I remember once flying to a meeting in another country and working with a group of people to annotate a proposed standard. The convener projected a Word document on the screen and people called out proposed changes, which were then debated in the room before being adopted or adapted, added or subtracted. I kid you not.

I don’t remember exactly when this was, but I know it was after the introduction of Google Docs in 2005, because I do remember being completely baffled and frustrated that this international standards organization was still stuck somewhere in the previous century.

You may not have experienced anything this extreme, but many people will remember the days of sending around Word files as attachments and then collating and comparing multiple divergent versions. And this behavior also persisted long after 2005. (Apparently, this is still the case in some contexts, such as in parts of the U.S. government.) If you aren’t old enough to have experienced that, consider yourself lucky.

A note from the development of the White House AI Action Plan

This is, in many ways, the point of Arvind Narayanan and Sayash Kapoor’s essay “AI as Normal Technology.” There is a long gap between the invention of a technology and a true understanding of how to apply it. One of the canonical examples came at the end of the Second Industrial Revolution. When first electrified, factories duplicated the design of factories powered by coal and steam, where immense central boilers and steam engines distributed mechanical power to various machines by complex arrangements of gears and pulleys. The steam engines were replaced by large electric motors, but the layout of the factory remained unchanged.

A marine engine factory in Greenwich, England, 1865

Only over time were factories reconfigured to take advantage of small electric motors that could be distributed throughout the factory and incorporated into individual specialized machines. As I discussed last week with Arvind Narayanan, there are four stages to every technology revolution: the invention of new technology; the diffusion of knowledge about it; the development of products based on it; and adaptation by consumers, businesses, and society as a whole. All this takes time. I love James Bessen’s framing of this process as “learning by doing.” It takes time to understand how best to apply a new technology, to search the possible for its possibleness. People try new things, show them to others, and build on them in a marvelous kind of leapfrogging of the imagination.

So it is no surprise that in 2005 files were still being sent around by email, and that one day a small group of inventors came up with a way to realize the true possibilities of the internet and built an environment where a file could be shared in real time by a set of collaborators, with all the mechanisms of version control present but hidden from view.

On next Tuesday’s episode of Live with Tim O’Reilly, I’ll be talking with that small group—Sam Schillace, Steve Newman, and Claudia Carpenter—whose company Writely was launched in beta 20 years ago this month. Writely was acquired by Google in March of 2006 and became the basis of Google Docs.

In that same year, Google also reinvented online maps, spreadsheets, and more. It was a year that some fundamental lessons of the internet—already widely available since the early 1990s—finally began to sink in.

Remembering this moment matters a lot, because we are at a similar point today, where we think we know what to do with AI but are still building the equivalent of factories with huge centralized engines rather than truly searching out the possibility of its deployed capabilities. Ethan Mollick recently wrote a wonderful essay about the opportunities (and failure modes) of this moment in “The Bitter Lesson Versus the Garbage Can.” Do we really begin to grasp what is possible with AI or just try to fit it into our old business processes? We have to wrestle with the angel of possibility and remake the familiar into something that at present we can only dimly imagine.

I’m really looking forward to talking with Sam, Steve, Claudia, and those of you who attend, to reflect not just on their achievement 20 years ago but also on what it can teach us about the current moment. I hope you can join us.

AI tools are quickly moving beyond chat UX to sophisticated agent interactions. Our upcoming AI Codecon event, Coding for the Agentic World, will highlight how developers are already using agents to build innovative and effective AI-powered experiences. We hope you’ll join us on September 9 to explore the tools, workflows, and architectures defining the next era of programming. It’s free to attend. Register now to save your seat.