We follow our intuition in guiding agents through Harness engineering. Our intuition is based on the things we have done to boost our teams and our own productivity.
The way AI and humans build software has overlap. We should aim to start to build the harness that models our own development processes. The future will tell us how to be better, but this is where we can start right now.
If it helped us, it should help the agents. Whoever you are, treat yourself as the Exemplar that your models should follow.
What?
Every single person should consider themselves an exemplar for the AI to follow. Even if we aren’t strong or experienced, this is what we know right now. We can’t defer to the AI and think that it knows better than us. If LLMs get to the point where they are capable of higher order thinking than we are, then we won’t be building a harness for them. So long as we are building a harness, we should always assume that we know what is right and be skeptical of what LLMs produce.
Establish confidence in ourselves, be humble. “We are the exemplars.”
Keep ego in check
All of the experiences and successes I have had over the years give me confidence in the way I approach software development and writing code. My workflow and my practice is the exemplar I want AI agents to follow. This isn’t because I am the GOAT, it’s because I truly do believe a harness modeled after the things I have done and the things the teams I’ve worked with have done is going to do well.
I’ve always tried to push the bar a little bit higher. I’ve tried to inspire engineers to grow, to challenge them, and to show them a little bit more of what is possible. I’m obsessed with DX & developer productivity.
These are the traits that I feel strongly about.
The traits that you feel strongly about should be the ones you focus on as you are progressing along this journey into your AI-assisted workflow.
- Have confidence in yourself and what you do and push for your ideas to be considered
- A diverse team with different individual core tenets should produce a more diverse and capable harness
- There are other people who are more intelligent than we are in many areas, use them as the exemplars if you’d like
Why we are the exemplars
The two most important reasons that we should model our harness after ourselves:
- We cannot describe what we do not know, so we cannot instruct the AI to base their work off of something we do not understand
- We should not pretend to believe we can instruct the AI to work at a level above us, as that would imply the AI knows better than we do (and thus we are unnecessary)
Everybody should follow this philosophy and have confidence in how they work. If you are a more junior developer, to the agent, you are still the exemplar.
If you want the agent to do better, level up yourself: improve yourself => improve the exemplar => improve the agent.