Making, fast and slow

Tools change fast; many human needs don’t change that fast

The last two posts kept circling one question.

Leviathan was written in 1651.

375 years later, the trust, cooperation, power and coordination Hobbes discussed can somehow still be carried all the way to companies, crypto and AI agents.

Then I came across Stewart Brand’s pace layers:

the things in a society never change at the same speed to begin with.

Technology, products and commerce can be fast; infrastructure, institutions and culture are much slower; further down, some human needs and problems of living together may be slower still.

Following this line further, I started to feel:

making things itself seems to happen between two different speeds too.

01 | What’s fast is the tools, and “what can finally be done now”

Building AI products these past few years, it’s easy to be drawn by new capabilities.

• The model gained another ability.
• Coding agents can independently finish longer tasks.
• Voice can converse in real time.
• Agents can search, read files, call tools and operate on real systems.

Every time a new capability appears, it’s natural to ask:

what else can it be used for?

That’s certainly a sensitivity builders should keep.

Because one of the most important effects of technological progress is constantly changing how feasible something is, and what it costs.

• What used to take a team weeks, maybe one person can do in a day now.
• Services nobody offered because they cost too much may become commercially viable for the first time.
• Work only professionals used to be able to do may now be largely done first with AI’s help.
• Software that used to need someone watching it constantly can start running continuously on its own.

These changes are very fast.

But:

what technology can do and why people need it are two different questions.

02 | What’s slow is many deeper needs

Transport may be the most intuitive example.

People went from walking and riding horses to carriages, trains, cars and planes.

Speed, cost and reachable distance changed enormously.

But many of the reasons for setting off aren’t that unfamiliar:

to look for new opportunities.

To see a world you’ve never seen.

To leave a place you don’t want to stay.

Or just hoping to see someone you miss a little sooner.

Communication too.

Letters, telegrams, phone calls, text messages, instant messaging, video calls.

A message went from taking months to arrive to arriving almost in real time.

What changed is the medium.

But much of the time we’re still just expressing some very old needs:

I want to know how you’re doing. Something happened, and I want to tell you. I need your help. I want you to know I’m still thinking of you.

AI is like this too.

It can help people look things up, organize information, write code, arrange work, explain problems and offer a second opinion.

These capabilities are very new.

But many of the needs they serve aren’t that new:

we’ve always wanted to do a little less repetitive work.

Always wanted to find the information we need faster.

When we don’t understand, we hope someone will explain.

When there’s a lot going on, we hope someone will help us sort it out.

When making decisions, we hope for another perspective.

When things are hard, we hope someone can help a little.

These used to be handled separately by colleagues, teachers, friends, secretaries, search engines, Excel or professional services.

What AI changes is:

who provides this help, and its cost, speed and scale.

Of course, human needs haven’t stayed completely unchanged over the centuries either.

Only after smartphones appeared did we get today’s kind of notification anxiety; social media also created specific problems — likes, followers, algorithmic recommendations — that didn’t exist before.

So more precisely:

compared with the forms technology takes, many deeper needs change much more slowly.

03 | Many product opportunities appear where fast meets slow

For builders, this is the layer I find most interesting.

Now when I see a new AI capability, I increasingly don’t only want to ask:

what can it do?

I also go on to ask:

which existing problem does it let us solve, for the first time, in a clearly better way?

Because “the model can do X” isn’t a product in itself.

Coding agents can write code, for example.

That’s a capability.

But what it might really solve are completely different needs:

• people who can’t code can still build an idea;
• engineers can cut a lot of repetitive implementation;
• a very small team can maintain a project that used to need more people;
• code review and testing can happen more continuously.

The same technical capability can connect to completely different problems.

What’s really worth looking for is often that intersection:

a long-standing problem that, because technical conditions changed, suddenly has a new solution.

Sometimes what changes isn’t even “it works a bit better.”

It’s that the whole cost structure has changed.

For example, giving every ordinary person a round-the-clock personal assistant used to make almost no economic sense.

Today, although an AI assistant is far from equal to a real human assistant, “everyone can have some kind of help on call” is becoming feasible for the first time.

The bar for a non-technical person to build software themselves used to be very high too.

Design, front end, back end, deployment, testing — many different abilities were needed.

Once coding agents drove down part of the execution cost, the boundary of “one person building a product” started moving too.

So some product opportunities aren’t things nobody thought of before.

It’s just that before, they were unaffordable, too slow, or could only serve very few people.

04 | The tools are new, but the problems they meet in real life may be old

Still, having a need and a new technical capability doesn’t mean a product naturally holds up.

This brings us back again to Leviathan, which I wrote about earlier.

AI agents are something very new.

But once several agents really start handling real work, some very old questions quickly pop up:

• who has the power to decide?
• who can act?
• what counts as a trustworthy fact?
• what happens when two agents disagree?
• who approves high-risk operations?
• who’s responsible when something goes wrong?
• how is it undone?
• why should users trust it?

In other words, once a new capability really enters people’s lives, it still has to deal with:

trust, permissions, responsibility, coordination and control.

These problems don’t get a version update just because technology updates fast.

That’s also what I’ve found interesting about Leviathan all along.

Centuries ago, people were discussing these things in states and communities.

Today, with AI agents instead, we’ve run into similar problems in a completely different form.

The environment is very new.

Part of the problem underneath is very old.

05 | So now I look at a new technology with three questions

1. What does it suddenly make easier?

Has the cost dropped sharply?

Has speed gone up?

Or is something only a few people could do before becoming widespread for the first time?

This looks at the change technology brings.

2. What human need does it meet?

Has this need existed for a long time?

How do people solve it now?

What exactly is better about the new way than the old?

This looks at the layer that changes more slowly.

3. To really be used, which old problems does it have to solve again?

Where does trust come from?

What happens when it gets something wrong?

Who bears responsibility?

How much control do users have?

Why would people be willing to hand this over to it for the long term?

This looks at what’s still missing between the technology and real life.

Put together, I think these three questions — compared with just looking at:

what can AI do now?

— are closer to how a product opportunity actually comes to hold up.

Making things may always have lived between fast and slow

Builders certainly need to chase the fast things.

If you don’t know a new capability has appeared, it’s hard to see new possibilities.

But if you only chase capabilities, it’s easy to end up:

with a new hammer, looking for things to hit.

On the other hand, if you only look at long-standing problems, you might completely miss that:

technology has already rewritten the old solution.

So making things may always happen between two speeds.

On one side, fast-changing tools, capabilities and new possibilities.

On the other, needs, relationships, institutions and human lives that move much more slowly.

Using new technology to solve some very old problems again.

• From carriages to planes.
• From letters to instant messaging.
• From bank ledgers to blockchains.
• From human teams to AI agents.

The forms keep changing.

But whether something can really last may still come down to:

whether it meets a real, enduring need that deserves to be solved better.