You are. Citizens Bank and Worldpay run Grep on decisions a regulator can ask about, at 99.7% accuracy. The launch should say so in ninety seconds. Three films, one story, the first two live by the 14th.
One film from each. The teaser that built the room for Adaptive, the launch film that gave Athena its voice, the product film that showed Mem doing the work. Grep gets all three from one shoot.
→ Athena had to make trust a category. Mem had to make an AI worth opening every morning. Grep's job is harder and better: the agents are already inside banks. Nobody has put that on film.
Parcha became Grep. Compliance agents became a platform for any work that can't be wrong. The post was good. Then the blog went quiet, and the proof went to a page nobody lands on. Citizens Bank, Worldpay, Airwallex, 99.7% accuracy across hundreds of thousands of live decisions, all of it sits on a secondary page while the front door opens with a terminal command.
Meanwhile the category got loud. Demos that work on stage, agents that work in a thread, a new "autonomous" something every week. Nobody is operating agents at scale on tasks with consequences. Grep is. Almost nobody knows.
"Custom AI Agents for High-Stakes Work"
A category label. A hundred companies could put it on a slide. One of them has run agents inside banks for three years, and the line doesn't say which.
"Agents you hire for the work you can't afford to get wrong."
More AWS than Cursor. The film has to feel trustworthy before it uses the word.
Hundreds of thousands of compliance decisions, in production, inside regulated companies. The number that ends the "why not just use Claude" conversation:
accuracy on the work you can't get wrong. It has never been on screen.
Built in the order the audience meets them. The teaser earns the click, the launch film earns the belief, the product film earns the signup. All three cut from the same shoot, the same script, the same story.
One question, no answer. Runs on X and LinkedIn a few days before the launch. Its only job is to make the feed stop and the date matter.
Before · builds the room
The hero. States the problem plainly, sets the landscape, then shows the product on three real tasks: Money, a bank onboarding a customer. Reputation, a fund checking a founder before the wire. Agency, a standing agent catching the lawsuit the morning it's filed. AJ carries the voice and the product carries the proof.
Launch day · carries the story
Product only. Fast, precise, no narration it doesn't need. Describe the task, watch the harness plan it, source it, deliver it. Lands two days after launch as a second beat, then lives on grep.ai for a year.
After · converts the curious
Every film ships with its cuts: an X hero cut, a LinkedIn cut built for the Chief AI Officer, teaser verticals, three product clips, and a stills set. Eight cuts, one shoot, one story.
You said more AWS than Cursor. We took it as a production brief. Scripts, boards, the shoot, motion, score, mix, every cut: all ours to carry. Three rules on top.
This is a sprint. We clear the calendar, you clear the approvals. The teaser and the launch film land on the 14th. The product film follows two days later, which gives the hero the whole week and the product film its own moment. The founder session on Wednesday is the one date that cannot move.
Week of Sep 7 · Story, shoot, build
Forty-five minutes with AJ. The problem, the landscape, the three acts, the line. Product access opens the same day. Scripts start the same afternoon.
All three films scripted and boarded by morning, one round of notes by midday. Real tasks run on the real product and recorded in the afternoon, one per act. Motion and score start from the boards.
Ninety minutes with AJ on camera. Controlled light and sound, multiple takes, the script already in his head from Monday.
Teaser final and ready to schedule. Launch film first assembly with motion and score in place.
First cut in the morning with motion and score in place. Your notes by end of day. One round, then picture lock.
Weekend · Finish
Launch film finished. X cut, LinkedIn cut, teaser verticals, stills. Teaser goes out Sunday. Product film first assembly in parallel.
Teaser and launch film on grep.ai and the YC page, cuts scheduled, stills in the post, the day the Fall batch kicks off. We're on call through the day. Product film first cut to you the same afternoon.
Notes Monday, picture lock Tuesday, live Wednesday with its three clips. A second beat in the feed two days after the first.
Seven days to the launch film leaves no buffer, so the two approval points are the whole plan: boards Tuesday midday, first cut Friday. The product film on the 16th is the release valve. If anything slips, it moves, and the 14th holds.
Concept, scripts, boards, shoot, motion, post, delivery: all ours to carry. What we need from Grep is direction and access, not production time.
Three films, eight cuts, the stills, and every step from script to launch day. One shoot in San Francisco. No line items.
Less than one month of one enterprise contract. It has to help close one.
AJ, we built this page the way we'd build the film. Plain and precise, product first. If it feels right, text me yes. The deposit invoice and the Monday calendar hold follow within the hour. Half to book, half on delivery.
Questions first? Twenty minutes · Not on a phone? Email the same yes
-Rob