Reading · AI Gold Rush · RSS Amplifier

Enterprise demand for AI keeps growing, and software companies are responding by shipping AI features at a remarkable pace.

There is just one problem.

A lot of customers aren’t actually using them.

A recent survey of more than 260 vertical market software companies found that 51% of software providers had fewer than one in four customers using the AI capabilities they had built.

That creates a strange gap in the enterprise AI market. Companies are building faster than ever, customers say they want AI, yet adoption is lagging far behind development.

The problem increasingly comes down to readiness, workflow changes, security, data quality and whether the new AI feature actually solves a meaningful business problem.

The speed of software development has changed dramatically.

AI can help companies build, test and deploy new features in weeks instead of months. That’s a huge advantage, but it also creates a new problem: companies can now build things faster than customers can figure out how to use them.

Enterprise customers have plenty of reasons to hesitate.

AI features may need to fit into existing workflows. Security teams need to understand how data is handled. Employees may need training. And generative AI can produce unpredictable results, which makes businesses cautious about putting it into important processes.

There’s another issue that is harder to measure.

Many companies still haven’t figured out exactly where AI can create the most value.

Wanting AI and knowing what to do with it are two very different things.

Ben Schein describes the situation with a term worth remembering: tokenmaxxing.

Employees can consume huge amounts of AI tokens, giving executives the impression that AI adoption is booming.

But usage alone doesn’t tell you whether the company is getting anything useful from it.

A team might generate thousands of AI-assisted documents, write huge amounts of code or run countless queries while making almost no measurable improvement to revenue, costs or productivity.

The better question is simple:

What did all that AI usage actually accomplish?

Schein also points to another major problem: data.

AI systems are only as useful as the information they can access. Companies need clean, accessible and properly governed data if they want reliable results from AI.

That means data governance is becoming part of the AI strategy, whether companies planned for it or not.

For software vendors, the lesson is equally important.

Building an AI feature and putting it in a product doesn’t guarantee that customers will find it useful.

Instead, vendors need to spend more time understanding how customers actually work and where AI can change those processes.

Companies should think beyond simply making an existing workflow faster.

Sometimes the bigger opportunity is redesigning the workflow entirely.

That requires much closer collaboration between software companies and their customers.

Vendors can launch quickly, measure adoption, see what works and adjust. Customers can provide feedback on usability, security and the practical problems that don’t show up in a product roadmap meeting.

That feedback loop could become more important as AI development accelerates.

The enterprise AI race has created a new measurement problem.

For years, software companies could point to features shipped as evidence of progress. In the AI era, that metric is becoming much less useful.

The real question is whether those features change something that matters.

Does the AI reduce costs?

Does it save employees time?

Does it increase revenue?

Does it help customers complete important tasks faster?

Does it improve the quality of decisions?

Those are the numbers executives should be watching.

For companies that haven’t started their AI roadmap, the answer is still to move quickly. But speed should come with measurement. Start experimenting, track the results and put more resources behind the capabilities that actually create value.

And for software vendors, the lesson is even clearer.

Building AI is getting easier. Getting customers to use it well is becoming the harder part.

The companies that figure out that second problem may have a much bigger advantage than the ones simply shipping the most AI features.

Anthropic says it will begin adding invisible, machine-readable watermarks to text generated by Claude across its models, Claude Code, Claude Cowork, Claude Tag, its API, and supported cloud deployments. The watermarks are designed to remain detectable even after copying and some forms of editing, and Anthropic says the move is tied to the EU AI Act’s transparency requirements for AI-generated content. New models launched in the EU will include the technology from day one, while older models will be retrofitted.

However, the watermark won’t prove that Claude originally wrote a piece of text. Claude can also watermark content it only translated, summarized or edited, while heavy editing, paraphrasing, translation, screenshots and file conversions can potentially remove the marker. Anthropic has not disclosed the technical method or detection accuracy, and it does not currently offer a public detection tool. This leaves open questions about how reliable the system will be and whether attackers could eventually develop ways to remove or falsely add the hidden markers.

Japan is facing labor shortages, an aging population, and weak productivity, but AI adoption remains far behind countries such as the US, UK, and Singapore. Only 8.4% of Japanese workers use AI at work, compared with 50% in the US and 32% in the UK. Experts attribute the slow adoption to Japan’s conservative corporate culture, low tolerance for mistakes, concerns about reputational risk, and a preference for using AI only for low-risk tasks such as writing, summarizing, and information gathering.

The government is trying to accelerate adoption through its AI Promotion Act and aims to make Japan a leading country for AI development and use. However, outdated IT systems, low digital literacy, a shortage of nearly 800,000 IT workers by 2030, and concerns about job losses continue to hold companies back. Some businesses are now prioritizing AI literacy when hiring graduates, and adoption is gradually increasing, but autonomous AI agents remain rare. Without greater willingness to embrace technological change, Japan risks missing out on the productivity gains AI could provide.