The early wave of generative AI adoption was all about experimentation and high expectations. Nearly every major organization has now tested AI in some form. The real question in 2025 is no longer if companies will adopt AI, but how quickly they can turn experiments into measurable productivity gains.

Why Results Vary So Widely

Some companies are reporting dramatic performance improvements, particularly in software development and sales, with efficiency gains of up to tenfold. Others, however, are still struggling to move beyond pilots with only incremental results. A recent study by MIT found that 95% of generative AI pilots failed to achieve their intended results.

Two factors explain much of this gap:

• Integration with existing IT systems. Startups with modern tech stacks connect AI tools easily. Established corporations, weighed down by legacy infrastructure, find integration far more complex.
• Organizational mentality. Companies that go all-in see the biggest gains. Those that treat AI as a side experiment tend to see only marginal improvements.

From Hype to Evidence

The honeymoon phase of “AI for AI’s sake” is over. Executives have grown more demanding after experiencing failed AI projects. Buyers now expect clear ROI, not just flashy proofs of concept.

Adoption speed still varies by industry. Tech and finance are leading. Manufacturing is catching up. Other verticals remain cautious, often slowed down by regulatory hurdles that delay enterprise roll-outs.

The “Coffee Break Problem”

Many AI tools still fall into what we call the coffee break problem: they look impressive in short demos, sparking excited conversation during a coffee break, but fail to transform core processes.

The projects that deliver real impact tend to be highly customized, addressing core business processes and integrating with existing systems. Workflows worth automating are rarely generic, which is why off-the-shelf solutions often disappoint. Increasingly, AI startups deploy forward-embedded engineers, a model popularized by Palantir, to tailor products directly inside client environments.

This points to an important reality: today’s phase of AI adoption requires services as much as software. Close engagement with customers unlocks transformative results, but scaling services-heavy implementations demands careful resource balancing for startups.

A Turning Point

Enterprise AI is at a crossroads. Flashy demos are giving way to the harder work of deep integration and organizational change.

The companies that win in this next stage won’t be those with the most pilots but those willing to commit, embedding AI across workflows, adapting processes, and embracing the messiness of transformation.

Those who remain at the “coffee break” stage may soon find themselves left behind.

This piece is the first in a three-part series by Andreas on how AI is reshaping organizations. Next, we’ll explore how AI is moving beyond productivity gains to become a new coordination layer, transforming the way companies align and operate.