# The Key to AI Adoption is Encoding Judgment
*Published: 2026-07-18*
*Tags: ai, encoding-expertise, agents, for-executives*
*Source: https://chrislema.com/encoding-judgment*
---When talking with execs and business owners on AI, I regularly refer to “encoded judgment”. 

The only way you get to where you want to be, when it comes to adopting AI, is when you can let it run!

But letting it run wild freaks everyone out. And I get it. 

The only answers are either putting yourself back in control, putting AI on a leash where you’re the human in the loop, checking each step.  

Or, alternatively, you coding your own decision making so the AI is constrained by those guardrails but can run with them. 

One stops you up. The other lets you run. There would be no decision to make if we were all good at encoding judgement. But we’re not. 

Let’s say I want to surface a potential partner. We can talk about it. But when it comes to automation, it means thinking about what I look for (more instinctively). 

I might say, I look on LinkedIn for potential partners. But where? And what tells me they could be good?

I have to think about it. 

I’m looking for people who are engaging. What does that mean?

I have to think about it. 

They have influence. How do I judge that? Follower count? Not completely. 

I have to think about it. 

I look at posts about our topic. I look to see which names I recognize (because I’ve seen them before). I look for the discussions they’re leading (quantity and quality of comment threads). Consistently, over time. 

So I might:
1. Define an industry / topic
2. Define the watch words to identify posts that are talking about stuff I care about
3. Then look at all commenters. Grab their names. 
4. Look at their comment counts over time. 
5. Give them points for smart posts, points for number of threads. Points for number of comments in threads. 
6. Take the top scoring influencers and enrich their profiles to create battle cards I can give my head of partnerships. 

Now I have something I can have an agent run on a daily, weekly or monthly basis. 

But that’s only the first part, right?

I also need to close the loop. 

What happens when my guy closes the lead - positively or negatively. It’s not just a status change. 

It’s also feedback on everything else - on the topics, weights, and all the calibration that surfaced that opportunity. 

As time goes on, with enough feedback, I can have another agent use the feedback data to adjust the system. And monitor its changes. 

Did it bring leads to zero? Or just dropped the list by 10%?

This is encoded judgement. The freedom to build systems, not prompts. Systems that take your own smarts and pt it to work.