I spent 35 years around organizations deciding whether to adopt something new. I watched a lot of technologies show up with fanfare and leave with a shrug. So when people compare AI agents to sliced bread, I want to know if that comparison is fair. Or if it’s just a good line.

It turns out it holds up better than I expected.

A short history of sliced bread

Otto Frederick Rohwedder was a jeweller from Iowa. In 1912, he started building a machine that could slice bread. His prototype burned in a fire in 1917, along with his blueprints. He spent another eleven years rebuilding it.

On July 7, 1928, a bakery in Chillicothe, Missouri sold the first commercially sliced loaves. The local paper called it the greatest step forward in baking since bread was wrapped. By 1930, Wonder Bread was selling pre-sliced loaves across the country. By 1933, in the middle of the Great Depression, about 80 percent of bread sold in the U.S. was pre-sliced.

Five years to go from a novelty to the normal way of buying bread. That is fast.

Why it worked

Sliced bread succeeded for a few clear reasons. I think these same reasons explain most technologies that actually stick.

• It solved a small, everyday problem.
• It asked almost nothing of people. You bought bread the same way you always had.
• It used systems that already existed, like the bakeries, the trucks, and the grocers.
• It had a partner technology. The pop up toaster worked much better with even slices, so each one made the other more useful.

Within a few years, nobody called it “sliced bread” anymore. It was just bread. That, to me, is the real sign a technology has succeeded. It stops needing its own name.

How it compares to other fast adopters

Sliced bread set a high bar, but it wasn’t the only fast one. Radio reached half of U.S. homes in about nine years. Television took eight to ten. Smartphones, after the iPhone launched in 2007, went from about 35 percent of American adults in 2011 to roughly 68 percent by 2015. That’s six to eight years to reach a majority, helped along by app stores in much the same way the toaster helped the loaf.

Even so, five years to 80 percent still puts sliced bread in front of all of them.

Enter AI agents

Tools like ChatGPT and Copilot showed that a machine could write, code, and create images when asked. AI agents take this further. Instead of waiting for a prompt, an agent can plan a series of steps, connect to your systems, use the software you already have, and report back once the job is done.

That’s a real shift. It moves the technology from assistant to delegate. I think that is why people are paying such close attention right now.

Where the numbers stand

In 2023, about a third of organizations used generative AI regularly. By 2024, that number had roughly doubled. By 2025 and into 2026, most surveys put 80 to 90 percent of organizations using AI in at least one part of their work. McKinsey’s 2026 survey found that 40 percent of large enterprises were scaling agents in at least one function, up from 27 percent the year before.

Two to three years to reach majority use. That pace could beat the loaf, though I would want to see those McKinsey numbers confirmed before I bet on it.

Where I think this goes

Over the next year, I expect less noise and more quiet progress. Agents will start showing up inside tools people already use, like their inbox, their office software, and their accounting system. That matters because it skips the training that most new technology requires. Routine tasks will start moving over to agents, with a person still signing off on the result. Rules about who can use what, and on what data, will get written down. Usually after something goes wrong first.

I have sat in enough rooms to know that last part is true. A new system often gets adopted before anyone agrees on the guardrails. It rarely goes smoothly.

The one real difference

Agents check almost every box that sliced bread did. A common, unglamorous problem. Almost no change in behaviour required. Existing systems to build on. A partner technology in the models underneath.

But sliced bread only had to earn one day of trust. You bought the loaf, saw the slices, and decided again tomorrow. An agent has to earn trust with your systems and the decisions you are responsible for. That kind of trust does not build in a single try. It’s the real reason adoption, even at this pace, is not yet a straight line to 80 percent.

I still think about Rohwedder. Not because the machine itself was brilliant. It wasn’t, really. But everyone who told him bread would never sell pre-cut was making a bet about human nature. They lost that bet in under five years.

The real test for AI agents is the same one the loaf passed. Not whether the technology impresses anyone today, but whether, a few years from now, we’ve stopped calling it anything at all.