Mastermind · May 22, 2026
Mics in Every Room
Tom Bloomfield said something in a recent YC talk that I keep coming back to. He said he wishes there were mics in every room at the event, catching what everyone is saying.
It sounds invasive at first. It is also the right instinct.
The conversation is the data. The data is the company.
Most teams I talk to track the wrong north star. Revenue. Signups. Active users. These are all real numbers, but they are downstream of the thing that actually compounds. The upstream metric, the one that decides whether you will matter in two years, is whether you are collecting data that nobody else has.
Data is the new oil. That line is tired. The version of it I actually believe is sharper. If you have asymmetric information about the world, and you can pipe it into a model, you generate insights no one else can. That is the moat. Not the brand. Not the funnel. The data underneath.
So the question I keep returning to is this: what is meaningful data, and how do you collect it?
My working definition is that meaningful data is data whose value is defined by what you can build on top of it. Two halves. The value of the data itself. The value of the layer you put over it. Either one without the other is not a business.
This is why I think every team should be asking three questions every Monday morning:
- What data are we going to collect this week?
- How are we going to collect it?
- How are we going to analyze it to measure utility?
If you cannot answer those three on a Monday, you are not running a data company. You are running a service business with a CRM bolted on.
The frame that helps me think about this is Input → Brain → Output.
Most teams optimize the Brain. They tune the prompt, swap the model, add tools, build agents. All real work. None of it the leverage point. The leverage is in the Input. If your input is the same public web everyone else trains on, your output is commodity. If your input is something only your product can see, your output is differentiated by definition.
Take hackathons, the world I live in. Most hackathon platforms collect the final submission. A demo link, a GitHub URL, maybe a video. That artifact is the tip of an iceberg, and most platforms are storing the tip and throwing the iceberg away.
The actual signal is the build itself. Every prompt sent to an agent. Every false start. Every decision to abandon an approach and pick up another. Every conversation between teammates about what to ship. Every judge's question at the demo. The artifact is the receipt. The build is the company.
If you can capture the build end to end, platform-agnostic, harness-agnostic, you have something nobody else has. You can benchmark on it. You can train on it. You can use it to surface the next builder for the next problem. You can sell access to it. The product is downstream of that data, not upstream of it.
This is what Bloomfield was pointing at with the mics. He was not being weird. He was being early.
There are two failure modes I see teams fall into here, and I am not exempt.
The first is a leaky funnel. You have the events, you have the participants, you even have the platform, and the data still walks out the door at the end of the weekend. You support the hackathon end to end and capture nothing structurally useful. The infrastructure is built for the event, not for the data the event throws off.
The second is no analysis. You finally start collecting something, and you never go back and ask whether the collection is doing what you hoped. You do not have hypotheses, so you cannot be wrong. You do not revisit assumptions, so you cannot learn. You accumulate, but you do not compound.
The fix is unsexy and operational. Hypotheses on paper before the collection starts. A weekly review where you ask: did the data we collected this week sharpen or refute the hypothesis we wrote down last week. Then repeat. The pattern is the same one that makes science work and the one most startups skip.
I think this matters more than revenue right now. That sounds reckless. It is not. Revenue is the lagging indicator. The leading indicator is whether the data you control today gives you a position you can defend tomorrow. If the answer is no, the revenue you book now will get arbitraged away by whoever does control the data.
So the working principle I am trying to hold the team to is this. Every project, every product, every event, the first question is what meaningful data does this generate, and where does it land so we can build on top of it. If a project answers that question well, it ships. If it cannot, it does not matter how exciting it looks. We are trying to stop building flow-throughs and start building catchments.
Tom Bloomfield wants mics in every room. So do I. The next year is about figuring out where to put them.