Workday’s Jon Lexa and Diginomica’s Jon Reed share key insights on the reality of enterprise AI. Read their practical takeaways for scaling business technology today.
Sydney Scott
Editorial Strategist, AI
Workday
Audio also available on Apple Podcasts and Spotify.
A person is sitting at their desk when the red light on their computer mouse starts to blink. Unsure of what that could mean, they ask an AI agent. Trained on a trillion data points, the agent confidently explains that the blinking red light indicates the mouse is broken. But the agent is wrong and that blinking red light just means the mouse wants to pair with the computer.
That small, dumb failure is the whole story. Enterprise analyst and Diginomica co-founder Jon Reed told the anecdote to host Jon Lexa during a live taping of the Future of Work podcast in New York City, and it points to a bigger problem sitting inside corporate boardrooms right now.
Companies are pouring millions into giant AI systems that still don't know basic, everyday facts. Reed says the fix isn’t a bigger model, but smarter setups built around smaller ones, with real context and human experts kept firmly in the loop.
Skip that and teams don’t just risk wrong answers, but security holes and operational breakdowns. Here are five hard truths that Reed and Lexa revealed during their conversation:
Report
"LLMs out of the box are just not good enough."
Jon Reed
Enterprise Analyst and Co-Founder
Diginomica
So many companies buy expensive AI models, believing bigger software solves every operational problem. Yet the core model only supplies a fraction of total system capability, “LLMs out of the box are just not good enough, frankly, for our purposes here,” Reed said.
Citing Google’s 50-page playbook, he shared that base AI models deliver just 10 percent of real-world system power. Context engineering and harness layers supply the remaining 90 percent. Surrounding system design gives these digital tools their true strength.
That’s good news for budgets. Reed pointed to research showing companies save up to 90 percent on computing costs by pairing smaller models with rich context layers. It’s why Reed champions a clear shift in strategy, "One of my rallying cries lately is sort of independence from frontier models."
Old-school databases and static rulebooks weren’t built to keep up with how we work today.
Reed argues AI hits its toughest bottleneck at the context layer and the fix for it is what he calls real-time organizational truth, decision traces that bring a rulebook to life the way case law brings a constitution to life.
The trouble is, most of that real-time judgment never makes it into formal files. Important decisions happen every day in quick messages and project channels that legacy software can't see.
Lexa framed the stakes of that shift directly, noting that a company might scale a successful project only to hit real barriers, “whether it's the context or real time organizational truth that limits it, or maybe just pure, good old fashioned security and governance."
Newer tools are bridging that gap, tying official records directly to daily workflows so AI tools finally get the full story. That missing foundation gives companies a clear path to turn promising experiments into trusted systems.
Base AI models deliver just 10 percent of real-world system power.
There is a persistent myth in corner offices that machines have to run entirely on their own to pay off. Reed pushes back hard on that idea, and the alternative he offers is reassuring rather than restrictive.
"One huge piece of BS is that you have to be fully autonomous to achieve ROI,” Reed said during the live session, “We've documented so many cases where you can use all kinds of AI to achieve significant ROI, while still including humans in important places in the loop. In fact, a lot of the most embarrassing AI missteps have been over autonomous situations.”
To show what those safeguards look like in practice, Reed pointed directly to Workday’s built-in guardrails when during a live demo he attempted to push a custom agent to every Workday employee. The platform’s security controls immediately intercepted the command.
Reed pointed to it as proof that smart guardrails work in real time to prevent the exact kind of runaway automation that causes headline-grabbing missteps.
The data mess inside most companies runs deeper than boardrooms often admit.
A Workday report found that 77 percent of employees still reconcile conflicting data across separate systems. Tired of clashing apps, some employees are already building their own AI tools or using shadow AI to get work done.
Despite the disconnected apps, Reed sees workers’ enthusiasm for building and seeking new AI tools as proof that the appetite for this technology runs deep across the workforce. He says, "I think what individuals are realizing is that there's ways to solve real workflow problems using AI in new ways."
Giving employees a safe space to experiment can turn a cool side project into a reliable tool for the entire business.
"One huge piece of BS is that you have to be fully autonomous to achieve ROI."
Jon Reed
Enterprise Analyst and Co-Founder
Diginomica
A common fear across the industry is that AI will replace human expertise. Reed calls that premise dead wrong, pointing out, "That's totally wrong. In some cases, it's only your domain experts that can identify what went wrong with certain kinds of output."
When an AI system produces a bad answer, someone with real domain knowledge still has to catch it. AI won’t replace human expertise or creativity. Instead, it is surfacing exactly where teams need to collaborate better.
Companies that communicate this openly can get ahead of worker anxiety. Highlighting pharmaceutical giant Pfizer, Reed referencing their conference session on the value of talking openly with staff to ease job-loss fears—a reminder that transparency, not secrecy, is what builds trust in these tools.
Executives get to choose the culture they build around AI, and the companies leaning into openness are the ones seeing it pay off in adoption.
Report
A gap between effort and impact is the real ROI problem. Discover the solution.