ENZO

turning people into high-level operators

[KNOX @knoxtwts]

AI filmmaking right now is like discovering Bitcoin in 2009… 💀

Higgsfield just made every prompt and asset behind Hell Grind public: their $500,000 feature film that screened at Cannes Market.

anyone can copy the exact techniques behind a press-covered AI movie and enter the $1M Higgsfield film festival with them.

you missed dropshipping in 2016
you missed meme coins in 2022

don't miss AI filmmaking in 2026… 💰

[ENZO @enzoprints_]

@knoxtwts @keetonny where do i sign up

[KNOX @knoxtwts]

so you're telling me the co-founder of Pixar...

- 5 Oscars
- a Turing Award
- ran Walt Disney Animation Studios

...is now judging a $1M contest for films made by people who've never touched a camera?

people said computer animation wasn't real filmmaking in 1995 too. he's seen this movie before.

[KNOX @knoxtwts]

hackers pulling up to a vibecoded bolt app after the security agent already patched everything

(it's free btw)

[ENZO @enzoprints_]

@knoxtwts crazyy

[KNOX @knoxtwts]

Seedance 2.5 just crossed the line where you can't tell anymore. vlog-style videos are where it's scary good.

no crew, no camera, no actress. one generation, & it can now...

> hold a 30 second shot in one pass & the character doesn't drift
> keep the same face through every scene & lighting change
> nail the human stuff: pores, hair, the tiny involuntary movements
> take 50 inputs so your cast, props & style refs all travel together

watch it twice & try to catch the moment it stops being real. i couldn't find it.

Seedance 2.5 & it's coming to Higgsfield soon

[KNOX @knoxtwts]

i found a way to run a full agency-grade marketing audit on any website in 20 minutes...

example below...

agencies charge four figures for this and take two weeks to deliver. most of that time is a junior filling out a template

the actual audit is 6 questions asked in the right order. so i turned the one we run on real clients into 6 claude prompts

it reads your page like a stranger with money, finds the exact line where the message loses them, and hands back the one fix to make first

the full prompt stack is in a free guide, every prompt paste-ready

like + comment "AUDIT" and i'll send over in dm's

(following + rt for priority access)

[KNOX @knoxtwts]

https://t.co/rB9EOgDeNL

[KNOX @knoxtwts]

your marketing department can now be a folder of text files

this is the part people don't believe until they see it

voice rules, offers, winning hooks, each client's context.. written down as plain files. claude reads the folder & becomes the department. new campaign, it already knows the voice. new client, it reads their file & starts the day briefed

the wild part: the folder compounds. every correction i make gets written back in. the department gets smarter every week & it can't quit & take the knowledge with it

most businesses have that same knowledge scattered across six brains & zero files

the starter version is five files & an afternoon:

→ voice.md.. how you talk, words you'd never use, three examples of you at your best
→ customers.md.. who buys, the exact words they use for their problem
��� offers.md.. what you sell, the price, the promise, the proof
→ winners.md.. every hook, ad & subject line that has ever worked
→ lessons.md.. every mistake you don't want repeated

paste the folder into claude before every marketing task & feel the difference between briefed & guessing

six months from now that folder is the most valuable thing your business owns that isn't on the balance sheet

[KNOX @knoxtwts]

so you're telling me your whole team can now...

> dump the design system, brand voice & seo rules into skill files
> prompt Bolt like normal
> have it auto-pick & stack the right ones on every single build
> get everything out the door to the same standard

without anyone invoking anything?

we're not going back

[KNOX @knoxtwts]

token prices dropped all year and somehow every enterprise ai bill went up. nobody finds this weird?

chamath says token costs are doubling every 45 days for a 5% productivity lift btw…

the glean founder’s take is that everyone’s measuring the wrong number. cost per token means nothing. cost per successful task is the metric. a cheap model that produces rework isnt cheap, you just pay for it twice

[KNOX @knoxtwts]

you ask an AI for deep research, hit go, and come back an hour later hoping it actually finished.

Octen just ran one of these, 129 searches into a full sourced report, in 2 minutes 35 seconds.

and it scored higher than OpenAI, Gemini, Grok and Perplexity on the benchmark, not lower.

we all quietly accepted that good research means waiting.

turns out we were just waiting on slow search.

[KNOX @knoxtwts]

Ramp pulled the token bills of 1,000+ companies and almost everyone is reading the wrong number off it…

the teams overpaying the worst were barely touching frontier models AT ALL

> inefficient teams: 48% of spend on frontier
> efficient teams: 51%
> the inefficient ones still pay 4.15x more per million tokens

the separator was cache reuse. 28% hit rate against 88%

so every team downgrading to a cheaper model to fix the bill is solving the wrong problem...

the same context gets re-sent uncached on every run either way

what that actually means: you cannot shop your way out of this. the bill only moves when you can see what each individual request cost you

most teams are flying blind on the fastest growing line item they have

[ENZO @enzoprints_]

@knoxtwts sheesh

[KNOX @knoxtwts]

the most interesting job signal i've seen this year is one company recruiting substation electricians and backend engineers on the same careers page.

anthropic committed $50B to compute buildout. fluidstack is the one actually building it, and they are hiring across basically every function to get it done.

situational awareness, blackrock and google all sitting on the same cap table is not a combination that lines up by accident.

if you have been waiting for the part of ai where the work is physical and the deadline is real, this is that.

the interesting infra jobs stopped being at the labs.

[KNOX @knoxtwts]

i found a way to steal the structure of any winning meta ad...

example below...

the ad library shows every ad your competitors run and how long it's been live. an ad that's been printing for months is a proven structure sitting there for free

everyone screenshots it and copies the words. wrong move. the words belong to their offer, and meta clocks near-copies anyway

the structure is what transfers. so i built a claude workflow that pulls out the skeleton, locks the character across 9 pillars, and writes the prompt that re-shoots the same structure with an ai actor for your offer

in the video: left is an ad running since june, right is the rebuild. nobody wrote a script

the full workflow is in a free guide, every prompt paste-ready

like + comment "SWIPE" and i'll send over in dm's

(following + rt for priority access)

[KNOX @knoxtwts]

so ai in hospitals now works like this:

> doctor talks to the patient, nothing else
> the notes, referrals and admin draft themselves in the background
> doctor reviews, signs, done
> 2.5 million patient visits a week already run on this

the most regulated industry on earth figured out background agents before most marketing teams did.

damn.

[KNOX @knoxtwts]

https://t.co/5SYgA0dC0S

[KNOX @knoxtwts]

pov: anthropic to the government

[KNOX @knoxtwts]

i default to ignoring anything labeled "legal ai" because it's almost always a gpt wrapper that reads one contract and hands back a summary.

this one reads every contract your company has already signed -> then scores the deal in front of you against your own history.

shows you which clauses your team has quietly caved on before and which ones you always hold.

so the review gets sharper every time you close a deal. vertical ai that actually builds its own memory instead of renting one from openai.

been around this space a while. this is a real one.