Forrester has released its predictions for artificial intelligence and technology leadership in 2026, indicating a shift from hype to practical applications. The report highlights that enterprises are expected to delay 25 percent of their artificial intelligence spending until 2027, as many companies struggle to connect AI investments to profitability. Frederic Giron, Vice President and Senior Research Director at Forrester, noted that the complexities of AI governance will lead 60 percent of Fortune 100 companies to appoint a head of AI governance by next year to navigate regulatory challenges.
A recent report from G2 challenges the narrative from the Massachusetts Institute of Technology, revealing that nearly 60 percent of companies have successfully deployed artificial intelligence agents, with a failure rate of less than two percent. According to the G2 2025 AI Agents Insights Report, 83 percent of businesses are satisfied with the performance of these agents, and organizations are investing over one million dollars annually, with nearly 90 percent planning to increase that investment in the coming year. Tim Sanders, head of research at G2, emphasized that AI agents are proving to be more effective than previously thought, with many companies reporting significant cost savings and faster workflows. The report highlights that the leading applications for AI agents are in customer service, business intelligence, and software development, with a notable trend of organizations adopting hybrid strategies that combine both in-house and ready-made solutions.
A recent study by Atlassian reveals a paradox in the adoption of artificial intelligence across organizations. Despite a significant increase in individual usage—where daily engagement with AI tools has doubled and the number of users deeming AI as “useless” has dropped by 78%—96% of businesses report no substantial improvements in efficiency, innovation, or work quality. The survey, which included responses from 12,000 knowledge workers and 180 executives from Fortune 1,000 companies, highlights that only 3% of executives see AI as having driven transformational change in their organizations. Executives express that teams continue to operate similarly but with “extra bells and whistles,” indicating a disconnect between expected benefits and actual outcomes.
Recent research from Anthropic reveals that poisoning large language models may be easier than previously believed, requiring as few as 250 specially crafted documents to disrupt their functionality. Collaborating with the UK AI Security Institute and the Alan Turing Institute, the study demonstrated that this minimal number of malicious documents could significantly compromise generative AI models, including Llama 3.1 and GPT 3.5-Turbo, regardless of their size. The researchers found that once the threshold of 250 malicious documents was reached, the models consistently outputted gibberish whenever prompted with a specific trigger phrase, indicating a successful poisoning attack. This alarming finding suggests that just 0.00016 percent of a model’s total training data could render it vulnerable.
Why do we care?
So, Forrester’s calling it — the AI hype wave’s cresting. They think a quarter of enterprise AI spend gets kicked down the road to 2027 because companies still can’t prove the ROI. Translation? The CFOs are winning the AI debate. And it’s about time someone asked: what’s this actually doing for the business?
They also expect 60 percent of Fortune 100s to hire “heads of AI governance.” That’s the new compliance officer with a fancier title. It’s less about innovation, more about not getting sued.
Meanwhile, G2’s saying AI agents are killing it — 60 percent deployed, 2 percent failure, huge satisfaction. But that’s likely the happy crowd — people already invested in AI tools. Atlassian’s research tells a different story: everyone’s using AI, and nothing’s really changing. More tools, same output. A very different picture than the MIT study that’s been kicked around.
Then Anthropic drops a bomb — just 250 poisoned documents can corrupt a large language model. That’s terrifying. If you’re using public or partner data, your AI system might already be vulnerable.
This is where MSPs and IT providers can step in — not as AI developers, but as AI enablers. Help clients set up governance, secure data pipelines, and actually measure results. AI’s not magic. It’s just another system to manage, govern, and make accountable. Whoever nails that — wins.