Before we get into that boof-a-rama, let’s look at the theory of communications. To diagnose a tricky situation, you have to understand the underlying mechanics. Communication isn’t a single event; it\u2019s like a traditional Unix Pipeline where you use prayer-based parsing and hope everything connects correctly. Here are the steps:<\/p>\n\n\n\n
- \n
- What you think.<\/strong><\/li>\n\n\n\n
- What you actually say.<\/strong><\/li>\n\n\n\n
- What they hear.<\/strong><\/li>\n\n\n\n
- What they think it means.<\/strong><\/li>\n\n\n\n
- What they remember.<\/strong><\/li>\n<\/ol>\n\n\n\n
At any step, this process can go pear-shaped. When there\u2019s a misunderstanding, you always want to think through these steps to find the point of failure.<\/p>\n\n\n\n
The Request: Solving for Attention<\/strong><\/h1>\n\n\n\n
At the time, I was the AI architect for M365. Perry and I would have these brainstorming sessions where he\u2019d firehose out a brilliant stream of ideas and I\u2019d go off to investigate them.<\/p>\n\n\n\n
Perry had a great idea: to improve user productivity, we needed to understand attention<\/strong>. Having a document open for an hour doesn’t mean anything if the user isn’t actually looking at it. If they are<\/em> looking, that\u2019s an interesting signal. He wanted to know: could we figure that out?<\/p>\n\n\n\n
That Didn’t Go The Way I Expected<\/strong><\/h1>\n\n\n\n
I tracked down some researchers who had models for this. They could track exactly where someone was looking on the screen. It was computationally expensive\u2014consuming an entire CPU\u2014but with a modern NPU, it was feasible.<\/p>\n\n\n\n
At my next one-on-one, I told him: “Hey Perry, I followed up on your request to track gaze<\/strong>, and we can do it.”<\/p>\n\n\n\n
Perry looked at me like I had a rat\u2019s tail coming out of my mouth<\/strong>. It was that visceral sense of disgust and total misunderstanding. I double-downed, explaining that we could reliably detect “gaze” and track it. I thought I was delivering the “What” he wanted, but I hadn’t checked the “Why.”<\/p>\n\n\n\n
Perry confusion turned to anger. I swear he was on the edge of firing me. He said, “Jeffrey, I never told you to track GAYS.”<\/strong><\/p>\n\n\n\n
![\"\"](\"https:\/\/www.jsnover.com\/blog\/wp-content\/uploads\/2026\/02\/image-4.png\")
<\/a><\/figure>\n\n\n\nThe Lesson: Getting Comfortable with Being Uncomfortable<\/strong><\/h3>\n\n\n\n
In that moment, I realized we had a failure to communicate. I was talking about eye-tracking (g-a-z-e<\/strong>), and he thought I was suggesting some horrific, HR-disaster surveillance of people’s sexual orientation.<\/p>\n\n\n\n
I was able to remedy the situation and keep my job, but it was a sobering reminder. If you get your communications wrong:
<\/p>\n\n\n\n
![\"\"](\"https:\/\/www.jsnover.com\/blog\/wp-content\/uploads\/2026\/02\/image-5-1024x572.png\")
<\/a><\/figure>\n","protected":false},"excerpt":{"rendered":"Technology is easy but people are hard. In particular, communications are hard. They\u2019re the physics of human interaction\u2014the $F=ma$ of how we get things done\u2014and when the forces don’t align, things get high-stakes fast. I learned this the hard way … Continue reading
- What you actually say.<\/strong><\/li>\n\n\n\n