olympian. crafting agents who craft agents
Melbourne
Joined March 2014
- Tweets8.5K
- Following958
- Followers6.8K
- Likes42.4K
Pinned Tweet
Be Nathan Fielder:
> grow up in Vancouver, Canada
> join your high school improv group alongside Seth Rogen
> graduate with a commerce degree in 2005
> decide accounting isn't for you
> start doing comedy
> land a correspondent role on This Hour Has 22 Minutes
> create a segment called Nathan on Your Side
> give real people increasingly ridiculous advice with a completely straight face
> turn the concept into Nathan for You on Comedy Central in 2013 at 29
> spend four seasons convincing real businesses to try increasingly absurd marketing schemes
> sign an overall production deal with HBO in 2019
> convince them to fund The Rehearsal
> build an exact replica of a New York bar so one man can practise a conversation
> co-create The Curse with Benny Safdie and cast Emma Stone
> return to The Rehearsal in 2025 and spend over two years learning to fly
>actually pilot a Boeing 737 carrying more than 100 actors
>secretly spend three years filming convicted Theranos founder Elizabeth Holmes
> co-direct You Can See Everything with Lance Oppenheim
> convince A24 to release the nearly three-hour documentary
> make it for roughly $7 million
> open in 2026 to $15 million from just 855 cinemas
> sell out more than 2,000 screenings in one weekend
deliver the biggest non-music documentary opening in over 20 years
nice
if you want to witness a little bit of beautiful magic today, you should be trying out autorouter.com
Meet AutoRouter, the most intelligent way to pick models
You’re wasting tokens selecting frontier models for every prompt, and sometimes for worse performance
AutoRouter handles the decision layer:
- Understands the prompt request
- Picks the best model fit for the task
- Saves you money at every turn
- Plugs into your current setup
Works with your plans & every leading model inside Codex/Claude Code
Beta is live: autorouter.com
The wild Seinfeld story
> grow up in Massapequa, Long Island
> your dad paints signs for a living and collects jokes
> graduate college in 1976 and decide to become a stand-up comedian
> spend years performing in New York comedy clubs
> make your first appearance on Johnny Carson's Tonight Show in 1981
> meet Larry David and pitch NBC a sitcom about nothing
> secure backend profit royalties for both of you
> convince the studio to let you act, but not Larry
> recruit Michael Richards, your co-creator's old colleague from Fridays
> cast Julia Louis-Dreyfus, who worked with Larry David on Saturday Night Live, and also a daughter to a billionaire
> pick Tony Award-winning Broadway actor Jason Alexander as George
> base George's personality on Larry's own disastrous social interactions
> premiere Seinfeld in 1989 to terrible audience ratings
> turn it into the number one show in America
> negotiate $1 million per episode for the final season
> pay George, Elaine and Kramer $600K each per episode
> they ask for a share of the show's profits
> they get turned down
> NBC offers you $110 million to make one more season
and you turn it down
> end the show in 1998 with 76 million Americans watching the finale
> watch the show generate billions in syndication revenue
> launch Comedians in Cars Getting Coffee in 2012
> sell the series to Netflix in a deal reportedly worth $100 million
> Hulu pays up to $180 million for Seinfeld streaming rights in 2015 again
> Netflix then re-pays more than $500 million for global streaming rights
> become a billionaire
> in 2026, 72 million Americans watch a show you stopped making 28 years ago
> turn 180 episodes about nothing into an estimated $7 billion business
listen to it NOW
i made this lightning fast terminal based stock market app, powered by @findatasets and @OpenAI
and so far I think @bot is the best at this
I think the personal AI agent race gets misunderstood as a model or feature race.
It’s a friction race.
Lower friction → more interactions → more context → more permissions → better actions → more trust → more delegation → higher switching costs.
I’m already seeing it in my own behavior. I use raw LLMs less, but AI more. The difference is that the agent is just there, inside the interfaces I already use.
That’s how an assistant becomes infrastructure.
I built a podcast insight ranker with Jev + DeepSeek via Vercel AI Gateway.
Jev scores 85 transcript sections from the @AcquiredFM Home Depot episode in around 380ms per batch.
Low-confidence scores go to DeepSeek, which also explains the top 5.
First quick test of Jev from @typesafeai
Install via @vercel gateway
Sending evaluation (input validation passed):
{
"model": "typesafe-ai/jev",
"state": {
"character": "2"
},
"questions": {
"kind": {
"type": "choice",
"instructions": "Is the character a number or a letter?",
"criteria": {
"number": "The character is a single numeric digit from 0 through 9.",
"letter": "The character is an uppercase or lowercase English letter from A through Z."
}
}
},
"maxRetries": 0
}
[Jev 60e7bb68] Evaluation received:
{
"choice": "number",
"probabilities": {
"number": 1,
"letter": 0
},
"durationMs": 1437,
"tokens": {
"input": 342,
"output": 31,
"total": 373
}
}
While I'm generally bullish on the market. The only stock which I'm probably finding myself thinking more about now is $CRWD I think the stock price and market is reflecting more of of the reality of the strength of the business. Something that I felt like I saw a few years ago. As a result not sure how much upside is left over the next few years, compared to some beaten down names like $META and $HUBS. The opportunity cost of capital is real but so is the cost of selling. I'm thinking of the possibility of reducing my $CRWD position by 20-30% in the next weeks/months.
How do you know if you’re good with AI any more?
Like the AI is so good that I can’t tell if I’m good or if it’s just the AI.
It was 2026 and every few days a guy called @thsottiaux would hit the reset button if we were lucky
So $HUBS is only a $13b market cap? Feels like it punches well above its weight as a brand. Have a small position in it but might look into it more.