Small Batches

TL;DR - Small Batches are easier to understand, faster to test e2e, and harder for mistakes to sneak into. Small Batches at scale lead to consistent iterations.

Overview

At hyper scale large batches can make sense.

• You get huge rewards from even small % improvements at scale
• You have the capital to expend large amounts up front to experiment and get the right process done
• You can work at the time scale of years to achieve these wins

For most other cases, small batches are better.

• You see e2e outcomes faster - which helps you determine if you're on the right path
• You expend less upfront investment to see these outcomes - so you waste less if you're wrong (you usually will be)
• You can course correct faster and more often
• It's easier for others to understand the full change - for feedback etc (like a code review)

Examples

Atomic Commits - Small Batches for faster, better software

An atomic commit is a change ("git commit") that contains one logical change. You can string these together into stacked commits to build larger features out of atomic commits.

The idea of atomic commits is:

• Software at scale is hard to understand, even seemingly innocuous changes can have far-reaching problems - at a probability that grows exponentially with the codebase (at some scale, no human can understand all the code)
• Big changes have more room for error, are harder to review, and generally take longer to get approvals on
• Smaller changes have less room for error, are easier to review, and typically get more thorough reviews (less LGTM!)
• Shipping in atomic commits leads to faster e2e lands and higher quality changes

Small Bets - Small Batches for faster iteration to PMF (Product Market Fit)

Small Bets are Minimum Viable Experiments (MVEs) of a business. You get an idea, ship it in a week, and see if the market buys it. If no one buys (which is likely) then you move on to the next without expending too many resources. If it works, you've got the seed of a good business to build up on.

With small bets you can build biz faster, build on the side of your main gig, and parallelize them.

The idea of Small Bets is:

• We don't know what's going to work in the market
• The only way to know is to test in the market
• So ship small and fast to test many ideas in the market
• When you find one, double down

Small Bets was coined by Daniel Vassallo.

Science Experiments - Aim to gain knowledge by testing the minimum possible change in input to output

If you think ab it, a science gains knowledge through experiments which themselves are small batches. A good science experiment gains knowledge by minimizing the input variables changed to better determine if the input change actually leads to the output change.