Did the AI prompt interface open Pandora’s box? Is the tiny text box an ultimate tool for shaping the future, enabling anyone to become a creator, coder, or strategist, or is it the new easy button that relinquishes control?
Paul Updike, senior director of technical marketing engineering at Nutanix, was working with Claude Desktop, Anthropic’s AI agent, on his personal laptop when he realized he was not only was he using the AI tool for large coding tasks, but to handle small errands, asking the tool to fetch details from code directories. He hit an existential fork in the road, or maybe it hit him.
“I very rapidly realized: ‘Oh my God, this is how we’re going to do everything,’” Updike said. “I’m not alone in this. But I hadn’t really realized on a personal level that I’m not going to click around anymore. We’re not going to type in a terminal window anymore. We’re going to be telling an AI: ‘I want to build a server, it’s going to look like this, it’s going to have the following applications on it, and I need to have it up by Tuesday.’”
“You just start talking,” Updike added, “and the thing talks back to you.”
Organizations and knowledge workers are, of course, still discovering and inventing the AI-powered processes that will presumably make up much of the future of work. So far, this has involved substantial back-and-forth, with humans uploading context-rich resources to large language models (LLMs) and asking the AI tools to pull from those materials during the co-creation of new software, content, presentations, and other outputs. But Updike, along with other observers, sees a future where all of this happens directly inside the AI prompt window.
In such a future, Updike said, users and organizations are likely to gravitate toward open technologies, rather than those that force them to use specific agents or protocols.
“More open, API-heavy technologies like Nutanix tend to be easier to integrate with AI in the fashion you want,” Updike said. “You can build your own agent and say: ‘I want to do whatever I want to do.’”
He explained that open API AI integration enables AI workflows to run on public cloud, on-premises and across hybrid IT infrastructures. In fact, it’s already happening at a fast clip, sometimes bringing unintended consequences. The Register reported on Forrester Research’s The State Of Agentic AI In 2026, which found that more than half of enterprises already report agentic sprawl.
AI-for-Everything Gets Expensive
Updike noted how easy it is to fall into a pattern of simply asking AI to do things for him. Rather than manually building out observability features using the open-source analytics and visualization tool Grafana, for example, he has found himself simply instructing Claude to create the features he needs.
“It’s going to do a better job than me, or at least as good of a job as me,” Updike said. “I’ve built some pretty incredible things without ever looking at the code. It’s amazing.”
Updike has also noticed these workflows seeping into other, more mundane, parts of his day.
“Writing code, administering things, even just checking your banking account and making sure that you’re not overdrawn…it’s all going to be the same,” he said. “And we’re getting there really quickly. Any complex, multi-step task is going to be easier to do in AI.”
Certainly, this all-AI-all-the-time model reflects how many people are currently working. However, the cost calculus is changing rapidly as both vendors and AI consumers increasingly treat token expenses as real money. Recent reports depict an immediate backlash as companies dial back their “tokenmaxxing” initiatives, where employees have been incentivized via internal leaderboards to consume AI resources as quickly as possible. And some AI vendors are beginning to switch from flat “all you can eat” fees to metered pricing, raising bills by as much as 100x for heavy users.
Tokens are small pieces of data that AI models process during a prompt, helping track AI usage and calculate costs. Tokenmaxxing hit the scene in early 2026 as new programs popped up to incentivize employees to maximize their use of AI tokens. These programs were often built around internal usage leaderboards, but indiscriminate use of AI quickly became too expensive and unsustainable.
Companies are beginning to feel the true cost of using AI for every small query, just as users are growing accustomed to making these queries through their AI tools. So: How can this AI-dependent working style survive in a world where tokens are rationed?
Updike thinks the answer may lie in how AI systems are architected. He recently read about a user who started running AI models locally on a $4,000 laptop, rather than consuming a steady stream of cloud resources.
“This person removed potentially $100,000 a year in spending with local models that were good enough,” Updike said. “Not perfect, but good enough. I think there’s a point where the frontier models are so advanced that the market optimization isn’t going to be: Can I use the best AI? It’s going to be: Can I use the right AI?”
AI and The Changing Role of the Knowledge Worker
In theory, AI shifts employees' work from execution and implementation to planning and judgment. Ideally, humans will set parameters and provide AI tools with the context needed to perform tasks, and then use their organization- and domain-specific expertise to tweak AI tools’ outputs until they meet rigorous quality standards.
But in practice, people often treat AI tools as something of an “easy button.” Rather than interrogating outputs, workers are often tempted to accept whatever AI tools generate, resulting in “workslop” that gets passed on to other employees and teams. Workslop, a low-quality AI-generated content, code or analysis that employees accept without sufficient review, is emerging as a documented productivity risk in enterprise AI deployments.
This sort of automation complacency is seen even in AI-powered tasks with physical dangers, such as autonomous driving. Although human drivers are supposed to intervene when autonomous vehicles make errors, most people mentally disengage when it seems the machine is in control, making it nearly impossible for them to quickly grab the wheel when needed. Similarly, AI tools can produce content, code, and other outputs that have the same outward characteristics as high-quality, human-produced work. Underneath the surface, though, the output may be riddled with errors that employees struggle to catch.
“I think it definitely creates a danger,” Updike said. “The danger is, it’s so very useful, and saying yes is so rewarding. It just does things for you, and then you get this little dopamine hit of: ‘Man, I accomplished something.’ But did you?”
To mitigate this risk, Updike said, organizations will need to align AI safeguards with the potential cost of failure. In low-risk scenarios, where a mistake simply means trying again, employees and organizations may be comfortable with more experimentation. But in high-risk areas like security, finance, healthcare, and legal work, the guardrails around AI systems must be stronger, Updike noted.
“If the cost of completely rebuilding something is a week and some tokens, do you slow down and do it right?” Updike said. “Or do you say: ‘Let’s fail fast and move on?’”
Updike imagined a scenario in which an AI-built app largely meets the business's current needs but must be rebuilt two years later to accommodate evolving needs. At that point, Updike said, it may be efficient to simply rebuild the app from scratch with entirely new code using the latest AI model.
“With humans, that’s practically impossible,” Updike said. “But with AI, that’s a Tuesday.”
Related stories:
Editor’s note: Learn about the Nutanix Enterprise AI solution and AI Gateway service that provides a unified, secure inference endpoint lets enterprises use cloud-hosted models (and token credits) alongside private LLMs with consistent authentication, observability, and token-based rate limiting.
Calvin Hennick is a contributing writer. His work appears in BizTech, Engineering Inc., The Boston Globe Magazine and elsewhere. He is also the author of Once More to the Rodeo: A Memoir. Find him on LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
