Generative AI for language has graduated from the playground and entered the messy, awkward jr. high years. The easy wins are behind it; now, enterprises must navigate the technology pimples and evolving behaviors ot tackle the hard work of helping genAI grow up. Our new report, The State Of Generative AI For Language, 2025, just dropped as a year-end recap. It’s time to understand where we are, recalibrate expectations, and double down on value-driven investments, reasonable expectations and — the most important thing of all — humans.

GenAI’s Awkward Adolescence Is On Full Display

Executive mandates are racing ahead of reality. Boards and CEOs often expect payback in 6–12 months, but disconnected strategies, messy data, and immature governance stall progress. While two-thirds of AI decision-makers say their organization uses genAI in production, only 15% report a positive impact on earnings, and just a third can link AI spend to profit and loss. Confidence in ROI is dropping, as well: In late 2024, 81% of firms reported 5% or greater ROI; by mid-2025, that fell to 62%! Without clear metrics, enterprises default to easy productivity wins that are hard to quantify — a fragile foundation for long-term value for executives who expect to see bottom- (and top-) line impacts quickly.

Trust Gaps And Workforce Anxiety Are Most Concerning

Trust is still a major obstacle as the industry realizes that today’s language models are inherently unpredictable and prone to errors, making them difficult to trust at scale. Privacy and security remain key concerns, as well, with leaders anxious about data leaks and model jailbreaks. Governance for GenAI also lacks maturity. For example, 69% of AI decision-makers do not fully grasp generative AI’s nondeterminism. These gaps create a “trust tax” that must be paid to implement agents that use generative AI for language as a foundation. This slows down decision-making and implementation timelines.

Additionally, employees are receiving mixed messages, creating confusion and disillusion. Nearly half of businesses have cut jobs due to AI, and 61% anticipate that some roles will disappear altogether. Yet automation often fails to keep up in replacing these positions. At the same time, demand for AI expertise is growing rapidly, creating churn, heightening anxiety, and potentially hindering adoption.

Six Broad Use Case Categories Have Emerged

The last time we wrote this report, use cases were only starting to emerge, and there were hundreds of them. In 2025, we saw six broad categories evolving across three time horizons:

  • Now: content creation, conversational assistants, and software development automation. Enterprises start with low-risk, high-volume tasks such as summarization, translation, and drafting marketing copy or RFPs. Conversational assistants are now common in lower-risk situations. Fixing software bugs and some coding automation are also delivering benefits quickly.
  • Short-term: productivity/decision support and governance automation. Automating work in these more critical and sensitive areas is taking longer to catch on because errors at scale can be costly.
  • Middle-term: autonomous systems/agents. The great hope for language models is that they can serve as the foundation of agentic systems, but that remains to be seen for high degrees of autonomy and critical processes.

2026 Will Bring Bubbles And Batteries

The economics of genAI have proved unforgiving. Token-based pricing clashes with early expectations of cheap AI, while longer prompts and deeper reasoning spike usage costs unpredictably. Providers are scrambling to recoup investments, but big AI tech firms are spending billions on infrastructure — Amazon plans to spend $100 billion over the next decade, while Microsoft nearly spent $80 billion in 2025 alone. This mismatch between costs and revenue has analysts whispering “bubble.” And then there’s energy: AI data centers could consume nearly 945 terawatt-hours by 2030, straining grids and budgets. In North America, more than half the grid faces a shortage risk by 2027. Energy is now a critical resource shaping AI’s future. Battery technologies to keep energy supply flowing in support of AI demand are exploding, as is the drive to build modular nuclear reactors and microgrids.

How To Work With Awkward Teenage AI

We think that enterprise clients need to do three things to navigate genAI’s early teens:

  • Wire every use case to a financial driver. Treat each deployment as a tough exam — success means a measurable impact on profit and loss.
  • Industrialize context engineering and knowledge infrastructure. Build strong habits — invest in AI-ready data and application context pipelines that are the foundation for reliable AI.
  • Upskill talent instead of cutting headcount. Invest in AI-powered humans and communicate how you will bring your people along the journey.

Schedule time with me to discuss how to apply these strategies and set realistic expectations for value in 2026.