Human Resources
From Copilots to Autopilots: The Era of Genuinely Autonomous Work Is Here
Agents of the future will find things to do, then do them.
Joel Hellermark
Chief AI Officer
Workday
This fall, we saw 10,000 Open AI agents collaborate to solve Navier-Stokes, one of math’s hardest problems, in 88 hours.
That news followed the skyrocketing popularity of Noah Shin’s invite-only AI assistant, Instinct, which tackles real-world tasks proactively for consumers, no instructions required.
The same capability is enabling a new class of agents at work that will look very different from the copilots we have built to-date.
We're now entering the era where AI agents do genuinely autonomous work.
Report
AI Agents Are Here—But Don’t Call Them Boss
Today’s AI Models Are Capped by Human Limitation
AI is already changing jobs and work at a dizzying pace. According to Workday research, 76% of global employees use AI in some part of their work. And in some industries, the percentage is much higher.
While adoption is no longer the hurdle, we have yet to fully realize AI’s potential because we are limited by the copilot model.
Most agents today follow the same pattern: the user prompts, the agent responds. These agents still add value, but there’s an inherent ceiling. The system is only as good as the user's prompt—even as LLMs get more advanced. In order to take full advantage of an agent, the user needs a full mental map of the agent’s capabilities and enough creativity to imagine every valuable use case worth prompting for. Even the most sophisticated users aren’t able to think of everything the model can possibly deliver.
Most agents today follow the same copilot pattern: the user prompts, the agent responds. These agents still add value, but there’s an inherent ceiling.
The result is a persistent, structural gap between model capability and realized value. That gap doesn't close when the underlying models get better. If anything, it widens, because capability is now advancing faster than user imagination can keep pace.
From Copilot to Autopilot
I believe the next era of agentic AI for finance, IT and HR will be built on a new model: autopilots.
Instead of waiting for a prompt, an autopilot sits inside the tools you already use, like your inbox, ticketing system, or ATS—and continuously observes activity.
It looks for repeated work and notices patterns a busy person might miss, anticipates what needs to be done, and proposes how to automate the work moving forward.
I believe the next era of agentic AI for finance, IT and HR will be built on a new model: autopilots.
Once you approve a new agentic workflow, the autopilot runs the work without further prompting, until it hits a built-in guardrail.
If an autopilot identifies missing information that could change how it should respond or get work done, it proactively asks for the context it needs to carry out the work safely and accurately.
Finally, getting more value out of these agents doesn’t require extensive user education, onboarding, or human creativity. It just requires the agent noticing more.
Some examples are already easy to imagine:
- A finance agent could scan an inbox, find unfiled receipts, and suggest filing them. Once approved, it would file future receipts automatically, so the user does not have to ask every time.
- A recruiting agent could audit an ATS or career site, identify unfilled roles, and suggest candidates to contact. Once approved, it would automate sourcing and outreach going forward.
- A support agent could analyze a ticketing system, find the backlog of unresolved tickets, and suggest a structure for resolving the highest-volume issues. Once approved, it would build and run that automation.
In each case, the pattern is the same: an autopilot observes what is happening, discovers opportunities to help, proposes a solution and then, with approval, runs autonomously. The user's job shifts from "think of what to ask for" to "approve or decline what's suggested.”
It is a dramatically lower bar that scales usage without scaling user effort.
The Agentic Future
The promise of AI at work is much more than productivity. Routine work will fade into the background. Entirely new ways of working will open up. Autopilots will hill-climb toward the global maximum of model capability and find problems to solve without waiting for the user to find that global maximum themselves.
Autopilots will hill-climb toward the global maximum of model capability.
And with agents working on autopilot, every employee will have agentic teammates that understand the business they’re in and how to proactively contribute. The experience of using these agents will be so simple and intuitive, people will no longer need to learn software to get their jobs done.
Report
AI Agents Are Here—But Don’t Call Them Boss
A gap between effort and impact is the real ROI problem. Discover the solution.