For most of my life, software waited.
You opened it. You told it what to do. You filled in the fields. You moved the object from one column to another. Even very good software was basically a room full of tools with nobody inside using them.
That is changing.
The old bargain
Software used to make a promise that was easy to understand: give me your labor and I will make each unit of that labor more productive.
Photoshop did not design the ad. Excel did not decide what the business should measure. Salesforce did not call the lead. WordPress did not decide which page the company needed next.
The human remained the operating system.
That sounds obvious until you notice how much of the modern economy is built around it. We created entire job categories whose practical function was moving intent between software systems. The marketer opened one tool, exported a CSV, uploaded it somewhere else, generated a report, interpreted the report, created a task, handed the task to another person, and then repeated the whole thing next week.
None of those tools were bad. They were designed around a constraint that is starting to disappear: software could store state and apply rules, but it could not reliably understand the work well enough to carry much of it forward on its own.
That constraint is disappearing faster than the org chart
Stanford's 2026 AI Index reports that 88 percent of surveyed organizations now use AI somewhere in the business, while actual agent deployment remains in the single digits across most business functions.
STANFORD HAI · 2026 AI INDEX · ECONOMY
I think that gap matters more than the adoption number.
It says the technology has already entered the building, but the building has not yet been redesigned around it.
Most companies are still putting AI into jobs that were designed before AI existed. They are giving a copywriter a better writing tool. Giving an analyst a better research tool. Giving a salesperson a better drafting tool.
Useful. Incremental. Still the old bargain.
The next bargain is different: give software an objective, context, permissions and boundaries, and let it carry part of the work to completion.
The verb changed
The first time most people used generative AI, the verb was ask.
Ask it a question. Ask it to summarize. Ask it to write.
Then the verb became help.
Help me code this. Help me analyze this. Help me prepare for this meeting.
Now the verb is becoming do.
OpenAI's Operator research preview in 2025 was explicitly built around an agent using a browser by clicking, typing and scrolling. The product later folded into the broader ChatGPT agent experience. The significance was not the browser. It was that the model crossed the boundary between describing an action and taking one.
OPENAI · 2025-01-23 · INTRODUCING OPERATOR
That is a much bigger change than another improvement in writing quality.
What I am building around
At YG3, the question I keep coming back to is not "how do we give a marketer a better AI tool?"
It is "why does the marketer have to manually operate all of this in the first place?"
A business has an identity. It has customers, competitors, markets, offers, constraints and a body of first-party information. Once the system understands that context, why should every channel start over from a blank prompt?
The website should know what the outbound system knows. The ad system should know what the site learned. The content system should know what customers are asking. The reporting layer should not be another place a person has to manually reconstruct the same company.
YG3 is my attempt to build around that premise.
Not because I think marketers disappear. Because I think the value of a marketer moves upward when the software can carry more of the execution.
The part I would not automate
There is a temptation, especially in AI, to treat "can be automated" as equivalent to "should be automated."
I do not believe that.
Judgment is not a bug in the workflow. Taste is not waste. Accountability should not disappear because an agent clicked the button.
Microsoft's 2026 Work Trend Index uses the language of human agency: as agents take on more execution, humans spend more time directing outcomes. In its survey, 58 percent of AI users said they were producing work they could not have produced a year earlier.
MICROSOFT WORK TREND INDEX · 2026-05-05 · AGENTS, HUMAN AGENCY, AND OPPORTUNITY
That is the future I find interesting.
Not software replacing the person who knows what good looks like.
Software removing enough execution that more people get to operate at the level where knowing what good looks like actually matters.
The work is not disappearing.
It is moving.
