The OpenClaw Hype Will Fade. The Real Replacement Will Not.

The more time I spend using OpenClaw, the more convinced I become of one thing:

This technology is still far from ordinary users.

That is not how the past few months have looked online.

On social media, people have been showing off workflows, counting how much they got done in a day, selling installation services, selling skills, selling tutorials, and selling the whole fantasy of an “AI employee team.” Once the heat picked up, plenty of people who did not even have a clear need got pulled in anyway. Nobody wanted to feel late.

It reminds me a lot of the DeepSeek wave last year.

First came the emotional spike. Then the mass attention. Then the rush to install, discuss, compare and panic. After that, the heat started to come down, for a simple reason: most people ran into the same question almost immediately.

You installed it. Then what?

The hype will fade, not because it is useless, but because most people do not really know what to do with it

If you judge OpenClaw by the standards of ordinary software, it is still nowhere near finished.

It is impressive. It feels like the future. But it is still a long way from something you can hand to ordinary consumers and expect them to use smoothly.

The system is extremely open, and it changes fast. Weekly major updates already tell you something. Crashes, memory loss, missing permissions, truncated rules, bloated context, over-designed roles and runaway token burn are not edge cases. They are routine.

Most tutorials still focus on installation. The real drop-off happens after installation, in the first few hours and the first week.

You think you are about to get a reliable AI employee. In practice, you become the operator, the product manager, the tester, the quality checker, customer support and emergency response. The system helps you, yes, but it also keeps sending problems, exceptions and unresolved decisions back to you.

For power users, that is early-stage leverage. For ordinary people, it is just a high bar.

And there is a more basic problem underneath all of this. Most people still do not really understand the relationship between OpenClaw, large models, tokens, skills and GitHub. If someone has to pay another person just to install the thing, it is already a stretch to expect them to grasp model routing, long-term memory, permissions, cost control and evaluation loops.

So I have started to see this whole Chinese wave of “pay someone to install OpenClaw, then pay someone to remove it” as a kind of tech farce.

Not because the technology has no value.

Almost the opposite. It has enough value to pull people in early. The problem is that product maturity, user ability and public expectations have not lined up yet.

For 90 percent, maybe even 98 percent, of ordinary users, there is nothing wrong with standing back for the next six months and just watching. Wait until the software is more stable, the costs are lower and the ecosystem is easier to understand. Then step in.

What blocks ordinary people is not just the technology. It is the cost.

People who have never used these systems heavily usually have no real feel for token consumption.

The chat era trained everyone to think in simpler terms. One box. A few questions. One monthly subscription. And, to be honest, most of the people reading this probably still have not paid a single dollar for any AI app at all.

OpenClaw is not that kind of product.

It reads rules. It reads skills. It reads tool definitions. It reads memory files. It calls models, calls tools, retries, loops and keeps running. A lot of the token burn does not come from deep reasoning. It comes from the system repeatedly dragging its own machinery across the floor.

Part of the reason is that the major coding plans have become harder to get.

When OpenClaw first showed up, many of those plans were easy enough to buy. Now the lite and pro tiers are often capped, rationed or simply hard to access. That tells you something. The providers themselves have realised that real-world token consumption is much higher than the early story suggested.

In other words, many of the subscription models that still look affordable are being held up, at least in part, by subsidy.

Once those subsidies tighten, or usage keeps climbing, ordinary users will feel the pressure first.

I have changed some of my own earlier instincts too. I no longer think everyone should rush to build an AI team, or that not using these tools right now automatically means you are behind.

It is not that simple.

At this stage, OpenClaw feels a bit like early Windows and a bit like the early internet. Important, probably transformative, but still not a finished consumer product. The people using it heavily right now are closer to early testers than everyday users.

The excitement is real. The value is real. But there is still a long stretch between that and something the average household can use with ease.

The strange part is that it may be far from consumers and still very close to companies

This is the part that feels genuinely unsettling.

OpenClaw may still be too rough for most ordinary consumers. It is becoming more attractive to companies at the same time.

That is not hard to explain.

For a company, this is not a toy for sampling the future. It is a new execution layer. Information gathering, basic analysis, document building, code generation, workflow stitching and repetitive operational tasks are exactly the kind of work that companies already know how to price.

Even if an agent is unstable today, even if it only gets to eighty or ninety percent, a company will still do the maths if that eighty or ninety percent already covers enough low-value work.

If roughly one hundred dollars of tokens a month can produce more than the marginal value of certain jobs, and do so without labour law, insurance, paid leave, moods or resignations, companies are not going to ignore that.

This is where I think many people are reading the story backwards.

The part that will fade first is the social-media imagination aimed at ordinary users.

The part that will keep moving is the quieter, far more practical replacement process inside companies.

The first behaves like a craze. It rises quickly and cools quickly.

The second is more like water seeping into a foundation. Quiet. Not dramatic. Easy to miss, until the structure has already changed.

For many countries, this looks less like a cure than two poison pills

At the level of national policy, the problem gets even more awkward.

If a country pushes hard on AI, productivity may rise, but labour pressure is likely to arrive faster too. Many middle-class cognitive jobs that once looked relatively safe may get squeezed earlier than people expect. No society treats that lightly.

But slowing down is not exactly safe either.

Capital, talent, infrastructure and industrial capacity will not pause and wait for the hesitant. A country that delays too long may find itself giving away future leverage to others.

The more I think about it, the more it feels like this:

For many countries, AI is not a clean solution. It is two poison pills.

One pill is acceleration. The price is earlier labour disruption and sharper structural shocks.

The other pill is delay. The price is strategic drift and falling behind.

Neither option feels good.

That is why I think it misses the point to ask only whether the OpenClaw frenzy is overhyped.

Of course the frenzy will cool down.

Many people who are not actually equipped to use it will eventually step away. A lot of the mythology around installation services, packaged workflows and easy promises will cool with it. Product maturity, cost structure and user capability are still badly out of sync.

But that does not mean the story is ending.

It may mean the real part is only just beginning.

Ordinary people do not need to rush in. It is perfectly reasonable to watch, to wait, to get clearer about the problems in your own life and work before deciding what to learn and how far to go.

But we should not mistake a temporary farce for a false direction.

The OpenClaw hype will probably fade. The real replacement probably will not wait.

That is the strange thing about this moment.

The technology is still too immature for most ordinary users, and already close enough to force companies and governments to recalculate.

That is the part I keep coming back to.