3 days ago
AI Agents Are Here, But Most Companies Aren't Ready
Farah Ayadi is a Staff Product Manager at Feedly, where she leads the development of AI-powered enterprise tools.
The enterprise rush to deploy AI agents is real, but so is the complexity. Behind the surge in adoption lies a stark truth: Most companies are unprepared for what it really takes to succeed.
In the past year, AI agent deployment has accelerated at an unprecedented pace. Enterprise usage tripled in a single quarter. Daily use of productivity assistants jumped from 22% to 58%, according to KPMG's recent pulse survey. Sixty-five percent of surveyed organizations have moved beyond experimentation.
But statistics tell only part of the story. For every successful rollout, there are failed pilots, unused licenses and unmet expectations. The gap between AI hype and execution is widening fast, and it's leaving many leaders exposed.
Agent adoption is exploding, but so are expectations.
KPMG's survey also found that AI agent pilots jumped to 65% between Q4 2024 and Q1 2025. Today, one-third of enterprises have agents in full production. Most executives now ask not whether to adopt agents, but how quickly they can do so.
Technology firms are leading the way: Seventy-seven percent of developers in some sectors now use coding assistants daily. Even highly regulated industries like finance and healthcare are accelerating adoption.
Still, acceleration doesn't equal readiness. Many deployments are premature, fueled more by fear of missing out (FOMO) than strategy. And that creates risk.
Where are agents delivering results?
In narrow, well-defined domains, AI agents are already driving real ROI, such as in:
• Customer Support: Top agents can now handle 85% of inquiries with 90% accuracy (registration required), delivering 24/7 service while reducing costs and wait times.
• Software Development: I've seen this transformation firsthand. One developer on my team recently used a code agent to complete three pull requests with 80%, 70% and 100% AI assistance, saving about two hours. They still needed a few agent follow-ups, but the gains were real.
• Research And Analysis: The biggest surprise for me has been how AI research agents have changed product research. What used to take me a full day, due to competitor analysis, synthesizing user feedback and reviewing technical documents, now takes two to three hours. The key is knowing what to ask and how to validate the results, but the time saved is undeniable.
These wins are real. But they're also the result of focused efforts, not plug-and-play deployment.
Why do most deployments struggle?
Salesforce's Agentforce offers a telling example (paywall): About 5,000 deals signed, but 40% of customers remain on free trials. Many haven't activated or fully adopted AI tools, not because the tech doesn't work, but because the organizations aren't ready.
In my experience building AI-powered products, the biggest barrier isn't technical but cultural. I've seen customers express genuine excitement about deploying agentic workflows. But when it comes time to implement, adoption often loses momentum. Teams report lacking the time, energy or bandwidth to rework their processes, even when they believe the tools could improve efficiency and outcomes in the long run.
Expectation mismatch is another major hurdle. Executives often approach AI rollouts expecting plug-and-play automation, immediate ROI and near-perfect accuracy. But what they actually encounter are users skeptical about hallucination risks and integration complexity.
What works? Starting small. The most successful deployments I've seen begin with narrow, high-impact use cases, like automating customer email triage. These projects may not be flashy, but they deliver quick wins, build internal credibility and create space to grow. Just as importantly, the rollout is treated as a change management challenge, not just a technical integration.
What do winners do differently?
Leading organizations follow consistent patterns:
• Start with data readiness before choosing a tool.
• Pilot multiple use cases in parallel.
• Invest in change management alongside technology.
• Set realistic expectations (80% to 90% accuracy is a win, not 99%).
• Build internal AI literacy to avoid over-reliance on outsourcing.
In contrast, laggards expect perfection, ignore cultural resistance and walk away after early friction.
Don't wait for perfection; start learning now.
Today's agents are imperfect but improving fast. The organizations winning right now aren't those with flawless rollouts. They're the ones that started early, learned quickly and improved continuously.
Because while AI agents are still evolving, the organizations that grow with them now will be best positioned to win when the tech matures.
The perfect time to start was yesterday. The second-best time is now.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
