Betting on Early Tech: How I Decide What to Chase

I argued at one point that Meta was too far ahead rather than wrong, and that the market would eventually catch up to the bet.

I was half right, which in practice means wrong. Being early and being incorrect look identical for long enough that the distinction stops mattering.

I have been early on things that worked and early on things that never arrived. Here is the filter I use now.

Question 1: Who Is Using It Daily Without Being Paid To?

The most reliable signal available.

If the only enthusiastic users are people with a financial interest, employees, or investors, the technology has not found a use case. Genuine unpaid daily use by ordinary people is hard to fake.

AI assistants passed this quickly and obviously. Metaverse platforms never did, at any point.

Question 2: What Is the Alternative, and Is It Actually Bad?

Adoption is a comparison. New technology has to beat what people currently do, including the switching cost.

Video calls were fine, so immersive meetings failed. Traditional search had genuinely degraded, so AI answers spread fast.

I now write down the current alternative explicitly before evaluating anything. It stops me from grading a technology against an imagined status quo.

Question 3: Does It Require Buying Hardware?

A hardware requirement multiplies the adoption timeline. Every time.

Software that runs on a device somebody already owns can reach scale in a year. Anything requiring new hardware takes a decade and usually needs a price collapse first.

This one question would have saved me most of the metaverse years.

Question 4: Who Is Making the Projections?

Market size forecasts for emerging technology are usually published by organizations selling services related to the transition.

That does not make them fraudulent. It makes them systematically optimistic and they should be discounted heavily.

I look for the number of actual paying users instead. It is smaller, less exciting, and real.

Question 5: Is the Bottleneck Physics or Software?

Software problems get solved fast. Physics problems do not.

Headset weight and heat are thermodynamics. Battery density is chemistry. These improve on decade timelines.

Interface quality, model capability, and developer tooling improve on annual timelines.

If the thing blocking adoption is physical, extend your estimate substantially.

How I Apply This

If a technology passes four or five of these, I invest real time. Learning it properly, building something, writing about it.

If it passes two or three, I stay literate and do not build. Read enough to have an opinion, skip the implementation.

If it passes one, I ignore it until something changes.

That framework would have had me on AI early and out of the metaverse quickly, which is roughly the opposite of what I actually did.

The Cost of Being Wrong Either Way

Being too late is a real cost and it is recoverable. You lose an early advantage and you catch up.

Being too early costs years and credibility. I wrote a lot of confident material about a thing that did not happen, and rebuilding from that is slower than catching up would have been.

The asymmetry argues for a higher bar than enthusiasm suggests. Wait for the daily unpaid users.

The full account of getting this wrong is in what happened to the metaverse, and the corrected version of the same analysis applied to AI is in the real limits of AI search engines.

For a rigorous framework on adoption timing, Gartner's hype cycle methodology is useful as a shape even though the specific placements are often arguable.

How Do You Tell Being Early From Being Wrong?

You usually cannot, in the moment, and that is the honest answer. What you can do is set a checkpoint in advance: a date and a specific observable signal that would tell you the thing is progressing. Without that, being early and being wrong feel identical for years.

The signal I use is unpaid daily use by people outside the industry. Not funding rounds, not partnership announcements, not conference attendance. Those all continue long after a technology has stopped progressing toward adoption, because they are produced by people with a financial interest.

Setting the checkpoint is the part that requires discipline. Write down, today, what you expect to be true in twelve months if this is real. Then actually look. Most people skip the second step, and the position quietly persists on momentum.

The asymmetry worth internalizing: being late to a real shift costs you an advantage and is recoverable within a year. Being early to something that never arrives costs you years and the credibility of everything else you say. That argues for a higher bar than enthusiasm suggests.

What I do now with things I find interesting but unproven: stay literate, do not build, and write about them with explicit uncertainty rather than confidence. Saying I am not sure yet is allowed and it ages considerably better than the alternative.

The honest summary of my own record: I was slow on AI and years early on spatial computing, and both mistakes came from the same habit of grading technology on what it could do rather than on what people would actually switch to. The filter above is the correction, and it is uncomfortable precisely because it tells you to wait when waiting feels like falling behind.

What are you early on right now? The honest test is whether anyone outside your industry uses it voluntarily.