I cancelled our ChatGPT subscription


Hi Reader -

Last quarter, I cancelled our company’s ChatGPT subscription.

I cancelled it because we’re betting that the next 12 months of AI won’t be won by people with better AI tools (or prompts).

They’ll be won by the teams that own their entire workflow. By people who know how to plug AI into their actual work: their inbox, their docs, their calendar, eventually their codebase - and not just use off-the-shelf tools for everything.

Let me rewind, because the edge keeps moving really fast.

**“My dear, here we must run as fast as we can, just to stay in place. And if you wish to go anywhere you must run twice as fast as that.”
- The Red Queen in Through the looking glass, Lewis Carroll

2023-24: the edge was the PROMPT. Whoever could whisper the magic words to the machine won. For about a year we all pretended prompt engineering was a real career.

2024-today: the edge became the TOOL. The wrapper of the week. They all say 'agent' while all they mean is that they're using other people's tools.

What we’re betting on is that the next wave of AI goes to the teams that actually build their own workflows - and doesnt ride on someone else’s tools.

And here’s the part I actually care about.

This isn’t one clever system I built. We’ve had those for more than a year now: BruteForce, Mission Control, the creative agents we open-sourced.

The thing I’m most proud of is that since we moved off ChatGPT, our WHOLE team started building.

Not the engineers. There are none.

These are growth and creative people who’d never opened a terminal until a month or so ago.

Every week we run a show-and-tell. Everyone brings one thing they built - and here is a selection of what came up just in the last 2 weeks:

  • A live dashboard that pulls a client’s complete metrics out of an MMP and refreshes itself every morning.
  • A frame-by-frame teardown of a client’s top ads, pulled straight off Meta, that caught the thing most teams miss: the creatives that won in testing were not always the ones converting to paying players at scale. Turned into a full root cause analysis - leading to a build-this-not-that deck for our creative strategy and design teams.
  • 90 days of data across twenty-three test batches, turned into an in-depth strategy deck with every winning ad hyperlinked into the slides.
  • A creative performance dashboard that reads ad names straight off Meta, matches each one back to its brief, and scores which batches and angles are actually hitting CPA.
  • A capacity-planning agent that maps the whole design team’s month, how many tasks each designer can take per client based on real hours-per-task, and flags exactly where we tip past capacity. It replaced a spreadsheet someone used to rebuild by hand every month.
  • A tracker that watches our past clients for activity signals and builds the week’s outreach list on its own.
  • A terminal wired into ClickUp that pulls every creative up for review and runs the first pass on its own.
    ....and a LOT LOT more.
    ​

Could our team have done all this on ChatGPT? Probably. With Codex, it isnt impossible.

The big reason we moved off ChatGPT was that Claude had a cleaner on-ramp to building with Claude Code(and Claude Desktop’s connectors). Plus there was something symbolic about a clean slate that gave people permission to try something they had never done before.

HUGE props to our team for taking the leap - and embracing the building(and only giving me a bit of grief for getting them away from their formerly beloved GPT ;)).

Anywayyyy: the divide was never Claude versus ChatGPT. It’s chat versus build.

A chatbot keeps you a user. Claude turned our team into builders.

The companies that win the next year are not the ones jumping on the latest ‘tool’ - they are the ones enabling their whole team to build(note how I didnt say ‘teach’ their team - it's more about set things up and get out of the way).

Anywayyy, if you’re spending $50k+/month and you’d rather hand your creative/growth to a team that builds from first principles than jump on others’ tools, reply “BUILD” and we’ll talk.

Or don’t. I’m sure the tool du jour is exactly what will crush your CACs in Q3 and Q4.

Later,

Shamanth


PS: Here is our internal memo(with a migration plan) that I shared with our team - that may inspire you in how to communicate a change like this to your team.

... and some links from our socials in the meantime:

​I built a Claude Fable 5 agent that runs the deepest Meta audit you'll ever encounter.​
​
​I built our AI creative strategy team using Claude. Steal it today.​
​
​I just analyzed 100 Remini AI Photo Enhancer ads with AI.​

​I just analyzed 53 Pok Pok ads with AI.​
​

Welcome to Intelligent Artifice, your weekly dose of insights to help you win with performance marketing in a Generative AI world.

Intelligent Artifice features hard won insights from the front-lines of performance marketing in a Generative AI world - delivered straight into your inbox. For free.

Read more from Welcome to Intelligent Artifice, your weekly dose of insights to help you win with performance marketing in a Generative AI world.

Hi Reader - "It is remarkable how much long-term advantage people like us have gotten by trying to be consistently not stupid instead of trying to be very intelligent." That's Charlie Munger, Warren Buffett's partner. Everyone in advertising is hunting the whale (or unicorn, or outlier - pick your metaphor of choice). You know the one. The ad that eats your budget, makes your year, and gets you high-fiving your CFO. Soooo every quarter somebody uncaps a marker and says "what's our big swing?"...

Hi Reader - There's a robot in The Hitchhiker's Guide to the Galaxy called Marvin. Full name: Marvin the Paranoid Android. He's the smartest thing in the universe, and he's miserable about it. In one scene, he tells a mattress he's thirty billion times smarter than it is. To prove it, he runs a little test. "Think of a number, any number." "Er, five." "Wrong. You see?" The mattress comes away very impressed. I think about that scene a lot when we use AI to analyze ads. Because the analysis...

Hi Reader - Years ago, I was working on a word game, and we were killing it in the US. So we ran an experiment in the UK. After all, they speak English too, right? I think it was George Bernard Shaw who said: England and America are two countries divided by a common language. (Whoever it was that said it, I was about to prove it.) Anywayyy for our word game, the UK did even better than the US. CPI was strong. Cost per first puzzle played was strong. D1 ROAS was extremely strong, and D7 held...