AI Gave Me More Capacity Than I Knew What to Do With.

I have become exceptionally good at starting things.

An idea can become a codebase, an MVP, a landing page, a waitlist, and the beginning of a marketing plan before I have completely decided whether it deserves to exist.

For a while, this felt like pure leverage.

Work that once required a team, a budget, or several uninterrupted weeks could suddenly happen in a weekend. I could explore ideas outside my expertise. I could build several versions instead of debating which one might work. I could move from thought to something tangible faster than ever.

Then I looked at how many projects I was carrying.

Several were individually promising. Collectively, they were exhausting.

AI had lowered the cost of starting them. It had not removed the cost of thinking about them, maintaining them, finding users, learning a new domain, or deciding what each should become.

Every project still charged rent in attention.

The constraint moved

AI gave me more capacity than I knew what to do with.

When I watch people use AI, including myself, I generally see three modes.

We can use it to do less.

We can use it to do more.

Or we can use it to do better.

These are not types of people. The same person can move between all three during one day.

And none of them is inherently wrong.

Do less.

AI can remove work from your life.

Let it summarize the document. Draft the routine email. Research the purchase. Clean up the notes. Handle the administrative task you would otherwise do at night.

Doing less is not laziness.

If AI helps someone meet the expectations of their job while reclaiming an evening, that is a legitimate use of the technology. Not every minute saved needs to be returned to an employer as additional output.

Sometimes efficiency should simply buy a person their time back.

Do more.

AI can increase your throughput.

An SDR who previously researched 20 prospects can research 50. An engineer can produce more code. A marketer can test more campaigns. A product manager can build a working prototype instead of waiting for space on an engineering roadmap.

“More” can also mean moving in a new direction.

Someone in customer support can build their own dashboard. An executive can create an internal tool. A designer can implement the interaction they see in their head. People can cross boundaries that were previously protected by years of training or organizational dependency.

That is incredible.

It is also the easiest mode to overvalue because the evidence is so visible.

More pull requests. More documents. More prototypes. More projects. More activity.

Capacity becomes productivity, and productivity becomes a growing pile of things someone must now care about.

I fell into this trap. AI made the first 20 percent of a project so inexpensive that I started projects without adequately pricing the remaining 80 percent.

The landing page was not the company.

The MVP was not product-market fit.

A plausible beginning was not evidence that the idea deserved years of attention.

AI made starting cheap. It did not make finishing free.

Do better.

This is the mode I find most interesting.

At work, my customers are other employees. Before AI, understanding them often looked like a few conversations, a requirements document, and then execution.

That process was reasonable. It was also lossy.

Now I can work with far more raw information: meeting transcripts, recordings, support questions, observed workflows, product usage, and repeated patterns across teams.

I can study both the large process and the strange little details inside it.

I can understand where someone says they work one way but consistently behaves another. I can preserve context from conversations that would otherwise become scattered notes. I can compare what different teams believe about the same problem before deciding what to build.

The building is faster, certainly.

But speed is not the most important change. The system I build can begin from a much deeper understanding of the person using it.

AI has also made measurement that once felt unjustifiable economically reasonable.

I wanted better visibility into how employees were adopting internal AI tools. Buying a full observability product could cost thousands of dollars. Building the infrastructure myself would traditionally compete with every user-facing project and probably remain an acknowledged gap.

With AI, I could create an internal telemetry system using open standards, analyze real usage patterns, and use those patterns to improve how I support the organization.

Instead of delivering the same generic AI workshop to everyone, I can identify where people are struggling and personalize training around what is actually happening.

That is not simply more training.

It is a tighter loop between behavior, learning, and the next intervention.

That loop is what “better” means to me.

Quality test

Better work is not work that looks more polished. It is work with a tighter connection to reality.

Doing more creates real leverage today.

But I am skeptical that raw throughput is where humans maintain a durable advantage over agents. Machines are already better suited to volume. They do not get bored, protect their weekends, lose focus after lunch, or become emotionally tired of repeating the same workflow.

If the competition is who can generate the most code, research the most accounts, or produce the most documents, that is unlikely to remain a human game.

Better is different, not because models are incapable of improving quality.

Models can critique other models. Multiple models working together can produce substantially better results than one model alone. Evals can measure performance, catch regressions, and preserve what a product has learned.

But there is a danger in creating a loop where AI builds the thing, AI uses the thing, AI evaluates the thing, and AI declares that the thing is good.

The person we are supposedly helping can disappear from the system.

Humans remain unusually close to human problems. We notice irrational behavior because we are irrational. We understand that what someone says, does, needs, and rewards may be four different things.

We also make strange connections. We jump between ideas that do not appear adjacent. We care about details that cannot yet be justified by an optimization function.

Most importantly, we remain accountable for whether the result helps another person.

If the customer is human, reality eventually needs a human vote.

The Jevons paradox originally described how greater efficiency in using coal could increase total coal consumption rather than reduce it. When something becomes cheaper to use, people often find many more uses for it. That insight dates to William Stanley Jevons’s The Coal Question in 1865.

AI may produce a similar effect across knowledge work.

If software becomes cheaper to produce, we may demand much more software. If personalized research becomes inexpensive, generic research may stop feeling sufficient. If every company can analyze thousands of customer interactions, operating from three anecdotes becomes less acceptable.

Efficiency does not necessarily reduce the amount of work available. It can raise expectations for what the work contains.

But the paradox does not tell us who captures that opportunity.

When everyone can produce more, more becomes less distinctive.

And as the amount of output grows, two things become scarce:

AI makes it easier to create more things while making it harder for any one thing to earn enough attention, including our own.

That is why focus becomes more valuable, not less.

I still use AI to do less. I celebrate the time it gives back.

I still use AI to do more. Some problems genuinely benefit from volume, exploration, and parallel attempts.

But I am trying to be much more deliberate about where I use it to do better.

I now periodically audit the capacity AI has created:

The hardest question is the last one.

AI makes an idea feel inexpensive by hiding many of its future costs behind an impressive beginning. Killing that idea can feel wasteful because so much already exists.

But the generated artifacts were never the scarce resource.

My attention was.

AI gave me more capacity than I knew what to do with. I am still learning how to spend it.

For now, my rule is simple:

The capacity rule

Do less where it gives you your life back.

Do more where volume creates real value.

Do better where the outcome matters.

And kill what exists only because it became easy to start.