What Six Shifts Taught Me About the Seventh

Every shift looked risky when I acted on it. That is the one thing that repeated. I have now been through six, and I am in the seventh.

I do not claim I predicted any of them. In each case I saw something small and close to home first, and then made a bet that some people around me thought was a mistake. Here is how each one looked from the inside.

1989: the customers asked

In 1989 I started a company recycling toner cartridges. Our customers kept asking us for personal computers. It became clear we could build them.

So I made a risky decision. We built and sold clone PCs. The company grew to two stores. I did not forecast the personal computer. I listened to people who were already asking for one.

1999: two bets at once

In 1999 I wanted to be part of the software industry. Most software at the time ran on a PC, and some of it was starting to run on a server in the customer's building. I convinced the CEO of a local software company to let us build a product for real estate agents that lived in our own cloud.

We also built it to be self-serve. A user could sign up and pay without talking to anyone.

Both moves were controversial and heavily debated. Both turned out to be fantastic bets. We started in 1999 with real estate agents, and by 2000 the product was running. It took until 2007 until we had more than 150,000 signed up for free and more than 6,000 paid, with no salespeople.

2008: the cash register stopped being the test

When we started Vendasta, we were building a product for homeowners. Then we noticed something. Businesses used to advertise in traditional media and had no idea if it worked. They found out when the cash register rang.

Then the phone got a real browser. People searched for what they saw in an ad, and often found something else. Google called it the zero moment of truth. We saw that a business needed listings to be found, reviews to be trusted and social media to have customers speak for it.

We moved our listings, reputation and social products to be sold under other companies' brands, by newspapers and yellow page companies. That was controversial and risky too. It paid off.

2014 and 2020

In 2014 we bet that our channel partners needed a platform to build, market, sell and deliver everything a small business needs. We pivoted to build it.

In 2020 the pandemic forced a change of plan. We added e-commerce and calendaring, and connected the video tools small businesses were suddenly using.

December 2022: the memo I would test myself on

In December 2022 I wrote to the whole company that AI had reached an inflection point. It is the best test of what I know, so I will say plainly what it said.

Use outside models. Do not build our own. The winners would be the ones with large systems of record and good places for AI to work. Humans check everything before it goes out.

Those three decisions have held up. I would put them today as: we rent the model, we own the logic.

I will also say what was missing. For five years we had spent small amounts on machine learning, with no serious investment and no significant results. The memo was about content suggestions. AI would draft and a person would decide. In my own three steps of AI at work, that is step one. It makes you more productive. It is necessary. It is not the product.

Amara's law

The memo opened with a rule I trust. We overestimate the effect of a technology in the short run and underestimate it in the long run.

That is why timing matters more than being right. Move too early and you pay for a market that is not there. Move late and you are copying someone. Each of my six shifts was a timing call, and each looked sudden to some people at the time.

The seventh

Look at the list again. Personal computing, hosted software, the zero moment of truth, the partner platform, remote and digital, AI. The first six changed how software is distributed and accessed. The seventh changes how software is generated.

That is the difference between step one and step three. In 2022 the AI suggested and a person did. Now it can hold a job. It runs whether or not someone is at the keyboard.

The signed line I use for this is short. Code got cheap. Correct is still expensive. The bottleneck moved from typing to judgment.

Six shifts is a small sample, and I should say so. I do not know that the seventh will pay off the way the others did. I know how I decide. I look for the thing people are already asking for. I weigh what it costs to be early against what it costs to be late. And I keep a person accountable for what goes out.

The pattern from the earlier shifts still applies. Listen to what people are already asking for. Make the bet that looks wrong to the people who like how things are. Keep a person responsible for the result.

In 2022 AI made suggestions. In 2026 it completes jobs.