An idea Sam Altman expressed in a speech at Stanford in 2014 is my guiding principle today in developing any new system.
-
The principle is simple - product development is built on an iterative cycle. In each cycle, we must check what the customer is missing in the product, implement the improvement, return to the customer for more feedback, and repeat.
If we manage to introduce a 2% improvement in each cycle, Altman said, while shortening each cycle from four weeks to four hours and persisting with this for a few years, we will find ourselves in a completely different place from where we started.
-
This idea is amazing not because it demonstrates the power of persistence and gradual improvement—we all already know those principles. It is amazing because it places the full weight on shortening the improvement cycle to the absolute minimum, and proper implementation of this principle leads to far-reaching results.
How can this be implemented? Here is an example.
On Friday afternoon, I set up an AI customer service chatbot on WhatsApp. It is connected to the products and orders on the website and is subject to strict guidelines to prevent abuse.
Setting up the bot took just a few hours, and I released it into production by Sunday morning, even though it wasn't 100% mature yet.
Alongside the bot, I created an agent whose job is to review all customer conversations with the bot twice a day. The agent analyzes the conversation flow, identifies gaps, and automatically generates a new version of the bot that will operate better and more accurately.
Since this entire process happens automatically, every customer who interacts with the bot improves it almost immediately for the next customer, and the resulting improvement curve is something that cannot be achieved in classic development processes.
I turned this principle into a standard procedure - in every new system, there is an agent that generates new versions of the code based on bug reports and feature requests from users.
But beyond user reports, there is another agent whose job is to scan the system daily, identify gaps, and suggest improvements, and it actually generates these improvements in practice through the reporting system of the first agent.
This process allows me to focus my attention on building the system's infrastructure, because its maintenance and improvement process already happens automatically as its usage grows.
-
This is true for technological systems, but not only those.
Continuous self-improvement will also yield significant results in the long run, but reducing the intervals between improvements will allow us to complete many more improvement cycles, and the result will be better.
The ability to improve rapidly over and over again should not just be a mode of operation, but also part of our worldview and something reflected in the solutions we build.
Nothing needs to be perfect right away; it just needs to be capable of improving in the best and most efficient way possible.
In the image: a screenshot from the new WhatsApp bot. The fact that the bot almost immediately redirects the customer to wait for a human representative is the kind of design flaw that the monitoring agent must identify and fix.
--
👋 Hi, I'm Shlomo Strauss and my posts are not written by AI.
Feel free to follow me for more interesting content on science and technology, or search for 'Shlomo Strauss's Blog' on Google.