Imagine having a list of every startup someone was building in 2007, including the ones that looked too strange to take seriously.
Most would have failed. Some would have disappeared without a trace. A few would have looked ridiculous. But mixed in with them would have been early versions of ideas that later became ordinary parts of the internet.
Early.tools offers a version of that view today. It is a human-curated directory of pre-launch startups and early-stage tools, with products grouped by stage from waitlist to launched. The site is less useful as a prediction market than as a record of what founders are trying to make while the boundaries of a new technology are still moving.
The current backlog makes that visible. Alongside conventional SaaS products are tools such as Entire, which stores AI agent sessions alongside code commits; Creed, which proposes a portable personal context file for AI systems; and Hilos, which puts people and coding agents in the same git-connected rooms. There are also stranger bets: Butter, a social app built around private mini-apps rather than feeds, and Oasis, a smart ring designed for keyboardless interaction.
Any one of these products could disappear. The more useful signal is the pattern across them. Founders are repeatedly experimenting with persistent AI memory, agents that act rather than merely answer, software built around autonomous workers, and interfaces designed for a world in which humans are no longer the only active participants in software.
That changes how early-stage companies should be watched. The question is usually whether a particular startup will become a viable business. For an individual investment decision, that question is unavoidable. For understanding technological change, it can be the wrong unit of analysis.
Consider AI memory. Early.tools currently lists multiple products attacking the problem from different directions. One might fail because the market does not want its approach. Another might discover a better architecture. A third might find a narrow use case that grows into a much larger category. Even the failures provide information about what does not work.
The directory therefore becomes a kind of time capsule. It preserves experiments before the market has sorted them into winners, losers, and obvious ideas. The products that survive will eventually be easy to find. The more valuable record may be the abandoned attempts that showed what people were trying to solve before the answer was clear.
That suggests a simple experiment: save today’s Early.tools directory and come back in 2029. Which products disappeared? Which founders built something else? Which ideas became mundane? Which strange experiment turned out to be early rather than wrong?
We cannot answer those questions yet. That is precisely why the directory is worth looking at now.
A mature technology market tells you what works. An early-stage directory captures what people are willing to try before they know what works.
You do not use it to see the future. You use it to remember what the future looked like before it became obvious.