I have noticed that a query planner that’s just okay, but rock solid in how it runs often outperforms a much fancier planner that’s built on a shaky or fragmented foundation.
The benefits of stable execution really add up over time.
Planner cleverness often does not.
I used to think planner sophistication was the main determinant of query performance.
Increasingly I think execution predictability matters more.
A stable execution pipeline with fewer surprises often beats theoretically smarter optimization layered over fragmented execution.
Most vector databases are actually two databases duct taped together.
One for retrieval. One for relations.
Then they are surprised when joins collapse, cache locality breaks, and planner behavior degrades.
Hybrid retrieval shouldn't fork execution architecture.
Compiler runtimes and database engines end up converging on similar problems surprisingly often:
predictable execution
memory locality
representation stability
materialization costs
execution path discipline
Different domains.
Same logic.
Want an easy way to monetise MVPs?
Here’s a pattern that’s both low effort and high perceived value for the user:
When making an offer, one of the most important concepts to keep in mind is value stacking — the more stuff you add for a price point, the more likely you are to get the sale
This is why info marketers include 30 different things with their course, why you get 3 replacement heads whenever you buy an electric toothbrush, and why you get all kinds of random bonuses when you purchase a collector’s edition game
This is a very powerful tactic, but it takes a lot of effort to implement. If you want to offer 10 different things to your app’s offering, you have to build 10 different things to offer. This is in opposition to the core principle of shipping the MVP as fast as possible…
Except it doesn’t have to be
Here’s how you can create a 1-feature value stack
First, offer the core feature for free. You do this in order to increase engagement time
Then, offer customisations for the app — skins, personalities, pets… anything that can be added quickly, without much development cost
Each of these collectibles comes at a fixed price point, let’s say between 1,99 — 9,99
This is essentially the freemium model games use. Think of how TF2 is free, but hats are paid
Now, let me add an extra layer to this:
Add a subscription (weekly works best) that unlocks all collectibles for the duration of the sub
If you price the sub at 4,99/w, it becomes a no-brainer (since one individual item is up to 9,99)
It’s an incredibly low effort way to value stack
Over time, you may want to change this model (since micro transactions can be a huge source of revenue), but as a starter it’s a great monetisation model
Scalable features can wait. 🛠️ What matters at launch is delivering core value to your early users. You can scale up once you validate your product-market fit. #indiehackers
The Greatest Storyteller of the 21st Century:
Robert Greene.
He's written 6 international best sellers and mentored Ryan Holiday.
Here're 10 of his best ideas to make your writing more powerful:
#buildinpublic on Twitter is not an effective way to get customers.
Don’t build in public here on Twitter if you need users/customers and your project is not in b2b else you will spend a large majority of your time focusing on the wrong target audience.