Why Is Every AI Company Focusing on Coding Now?

Why Is Every AI Company Focusing on Coding Now?

It’s simple. AI coding is currently one of the very few areas where investors can see even a glimmer of hope—one of the fields most likely to become profitable and achieve a positive ROI in the near future.

And I do mean hope, nothing more.

AI coding meets several conditions:

1. It consumes a huge number of tokens. Whether generating code or reviewing it, tokens are burned extremely quickly. By comparison, even if you push AI fiction generation to the limit, it still won’t consume that much.

2. Demand is high on both the B2B and B2C sides. The B2B demand needs no explanation. On the consumer side, countless product managers who believe they are “just one programmer away” from realizing their dreams are now relying on AI coding to make those dreams happen.

3. Its development costs and objectives are relatively well-defined. There are fairly stable and objective standards for evaluating whether a coding model is good. That is not the case with art or fiction. Some people may think an AI-generated work is good, but once that kind of content floods the market, even “good” stops feeling good.

To put it another way, in their current forms, AI fiction and AI-generated art are barely alive in either B2C or B2B markets. There is no point stubbornly insisting that AI art is replacing artists or that AI writing is improving productivity by generating answers and novels. Investors put money into things to make a profit, not to protect your feelings.

Judging by usage volume, usage frequency, and actual results, text and visual-content generation currently come nowhere close to covering the costs invested in them.

Of course, that does not mean there will never be another breakthrough or that someone will never discover a viable business model. Perhaps one day these fields will suddenly advance again, find a path to profitability, and give investors a reason to believe they can make money.

But as of today, there is not much reason for optimism.

There is no need to talk about some distant future or grand historical trend. Investment inherently has a herd mentality. Once a field becomes fashionable, it aggressively siphons resources away from every other sector. It is like a store with a queue outside: the longer the queue becomes, the more people join it. Meanwhile, a store with no customers only becomes emptier.

When the OpenClaw space exploded in popularity a while ago, people genuinely hoped that agents would carry the banner and create a profitable direction for AI. But after all that experimentation, hardly anyone is talking about the “lobster” stuff this year. In the end, the only direction that seems able to survive is code agents.

That is also why hardly anyone talks about AI-generated PowerPoints, AI email sending, A2A, or similar office-assistant applications anymore. It is not that they are completely useless. The problem is that their useful applications are too limited. Most of what they can produce consists of low-value daily or weekly statistical reports that managers and employees use to go through the motions—documents with very little actual information.

Most importantly, they do not make money.

From today’s perspective, even the heavily promoted concept of the “one-person company” is on shaky ground. There are certainly small niches where modest profits can be made, but those niches mean very little in the context of the AI industry’s massive investments in foundation models.

The number of directions in which people can genuinely see hope can be counted on one hand. One is AI coding. Another is AI-generated short-form video.

As for AI coding, I am personally somewhat pessimistic.

Unless the industry develops a genuinely new paradigm or a new way of using it, AI coding will end up as either an oversized “Python Plus” or a more convenient code reviewer.

The technical problems may actually be less important. Whether the code is correct or whether the architecture is good can be discussed separately. The most fundamental problem is that there simply is not that much demand for software development.

Before AI arrived, the demand for all those management systems, ERP platforms, and inventory, purchasing, and sales systems had already been squeezed nearly dry. All the low-hanging leaves had already been stripped from the trees, never mind the fruit. There is not even any soup left to drink.

And that is before we even mention the endless flood of generic apps and mini-games: bookkeeping apps, Pomodoro timers, text adventures, and countless other copycat products.

Meanwhile, high-barrier industrial and productivity software—as well as commercial software requiring years of continuous iteration—remains beyond the capabilities of today’s AI. Even if AI eventually breaks through those barriers, how much new demand would that really create?

Do we genuinely need that much software?

Foundation models are also improving so quickly that the market will inevitably turn into a price war before long. Forget recovering the original investment—the companies involved will be lucky if increased usage does not cause them to lose even more money.

No matter how you look at it, fiction, visual art, and programming all face the same ultimate problem:

There is not enough demand to support them.

If AI is going to maintain its momentum, it must continually create new categories of demand that give investors hope. Otherwise, it dies.

Honestly, the whole thing feels rather uncertain. It could collapse one day, and there is probably a fairly high chance that it eventually will.

AI coding has already entered the later stage of its novelty cycle. Whether in B2B or B2C markets, finding a few bugs, solving a few mathematical problems, or creating yet more slopware does not generate much revenue.

To put it bluntly, most of this merely gives people inside the industry something to congratulate themselves about.

Think about it: how many users are actually willing to pay you because “I used AI to discover several CVEs and solve a few mathematical problems that I cannot understand or even explain”?

I do not know how much longer AI coding and AI-generated short dramas can keep this going.

In the long run, agents may have potential. But today, I cannot see where that potential will come from.

As for the other directions, we still do not know what a winning path to profitability would even look like.