#128: Scaling Paid Ads Programs, How Anthropic Lead Engineers Prompt, What Comes After Newsletters

Hi All!

We are pleased to welcome you to this week's edition of Undiscovered, a newsletter with exclusive resources and insights expanding from the material found on our main site - becketu.com.

This edition is brought to you by 1440 - the knowledge company for the insatiably curious. Be sure to check out their exclusive offer in the P.S.

This week we will take a look at how a lead engineer from Anthropic optimizes their prompts, what comes after newsletters, how to build a paid ads program, and more.

Let's dive in:

How Anthropic Lead Engineer Prompts

​Felix Rieseburg is a Lead Engineer at Anthropic, who recently shared an insightful essay titled How to be more ambitious: A tutorial for humans with models.

In order to be more ambitious, he claims we must think along four dimensions -
​
1. Speed: The same work in less time, the bread and butter for AI’s economic value: If you need some code written, (models will do it faster). Well established and requires little creativity in how you approach AI
2. Altitude: Going up in abstraction and considering the outcome rather than the individual task.
3. Depth: Considering a task the way a field-leading expert would. Rather than operating at the task’s layer, we go one level deeper and touch the underlying data, tools, and problems.
4. Range: Drawing on multiple fields rather than just the one the task might need at the surface, using knowledge, tools, and techniques you normally wouldn’t.

One of the biggest takeaways you can immediately start implementing in your workflow today, is a prompt he uses to accomplish more 'depth' in his work:

Newsletters Will Die, What Comes Next?

​Dan Shipper, whose company Every runs a newsletter with 120,000 subscribers (as its core business), just declared "the newsletter is dead". His reasoning is that persistent AI agents now filter his inbox so completely that he almost never looks at it, and within a year he expects to do all of his email through his agent.

Having someone's email address no longer guarantees a relationship, because the agent is now the gatekeeper deciding what gets seen. The new relationship is with the agent itself. So what's next?

His move is to get installed in the reader's agent harness of choice, so the agent notifies them when you publish instead of competing for space in an inbox they never open.

The deeper shift is that it is now cheap enough for any creator to build their own app and feed, so the winners will own the channel the agent lives in rather than rent attention in someone else's inbox.

How to Test Creatives with Low Budget

Marco (@maxzrco) laid out a clean fix for a problem every small advertiser hits.

You put ten creatives into one ad set, the algorithm dumps 90% of the budget into one of them, and the other nine barely get impressions, so you wrongly conclude they are losers when they never got a fair shot.

His structure is one campaign with two ad sets, a testing ad set where every new creative goes first and a scaling ad set where proven winners live. You load the testing ad set with at least ten creatives, then watch it and pause each one the moment it hits around 3,000 impressions, which forces the algorithm to redistribute budget to the next one. You repeat until every creative has been measured at the same impression count, which gives you real comparability because every creative actually got budget.

Judge them on CTR, not CPA, since the conversion data is too thin at this budget. Weak overall CTR means your hook or visual failed to stop the scroll, and strong overall CTR but weak destination CTR means you stopped them and then lost them, which is a messaging problem. The winners graduate into the scaling ad set, and new creatives always enter through testing.

A Lesson in Perseverance

​KAHALA shared a story about his son that quickly became one of my favorite X anecdotes in recent memory. His son tried out for his high school basketball team three years in a row and got cut every time. He is five foot six and not very fast, and the third cut broke his heart because he had practiced every single day that summer.

The morning after the final cut, the coach called and offered him a job as the team's statistician, because he had noticed his son was the only kid at tryouts who asked about defensive rotations. His son said yes before the coach finished the sentence, went to every practice and game, and sat at the end of the bench tracking everything. Shots from every spot on the floor, who played best together, which defenses worked against which teams. By midseason the coach was asking for his opinion at halftime, and by the playoffs he was building game plans around his reports. The team went further than they had in fourteen years.

At the end of season banquet, the coach gave a new award to the person who had changed how he coached basketball, and called his son's name. His son is now a sophomore studying sports analytics on a scholarship, and last month got an internship offer from a professional team. When he called his old coach to say thanks, the coach said he had seen it from the first day of tryouts. His son asked if that is why he cut him. The coach said, "No. You were just really slow."

A funny ending to a story with a very strong message!

How to Build a Paid Ads Program from Scratch

​Grant Miller broke down Hector Serrano's viral 'how to build a paid ads program' video in to 15 steps:

1. Pay on performance, not upfront. Creators earn a % of GMV for ecom or a % of ad spend for SaaS. Wispr Flow pays 7% of ad spend.
2. Set your % based on your CAC. Too high blows up your ad account, too low and creators stop caring.
3. Add a smart retainer. A guaranteed minimum you top up if commissions fall short. Once they out-earn it, they are on pure commission.
4. Use the retainer to recruit early. With no reviews or payout history, it is the reason top creators will bet on you.
5. Start with 4-5 creators, not hundreds. Mercenary creators make thousands of identical talking-head videos.
6. Recruit for gaps in your ad library, new formats, new ICPs, new regions, not just volume.
7. Find creators off-platform. IG keyword search, then dig through their followers and comments. Send 100-200 DMs over a week.
8. Treat creators like a sales team. Coaches, weekly calls with your media buyer, open access to your calendar.
9. Run giveaways and bonuses they actually value. Cash, camera gear, mics. Bonus, better gear means better ads.
10. Fly your top creators in for a retreat. Face-to-face builds loyalty that calls cannot.
11. Scale slowly. 5 to 10 to 20 to 40. Wispr went from 5 to 50 active creators in 4 months.
12. Cap around 40-60 and only add killers after that. Smaller groups feel more exclusive.
13. Pay a $100 referral bonus for creators at or above their level. Good creators know good creators.
14. Ban copycats. When creators replicate the top performers, Meta clusters near-duplicate videos and suppresses reach.
15. Push new concepts constantly. The leaderboard winners got there by experimenting, not by remaking someone else's hit.

P.S Can you please respond to this email and bring it into your 'primary' inbox? You can say 'Hi!', tell us the last book you read recently, or what your favorite resource was from above.
​
It would also help if you add us as a contact on your mailing list, with a visual tutorial below:

We appreciate any feedback you are able to provide here. What do you want more or less of? Other suggestions? Feel free to reach out to us on Instagram and give us a follow there, tag your friends on our posts, and please forward this newsletter along to anyone else who would enjoy it.

Disclaimer: Becket U is an Amazon Associate and purchases through Amazon links may earn a small affiliate commission, but the price is the same for you. We only recommend books we love and think you would love, too.

Always wishing you the best,

J.B.