Employers are pouring millions of dollars into artificial intelligence (AI) tools to increase productivity, but workers are still stuck with only a fraction of the technology's potential, according to a presentation by a top executive in the field who advises Fortune 500 companies on strategy and technology implementation.
Open Machine CEO Allie K. Miller spoke at the Fortune Brainstorm AI conference in San Francisco last week. From decades of experience at companies such as IBM and Amazon Web Services (AWS), I have argued that AI actually has four different modes of interaction that are becoming increasingly useful. Miller, who helped launch IBM's first multimodal AI team, said AI can be a microtask person, a companion, a proxy or a teammate, depending on the desired outcome.
The problem, Miller said, is that most users never get out of the first mode, where they use the AI for “microtasks,” essentially a glorified search engine that returns results for simple queries.
Her central critique focused on the fundamental way most employees interact with large-scale language models (LLMs). While traditional software (“Software 1.0”) requires precise inputs to produce accurate outputs, AI allows for inference and adaptation. Mistaking the former for the latter, she claimed, would mean wasting your annual ChatGPT, Gemini, or other subscription.
“Ninety percent of employees are stuck in this mode, and far too many employees think they're superusers of AI when all they're doing is asking the AI to write a mean email a little more politely,” Miller said.
This barrier is preventing companies from truly improving productivity, Miller added.
“With people stuck in this mode, annual subscriptions become worthless,” she said, implicitly encouraging organizations to rethink their AI investment budgets.
Miller's ideas are backed by data. A November study by software company Cornerstone OnDemand found that the “shadow AI economy” that thrives beneath the surface of corporate America is increasingly fragmented. The study found that 80% of employees use AI in the workplace, but less than half have received proper AI training.
To unlock the real value of enterprise AI, Miller's presentation outlined a shift to three more advanced modes: “companion,” “delegate,” and, most importantly, “AI as a teammate.”
By using AI in this mode of interaction, the technology acts not as a reactive answer provider, but as a collaborative partner who can participate in meetings, answer questions, and take action. OpenAI engineers are already doing this by incorporating the company's software engineering agent Codex into Slack, essentially treating them as colleagues, she added.
While “Delegates” handle 40-minute tasks such as managing your inbox, “Teammate” mode represents a fundamental change in infrastructure. In this mode, AI is ambient rather than transactional and “lifts systems and groups rather than individuals.” Miller predicted that current workflows will be reversed in the near future. “We will no longer prompt AI…AI will prompt us, because it will be built into our systems and help the entire team.”
But even for non-AI companies, embedding technology in this way essentially underpins the business tasks that employees perform every day, leading to greater productivity than an isolated curiosity about trivia questions.
“The big difference with AI as a teammate is that it powers systems and groups rather than individuals,” she added.
To bridge the gap between email rewriting and the deployment of autonomous systems, speakers introduced the concept of “minimum viable autonomy” (MVA). This is a twist on the old product design principle of the minimum viable product, a market-ready prototype. This approach encourages leaders to stop treating AI like a chatbot that requires “18 pages of perfect prompts” and start treating AI as goal-oriented software.
“We are no longer going to give step-by-step perfect instructions…we are going to provide goals and boundaries and rules, and the AI system is going to work backwards from the goals,” the speaker explained.
To operate this safely, Forecast proposes the implementation of an “agent protocol.” This is a strict guideline that groups tasks into categories of “always do,” “ask first,” and “never.” The speaker recommended a risk-diversified portfolio for these agents. Allocate 70% to low-risk tasks, 20% to complex cross-functional tasks, and 10% to strategic tasks that fundamentally change the organizational structure.
A warning for the next decade
The presentation ended with positive predictions for the immediate future. Speakers predicted that within months, AI will be able to operate autonomously for more than eight hours without interruption. Additionally, as costs fall, companies will go from running a single query to hundreds of thousands of simulations per market launch.
However, these advances come with caveats for legacy-minded leaders. The veteran concluded by reminding us that assessing whether AI is “good or not” is the new essential product requirement.
“AI is more than just a tool. Organizations that continue to treat AI like a tool will be looking at what happened over the next 10 years,” Miller concluded.
