53 chunks · 43 episodes
You need your data with you in your agentic loop. Your agentic loop is where all the value is generated. The more context and personal data in the loop with you, the better it can assist you. That compounding loop is so powerful that users won't brook not having it.
When energy aligns it creates emergent compounding value. "Divide and conquer" works because collectives have emergent power that is super-linear. If you break up one thing into fewer with the same mass you...
Jevon's paradox happens when elasticity of demand is extremely high. That is, when latent demand is significantly higher than realized demand. As cost declines, demand rises at a significant rate. LLMs have a compounding rate, because you can use tokens to create tools that consume tokens. This then
...to be 10% better. How do you take advantage of abundant cognitive labor to make compounding value? AGI will come not from the models but from the emergent use of them.
As a system gets increasingly cacophonous you get more superficial takes. That creates more cacophony. A compounding loop.
Actions that improve the worst case scenario have compounding value. You lock in a new worst case once, and then now all future instances have it for free: a compounding term. If you can then make it crowdsourced (loc...
It's easier to take a small thing that is known to work and extend it to generalize it. It's like pulling taffy. At every point, it exists and is viable. As long as you pull it carefully and consistently, it can expand. Compare that to starting with a theory and trying to build the real thing. The t
Things that are unstoppable start off as unstartable too. The trick is the thing that can be startable and become unstoppable. That's where compounding loops come in. A self-accelerating thing.
To get to a quality loop that learns from people's actions it has to be useful enough to actually be in their loop. That's very hard to do! A quality loop that is on the side can't ever get going. Typically you have to do it with a different, more quotidian primary use case, and develop the quality
Ask Claude Code: "What, if you had known it when you started, would have saved you time?" Then have it make those changes to the documentation. This is the key compounding loop in Compounding Engineering[gu]. This naturally accretes useful insights and smooths things down, in each loop. Things that
LLMs assume everything that happened before in the conversation made sense and try to keep it going. This is because they are excellent retconners. At every time step they have to figure out what token to output based on making the most sense of everything that came before. Naturally convergent; the
Vibecoding on an already healthy codebase does a good job at keeping it working. If it's a crappy codebase it makes it worse and worse at a compounding rate. AI is an amplifier.
A "yawning gap" happens when the two things are diverging at a compounding rate.
Resonant products have faster compounding loops. Not only do you like it, but you're proud to recommend it to friends.
100x Bot is doing something interesting. Seems like a combination of: the Skills / Learnings.md compounding loop Crowd-sourcing driving AI browsers. A catastrophically powerful combination. This kind of looks like RL if you squint. RL researchers might say this is an under-powered hack to get someth
A key approach for agent learning: accumulating insights in a LEARNINGS.md.[jq] After it does a task, have it distill insights that it gained that would have made this run easier. This makes the next run faster and higher quality. This is where the feedback loop closes and becomes a meta, compoundin
Data is not like oil, but sand. That is, individual grains of it are not valuable, but a large collection of it in one place is. Data gets more value the more it's aggregated, at a compounding rate. Both for individual users and for collections of users. This is why the aggregators have such large n
The inductive logic of folksonomies is not "from a standstill is this item good?" It's "given that others liked this, do you think it's good enough?" The "given that others liked this" is the compounding loop, a collective intelligence feedback loop. It allows people with little effort to go "... ye
Things that take a long time are impossible to predict. You'll mispredict how long it will take, and you won't get feedback until the end. Your estimates will get inaccurate at a compounding rate. Slice it into small slices that you can get feedback and learn quicker.
A phenomena of unfolding, where each interaction makes the thing increase in size, at a compounding rate. Positive version: blossom. Negative version: metastasize.