Thoughts For The Week - 2025.08.24
Development Process Optimisation, AI for learning and the cost of AI
1. Product Development Processes and Time Allocation:
John Cutler has a ton of insights, I usually get the feeling he’s sitting over my shoulder and spying on what I’m doing. He’s now going through a list of his top 30 doodles. My two favourites this week are:
When you’re thinking about optimising your development process then these provide a great baseline and way to think about responsibilities in terms of type of work and time allocation.
2. How to stop AI killing our ability to learn
Eva Keiffenheim gives some interesting perspectives in to making sure you can use AI to learn more effectively and avoid fully outsourcing our thinking to devices that undermines three core processes of deep learning: Automaticity, Schema Construction and Prediction Error.
A key suggestion is…
Before you use AI to help write a report, write the first draft yourself. Before you use an AI code generator, manually write and debug a small algorithm. Build your cognitive muscles in a safe environment first. Then bring in the power tools to accelerate your growth.
“Every time you face a gap in your knowledge, you make a choice. You can make a short-term withdrawal for an immediate answer, or you can make a long-term investment by doing the work to internalize the knowledge.
The first option feels efficient. The second is what builds your cognitive capital: the interconnected mental library that allows for creative leaps and resilient problem-solving.“
A Mind Without Knowledge Is a Body Without Muscle
Thinking works in exactly the same way. When you truly learn a concept, whether it's a math equation or a historical argument, you are bundling it into a compact chunk. This is the goal of all effective learning: to create a vast mental library of these chunks that you can access instantly. This is what frees your working memory for higher-order thought: synthesis, creativity, and strategic insight.
Constant cognitive offloading, through Googling, a “second digital brain,” and now ChatGPT, short-circuits this process. We get stuck at "knowing about" a topic, never reaching the automaticity of "knowing how."
3. The Cost of AI and who is paying
There’ll be some debate on the accuracy of the numbers but it’s clear there’s currently a lot VC funding which is subsiding AI use at multiple levels.
Agents caused in instant change in usage demand which breaks the financial models and business cases of a lot of companies. But as long as some people are willing the support this others have to pay up to stay in the game.
users became api orchestrators running 24/7 code transformation engines on anthropic's dime. the evolution from chat to agent happened overnight. 1000x increase in consumption. phase transition, not gradual change.
so anthropic rolled back unlimited. they could've tried $2000/month, but the lesson isn't that they didn't charge enough, it’s that there’s no way to offer unlimited usage in this new world under any subscription model.
Have a great week!