Engineers will need to think more like scientists to stay valuable in the age of AI.
Software engineers will need to learn more management skills to manage an army of AI agents with different tasks. The highest-leverage management will involve either
- great taste (making a good judgment based on intuition developed outside of the job) or
- great science (making a good prediction based on data collected on the job)
Developing scientific skills is one way new engineers with little experience can build great taste.
Thinking like a scientist includes:
- having a target metric to improve (a product or business KPI)
- having test data and validation data
- A/B testing new prompts and techniques
- thinking about edge cases and pulling in context that AI doesn't have yet
- measuring results
Humans might be better than AI at this for a while. They might not be able to process context nearly as quickly, but they'll be able to acquire context from non-digital places, e.g. the physical world. And while bots won’t be able to see this new context, humans will always be able to see the context that bots use themselves. Thus, humans will have a structural advantage.
Context acquired in non-digital sources will become extremely valuable and powerful when applied intelligently. It will be scarce. There will even be value in just feeding the bots new context without thinking scientifically at all.
The most valuable way to use this context will be towards changing the structure of investigations being done by the organization as a whole.
Will AI be able to do science of its own? Yes. They’ll be able to test new prompts on new models and evaluating them on new workflows. They’ll be able to field customer calls and gather new data.
But a human who ventures into a new data source without informing the bots will have a small advantage.