Garbage in, now dressed up in tech

I built some tools recently to understand my own fitness, using my Apple Watch data. I expected to learn something about my training. Instead I learned something about dashboards, and two findings changed how I read every number a machine hands me.

Two numbers I stopped trusting

The first was my VO2 max, and it read artificially high. When I dug in, the reason was the optical heart rate sensor on the wrist. From what I can tell on my own data, it misses brief spikes, the kind you get charging up a short hill. Drop the hardest few seconds from the record and the math concludes you produce a given pace at a lower heart rate than you actually do. The watch was grading me on a flattering edit of my workouts. I want to be careful about how far I push this: it’s what I found on my own runs, not a lab teardown of Apple’s hardware. But the pattern was consistent enough that I stopped taking the number at face value.

The second was stranger. I pointed an AI at my health data and it told me I wasn’t getting enough sunlight over the winter. Alarming, except in my case it’s almost certainly false. The daylight estimate comes from the ambient light sensor on the watch, and from late fall to late spring I’m outside with a jacket sleeve over it. If the same estimate leaned on GPS instead, it would tell the opposite story: plenty of time outdoors, sensor in the dark. The number wasn’t measuring my sunlight. It was measuring my sleeve.

Two sensor design decisions. Two wrong numbers. And a quiet chain running from bad measurement to confident analysis. If I hadn’t dug in, I’d feel great about my fitness and guilty about my sunlight, and I’d be wrong on both counts.

The garbage is better dressed than it used to be

Garbage in, garbage out is as old as computing. What’s new is how well the garbage is dressed. A number on a polished dashboard, with a trend line and a health ring around it, reads as truth. Feed it to an AI and you get confidently delivered, but misleading advice, with the sensor’s blind spot laundered out somewhere along the way.

I’ve started calling that last step AI-washing, and I mean something specific by it, not the usual sense of a company overstating how much AI is in its product. I mean this: the AI launders the provenance off a circumstantially wrong number, so a measurement artifact comes back to you as confident advice with no trace of how it was measured. The sleeve gets scrubbed out. What reaches you is “you need more winter sun,” clean and directive, and nothing in the sentence remembers that it started as a sensor sitting under a jacket.

That laundering is the dangerous part. A raw wrong number at least looks like a number, something you might squint at. Advice looks like a conclusion someone already checked.

Every company has a jacket sleeve

Companies are wiring agents to their data right now, and the corporate version of my jacket sleeve is everywhere. The CRM field nobody updates. The survey only the happy customers bother to answer. The metric that quietly changed definitions last year and kept its name. The agent won’t know any of that. It will analyze what it’s given, confidently, and hand back a clean conclusion with the provenance washed off, exactly the way the sunlight warning came back to me.

The fix isn’t to measure less. It’s to treat “how is this measured?” as part of reading any number, the same way you’d read the byline on an article. And it’s to drop the assumption that some measurement always beats none. A wrong number you trust does more damage than a gap you know about, because the gap keeps you looking and the wrong number tells you to stop.

None of this is new, exactly. A spreadsheet could always turn a bad measurement into a confident chart. What AI changes is scale and distance. It makes one more analysis nearly free, so more numbers get produced and acted on, and it stacks layers, a model reading a summary of a dashboard built on a sensor, until the output sits several steps from the thing that was actually measured. Every extra layer is another place the source and its reliability can drop out. The problem is old. AI industrializes it.

Where the data comes from matters more than ever

This is the part I care about at work, so let me connect it plainly rather than pretend the two are unrelated.

If the risk is agents reasoning over jacket-sleeve data, the answer is to ground them in data that doesn’t depend on someone remembering to fill in a field. That’s the bet behind the brain, our knowledge layer. Today it grounds the assistant’s answers in your actual work rather than a separate database you’d have to maintain, and the people-and-contact layer keeps itself current from the message stream, so who works with whom, and on what, isn’t a form somebody forgot to update. The broader goal, the brain keeping the full picture of the business current from your communication on its own, is what we’re building toward rather than something I’d tell you is finished today. The direction is the point: knowledge drawn from the work as it actually happens is harder to quietly falsify than a field nobody owns.

None of that removes the need to ask. Even the best-grounded number deserves the question. But it moves the source from “whatever someone typed in when they had a minute” toward “what the work itself shows,” and that’s a better place to be reasoning from.

What’s a number you trusted until you checked how it was measured?

Key takeaways

• Two numbers from my own Apple Watch data were built on bad measurement: a high VO2 max because the sensor missed my hardest seconds, and a low-sunlight warning because my jacket sleeve covers the light sensor all winter. Both are my own findings on my own data, not verified claims about Apple’s hardware.
• Garbage in, garbage out is old. What’s new is how well the garbage is dressed: a polished dashboard reads as truth, and an AI turns it into advice.
• The problem predates AI, which mostly industrializes it: more analyses run more cheaply, and more layers between you and the raw measurement, each one a place the source and its reliability can drop out.
• AI-washing, in the sense I mean it here, is the AI laundering the provenance off a circumstantially wrong number, so a measurement artifact returns as confident advice with no trace of how it was measured.
• Every company has a jacket sleeve: the CRM field nobody updates, the survey only happy customers answer, the metric that changed definitions. Agents wired to that data won’t know, and will analyze it confidently.
• The fix is to make “how is this measured?” part of reading any number, and to stop assuming some measurement beats none. Grounding agents in the work as it actually happens beats reasoning over fields nobody maintains.