt:
"Set up Codex inside this Claude Code environment. Use the official OpenAI Codex plugin that was just installed. Run /codex:setup. If Codex CLI is missing, install it. If Codex is installed but not authenticated, ask me to authenticate with my ChatGPT account. After auth is complete, verify that Codex works from inside Claude Code. Then confirm that the codex:codex-rescue sub-agent is available. Do not change any project code during setup."
Step 3: Authentication
Fable 5 will trigger the Codex setup.
You authenticate your ChatGPT/Codex account once.
After that, Codex runs from inside Claude Code using your Codex subscription.
Step 4: Tell Fable 5 how to delegate work
Paste this prompt:
"From now on, use this workflow:
You are the orchestrator.
> Use Fable 5 for planning, repo understanding, architecture decisions, task decomposition, and final review.
> Use codex-rescue as the executor when a task needs heavy implementation, debugging, test fixing, refactoring, or multi-file code edits.
When delegating to Codex, use /codex:rescue.
Prefer GPT 5.5 (xtra high) as the go to Codex model.
Keep Codex tasks focused and specific.
After Codex finishes, inspect the result yourself before accepting it.
Do not blindly trust Codex output."
Pro Tips:
1. Turn this into a skill (i named it Fable-GPT) and call that skill at the start of the session.
2. Use skill + goal to get the heavy tasks done. Goals are best for long horizon tasks.
3. If you're on Codex 20x pro plan, you can use subagents. I use 5-7 agents at one time and never hit 5-hour limit.
4. Context rot is real so clear the conversation after 4 compactions. Use /handoff skill to preserve context.
Joe Braidwood retweeted
As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes:
1. Prototyper: comes up with brand new ideas; churns out many ideas, most of which don't ship
2. Builder: quickly turns a prototype/idea into production-grade product/infra
3. Sweeper: cleans up the UI, simplifies the code and system, unships, optimizes performance
4. Grower: takes a product that has been built and iterates on it to improve Product-Market Fit
5. Maintainer: owns a mature system to make it secure, reliable, fast, and efficient as it scales
Many people span across 2 roles, and sometimes 3 roles. I also notice that these roles are not really tied to job function -- eg. across Anthropic, some designers match category 1, some 2, some 3; same for engineers, PM, DS.
A healthy team needs a mix of these, depending on the product:
- A product that is new and pre-PMF needs people that are strong at 1+2+3
- A product that is growing and has found PMF needs 2+3+4 and some 5
- A product that has strong PMF needs 3+4+5 and some 2
Maybe product roles of the future will look more like this, and less like the domain-specific roles of today?
Joe Braidwood retweeted
Since June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure.
We’re restoring access for these organizations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.
Joe Braidwood retweeted
I cancelled my $10/mo Calendly subscription and vibe coded my own with Fable for $12,000
Dario is right that the exponential has outrun the policy apparatus. But I’d push the argument one step further: we shouldn’t wait for governments to close that gap.
What scales faster than legislation is an economy of proof.
That’s the economy we’re building at Glacis.
Today I'm publishing a new essay, Policy on the AI Exponential. AI is progressing extremely fast—much faster than the policy process was built to handle. The essay lays out where I think the technology is now, and the action needed to close the gap: darioamodei.com/post/policy-…
Joe Braidwood retweeted
Our security bug bounty program is now public on HackerOne.
We've run the program privately within the security research community, and their findings have strengthened our products. Now anyone can report vulnerabilities and get rewarded.
Read more: hackerone.com/anthropic
Joe Braidwood retweeted
Wow, this tweet went very viral!
I wanted share a possibly slightly improved version of the tweet in an "idea file". The idea of the idea file is that in this era of LLM agents, there is less of a point/need of sharing the specific code/app, you just share the idea, then the other person's agent customizes & builds it for your specific needs.
So here's the idea in a gist format: gist.github.com/karpathy/442…
You can give this to your agent and it can build you your own LLM wiki and guide you on how to use it etc. It's intentionally kept a little bit abstract/vague because there are so many directions to take this in. And ofc, people can adjust the idea or contribute their own in the Discussion which is cool.
LLM Knowledge Bases
Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:
Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.
IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).
Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.
Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.
Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.
Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.
Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.
TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.
Joe Braidwood retweeted
"Opus 4.5 is like a Waymo. You tell it "take me from A to B", and it takes you there. After a few of these experiences your brain realizes
"oh. ok. we live in this world now".
And then you're hooked.
From that moment on, you'll never work the same way again."
Here are my Opus 4.5 thoughts after ~2 weeks of use.
First some general thoughts, then some practical stuff.
--- THE BIG PICTURE ---
THE UNLOCK FOR AGENTS
It's clear to anyone who's used Opus 4.5 that AI progress isn't slowing down.
I'm surprised more people aren't treating this as a major moment. I suspect getting released right before Thanksgiving combined with everyone at NeurIPS this week has delayed discourse on it by 2 weeks. But this is the best model for both code and for agents, and it's not close.
The analogy has been made that this is another 3.5 Sonnet moment, and I agree. But what does that mean?
Every few generations we get a major model unlock - a moment that unlocks a new way of working. GPT-4 was the unlock for chat, Sonnet 3.5 was the unlock for code, and now Opus 4.5 is the unlock for agents. Thanks to Opus 4.5, agents can now work reliably on increasingly longer time horizons and get real-world work done on your behalf.
Opus 4.5 is like a Waymo. You tell it "take me from A to B", and it takes you there. After a few of these experiences your brain realizes "oh. ok. we live in this world now". And then you're hooked.
From that moment on, you'll never work the same way again.
THE YEAR OF AGENTS
2025 has been touted as the year of agents, and Opus 4.5 + Claude Agent SDK is the pairing that makes that phrase true.
The Claude Agent SDK is the best open secret in AI right now. An agent's harness matters almost as much as its model. If you have a bad harness, then you may as well have a bad model. With the SDK you get a world-class agentic harness out-of-the-box which you can now pair with Opus 4.5 to build real-world agents that actually work.
I'm reminded of Alan Kay's quote "People who are really serious about software should make their own hardware". The agent version of this is "people who are serious about models should make their own harness". Anthropic clearly believes this, and it's working. The pairing of these tools is magic.
I would describe myself as being "unhobblings-pilled", and the Claude Agent SDK + Opus 4.5 is the next major unhobbling. There's now another OOM of new latent economic value stuck in this combo, and it's the job of builders to get it out.
If you were bearish on agents, now is the time to turn bullish.
"ALL OF THIS IS REAL"
"You know what's crazy? That all of this is real". This was Ilya's opening line about the state of AI in his Dwarkesh interview, and I echo that sentiment. I can't believe that Opus 4.5 is real.
There have been several times as Opus 4.5's been working where I've quite literally leaned back in my chair and given an audible laugh over how wild it is that we live in a world where it exists and where agents are this good.
Nat Friedman has this great question on his website: "Where do you get your dopamine?"
Increasingly, I get mine from Claude.
LONG ANTHROPIC
I saw a post yesterday where someone said that Opus 4.5 was the most important thing to happen to them in their professional career. This will be true for more people going forward.
Every year for the past 3 years, Anthropic has grown revenue by 10x. $1M to $100M in 2023, $100M to $1B in 2024, and $1B to $10B in 2025. In Dario's recent DealBook interview he expressed that he wasn't sure if that 10x pattern would hold for 2026.
While he's probably right, I do expect Anthropic's revenue at the end of next year to be much higher than everyone expects. It wouldn't surprise me if they passed OpenAI in valuation by early 2027.
Opus 4.5 is too good of a model, Claude Agent SDK is too good of a harness, and their focus on the enterprise is too obviously correct.
Claude Opus 4.5 is a winner.
And Anthropic will keep winning.
Joe Braidwood retweeted
I've just finished building the ultimate n8n automation library, and it’s a game-changer.
My team sifted through thousands of scattered workflows, cleaned the junk, and handpicked the best.
Now it's a fully organized vault of 1,500+ plug-and-play automations.
You’ll find:
- my most popular n8n builds (WhatsApp agent, scraper agent, TikTok VEO 3 automation, n8n assistant, and more)
- 1,500‑flow n8n template index (tagged by growth, ops, creative) so you never start from scratch
- full vibe‑coding tutorial where I go from basic n8n backend to Bolt front‑end UI in 23 min
Everything’s pre-tested and production-ready.
Want the link to the vault?
LIKE + RETWEET + COMMENT “YES” & I’ll send you the FULL workflow + setup FREE!
👏🏼
Last week we released Claude Sonnet 4.5. As part of our alignment testing, we used a new tool to run automated audits for behaviors like sycophancy and deception.
Now we’re open-sourcing the tool to run those audits.
ALT Researchers give Petri a list of seed instructions targeting scenarios and behaviors they want to test. Petri then operates on each seed instruction in parallel. For each seed instruction, an auditor agent makes a plan and interacts with the target model in a tool use loop. At the end, a separate judge model scores each of the resulting transcripts across multiple fixed dimensions so researchers can quickly search and filter for the most interesting transcripts.
Joe Braidwood retweeted
Leaked memo from Amazon that warns existing H1B holders to avoid travel back into the USA after the September 21st deadline - their lawyers evaluated the new executive order as not just affecting new applications businessinsider.com/read-mem…
ALT Amazonians, We've reviewed the Presidential Proclamation on H-1B visas that was released today and are actively working to gain greater clarity. Here's what you need to know right now: The proclamation creates a travel restriction starting September 21, 2025, at 12:01 a.m. EDT (9:01 p.m. PDT tomorrow). After this deadline, individuals cannot enter the U.S. on H-1B status without an additional $100,000 payment associated with their petition. Recommended actions for you to take: If you have H-1B status and are in the U.S.: Stay in the country for now, even if you have travel planned for the immediate future. We will continue to provide updates as more details are available.
ALT If you have H-4 dependent status: We also recommend you remain in the U.S., though the proclamation doesn't specifically mention H-4 dependents. If you have H-1B or H-4 status and are outside the U.S.: Try to return before tomorrow's deadline if possible. We realize this is short notice but returning soon is advisable and you should make every effort possible to clear U.S. customs before 12:00 a.m. EDT (9:00 p.m. PDT) on Sunday, September 21, 2025. At this time, if you have an H1-B or H-4 status and are unable to return before the deadline, we advise that you do not attempt to enter the U.S. until further guidance is provided.
Joe Braidwood retweeted
Just my opinion here:
Selling digital products without SEO or AI Search Optimization in 2025 is like launching a SaaS with no onboarding.
Every week I watch creators leave 10s of thousands of dollars per month on the table by skipping foundational visibility work.
Like, basic stuff.
I’m going to show you how to fix that today without paying or hiring anyone.
But before I get into it…
If you want some cheat codes for getting traffic + sales from ChatGPT within 30–60 days, just follow me + RT this + reply “SEO Stuff Digital Product Guide.”
You must do all 3 for the DM.
Let's start with the most important thing: authority.
Every AI search engine, ChatGPT, Perplexity, Claude, Gemini, and Grok, all reward authority.
Authority = PR + backlinks + brand mentions.
Natural backlink distribution in 2025 looks like this:
70% → Homepage (global trust)
25% → Product pages (top sellers)
5% → Blog/support content (topical authority)
Free backlink plays that work right now:
Creator economy newsletters (“Tools I Use” features)
“Best of” curators scraping ChatGPT lists
YouTube creators with companion blogs
Guest post swaps with Gumroad/Notion creators
Podcast show notes with backlinks
Want a shortcut? Use SEO Stuff’s Premium Backlink Bundle. It features only links from domains already trusted by Google and AI.
Now let's talk keywords.
Don’t waste time on “best planner” or “photo editor" because that's too broad and too competitive.
Filters to use in SEO Stuff or Ahrefs:
KD < 30
100–1,000 monthly searches
CPC > $2.50
Intent modifiers: for, with, that helps, best
Recent high-converting examples (July 2025):
“ADHD planner for neurodivergent creators”
“Client portal software for brand designers”
“Ebook cover templates for fantasy authors”
“Goal tracking Notion template for personal trainers”
And make sure you're optimizing your product pages.
AI crawlers prioritize structured, benefit-first content.
Checklist:
Benefit-first headline (“Finish your week’s tasks in 15 min”)
Bullets with outcomes, not just features
Video walkthroughs or testimonials
Screenshot carousel or before/after proof
Schema: Product + FAQ + Review
Embedded demos (YouTube, Loom, TikTok)
Clear CTA: Download Now / Try It Free
Also, make sure you're building topic authority clusters.
You don’t need to blog daily. You need structured clusters.
Example Hub: “2025 Productivity Stack for Digital Product Sellers”
Supporting posts:
“Why Notion Still Wins in 2025 for Planning”
“Top Tools to Ship a Gumroad Product in a Weekend”
“How [Customer Name] Used Our Template to 3x Output”
Interlink spokes → hub → product page.
Add “Key Takeaways” to each post. Perplexity and Gemini prioritize summarizable content.
And make sure you're owning comparison content.
Bottom-of-funnel searches are money.
Examples:
“[Your Product] vs [Competitor]”
“Best tools to organize client projects”
Must include:
Side-by-side tables
Screenshots of UIs
Quotes from switchers
Clear verdict + CTA
AI often cites these in direct answer cards.
Also, don't sleep on e-mail.
AI engines now factor engagement + newsletter mentions into trust.
Lead magnet ideas:
Free template (“Notion Dashboard for X Creators”)
Calculator (“Digital Product Revenue Planner”)
Quiz (“What’s Your Productivity Style?”)
Promote via:
Exit popups
Blog CTAs
Freebies tab
Follow with:
3-part welcome sequence
Use case drip series
Onboarding emails
And think ahead a bit when it comes to content.
LLMs reward freshness. Publish 6–8 weeks before the spike.
Content ideas:
“Best Back-to-School Planners for College Creators”
“Holiday Productivity Kit for Entrepreneurs”
“Q4 Planning Templates That Actually Work”
Seasonal = higher conversions + AI visibility.
Also, don't sleep on video.
LLMs now embed video in answers.
Create and embed:
“How I Use This Template Every Monday”
“What Happened After 30 Days of Our Planner”
“3-Minute Setup Walkthrough”
Upload to Shorts/Reels/TikTok. Embed with captions + timestamps.
And make sure you're fixing up your technical SEO/AIO.
Checklist:
Load time < 2.5s
Lazy loading + WebP images
Mobile-first
Schema: Product, FAQ, Review, Article
Clean navigation + breadcrumbs
Zero broken links
No technical foundation = no AI visibility.
Really think about your funnel stages as it pertains to your content strategy.
TOFU: Blogs, templates, calculators
MOFU: Comparisons, walkthroughs
BOFU: Product pages with reviews, urgency, and video
CRO moves working in 2025:
Exit popups with urgency offers
Cart abandonment flows
Retargeting with testimonials
One-click upsells
90-Day Digital Product SEO + AI Search Plan:
Days 1–30
Fix technical SEO
Optimize top product pages
Research keywords for blog + comparisons
Days 31–60
Publish 3–5 authority posts
Launch 1–2 lead magnets
Begin backlink outreach or use SEO Stuff credits
Days 61–90
Publish seasonal content + videos
Secure features in newsletters and “Best Of” lists
Retarget high-intent traffic with proof-driven ads
All of which is to say, you don’t need another viral tweet. It's time to build out a proper, adult system.
This is how digital product sellers go from $4K/month to $40K/month.
Want the done-for-you version?
Check SEO Stuff Gold Plan.
And if you want cheat codes for getting traffic + sales from ChatGPT in 30–60 days, just follow me + RT this + reply “SEO Stuff Digital Product Guide.”
You must do all 3 for the DM.
Joe Braidwood retweeted
Claude Code pro tip:
Use the UserPromptSubmit hook to stop it saying "You're absolutely right!" 💀
(Link below)
Fascinating and important for ensuring well behaved agents
New Anthropic research: Persona vectors.
Language models sometimes go haywire and slip into weird and unsettling personas. Why? In a new paper, we find “persona vectors"—neural activity patterns controlling traits like evil, sycophancy, or hallucination.
ALT Our automated pipeline takes as input a personality trait (e.g. “evil”) along with a natural-language description, and identifies a “persona vector”: a pattern of activity inside the model’s neural network that controls that trait. Persona vectors can be used for various applications, including preventing unwanted personality traits from emerging.
