
head of digital technology https://xitter.cf/t.co/Sumk8ovRGK \\š¦ @ugcpro_ co-founder \\
raea.io
Joined December 2021
- Tweets352
- Following211
- Followers14K
- Likes746
Pinned Tweet
Been training my desktop assistant to vibe and understand music in realtime.
Uses vision and audio towers fused to single realtime neural codec.
āWatch Modeā allows it to actively watch, help and take notes while you work, jam, or vibe (code).
Shane retweeted
If you needed 50 new ads next month, would you actually be testing 50 different strategic bets... or just 50 variations of the same thing? š
Because producing more creative doesn't automatically mean you're learning more from it. And that's one of the conversations I'm bringing to TikTok's NYC office next week! If you're in the city... checkout the event details below. š„³
Shane retweeted
And one day, if youāre lucky enough, youāll end up working with the love of your life not once, not twice, but yet another time (truthfully, Iāve lost count). š
Nearly 8 years ago, Shane Murphy and I met as direct response video editors at Mutesix in Culver City, California.
Fast forward to today, and weāre joining ContentJet as Head of Performance Creative and Head of Digital Technology⦠yes, you heard that correctly! We are leading our own parts of the business, TOGETHER. š„¹
And to think it all started with me asking the internet to help me find my fiancĆ©ās ābetter halfā in business. š Within minutes, the video went viral and the internet ironically reminded me that weāve always complemented each other in business too.
Apparently the algorithm gods decided they owed us one, because the Reel was served to the ContentJet team, they watched it, and made the first move. (Cassandra Kranjec, you rock! š)
So how will this marketing + machine learning duo work together in partnership with ContentJet? šāØš¤
The challenge with scaling paid media isnāt just spending more, itās building a creative engine that can keep producing enough of the right ideas, tests and iterations to support that spend without sacrificing quality.
My job is bringing a decade of tested performance creative strategy to ContentJet, shaped across thousands of accounts in nearly every vertical you can imagine, hundreds of millions in ad spend and billions in revenue, then turning that experience into the systems, testing plans and strategy our team uses to scale clients without sacrificing quality for volume.
Meanwhile, Shane is personally building the technology underneath all of it, creating the systems, infrastructure and machine learning capabilities that allow our creative engine to keep up as the volume scales. š¤
Eight years ago, we didnāt know it, but we were just getting started. Now we get to build what comes next, side by side, once again. @shinyobjectz_ #ContentJet
Claude started watermarking generations at the same time I noticed it now always say ācolourā not ācolorā.
Even in coding contexts.
Unless they are really just leaning into the British brand.
Marketing AI companies are selling you a lie, and it's usually hiding in the "creative intelligence."
Open source? Who verified the data.
Never read the spec docs? Then you don't know if it's good or if it just looks good.
80% of teams I talk to make the same mistake.
If you arenāt building with an ontology then you are likely wasting time in Claude Code and Codex.
Weāve been building skills and memory systems all wrongā¦
Check out the repo and see for yourself.
github.com/shinyobjectz/montā¦
Shane retweeted
my coding agents kept renaming my own concepts back to me. a "thread" becomes a "session", a settled idea gets reinvented under a new name, every css file grows another slightly different gray. none of it is wrong exactly. it just stops meaning anything.
github.com/socialite-ml/montā¦
my coding agents kept renaming my own concepts back to me. a "thread" becomes a "session", a settled idea gets reinvented under a new name, every css file grows another slightly different gray. none of it is wrong exactly. it just stops meaning anything.
github.com/socialite-ml/montā¦
monty vitals rolls everything up. each repo gets a verdict, tended, drifting, or untended, and every reason comes with the command that fixes it. i validated it against eight real repos in eight languages. flask reads tended. most repos will not.
the pattern underneath: flask's concept vocabulary is 49 words after 15 years, flat since 2019, because humans tend it constantly. agents make that kind of tending impossible by hand, so i made it automatic.
open source, local, apache 2.0
github.com/socialite-ml/montā¦
If you arenāt building ai products that start at the model layer or the infrastructure layer, you are building a toy.
Thereās no defensibility, just vibes.
But if you have the distribution, idk I guess you can do anything.


