As AI’s inexorable transformation of the enterprise continues to unfold, business leaders may be forgiven if they’re still coming to terms with the implications of large language models (LLMs) and agentic AI. But today there are new players on the field that could help bring about yet another potentially seismic shift: open-source AI agents.
These personal agents—OpenClaw is one example—can interact with other apps and use LLMs and software to carry out a variety of tasks on the user’s behalf: organizing files, responding to emails, staying up-to-date on the latest AI and market news, and executing repetitive tasks, to name a few. They typically “wake up” at frequent intervals to monitor the user’s dashboard, log files, or ecosystems of choice, and can proactively reach out or take action across platforms when new developments arise.
Personal AI agents can act in the user’s stead across a variety of contexts, effectively performing a role akin to a personal chief of staff. As the technology’s presence in the workplace grows, the implications for performance and security could be significant.
‘A Builder Economy’
Many open-source AI agents are now on the market, including also Hermes Agent and NemoClaw, often providing similar features that can give users access to more personalization, more control over computer-related tasks, and further security and operational controls. These solutions combine a model with an agentic harness including tool-calling software elements and memory.
Much of the early use of these tools is likely taking place among developers, hobbyists, and general AI enthusiasts. “There is currently a big moment around building with these agents,” says Jim Rowan, principal and U.S. head of AI with Deloitte Consulting LLP. “A builder economy is emerging.”
That momentum is dramatically accelerating the learning curve, says Ed Van Buren, a principal with Deloitte Consulting LLP and executive director of Deloitte’s AI Institute for Government. “It’s an exciting moment for personal exploration, where leaders can come face-to-face with applied AI, and it’s helping to drive AI literacy and fluency for the workplace.”
Already, that fluency is conveying into enterprise environments.
“Increasingly, people will probably begin showing up to work with their own agents,” Rowan says. “There will likely soon come a time when leaders will need to determine how to integrate these agents: how they log into work as distinct entities, where IP boundaries lie, who owns the resulting knowledge, and how it should all be kept secure.”
Memory and Context
Personalized agent capabilities have the potential to unlock numerous benefits both within and outside the enterprise, Rowan says, starting with personal productivity.
“A lot of users’ experience with AI so far has been through chatbots and early agents, but those tools are typically limited to specific purposes and parts of the enterprise,” he explains. “They’re often tools we’re instructed to use at work for productivity.”
Agents built with tools like OpenClaw, on the other hand, are more personal. “They are like your own productivity tool, with memory and context from past interactions you’ve had,” he says. “They can help uncover and enable your personal way of working.”
These personal agents also bring action to the forefront. “They’re not just helpful on the same level a chatbot might be, such as by offering suggestions,” Rowan says. “They perform tasks on your behalf the way you want them done—delivering a daily briefing tailored to your preferences, for example.”
Because they offer persistent memory over time and across platforms, these agents can also help fill in gaps among varying tools in the sometimes fragmented AI landscape, he adds.
‘Who Owns Your Knowledge Graph?’
OpenClaw agents are customizable for specific purposes using pre-established skill sets developed and shared by the OpenClaw community. With the ability to communicate via numerous messaging and other channels, “they meet you where you are, helping support a mobile workforce,” Van Buren says.
By fostering experimentation, open architectures can remove friction and help uncover the art of the possible for enterprises, notes Diana Kearns-Manolatos, a senior manager and Technology Transformation research leader in the Deloitte Center for Integrated Research with Deloitte Services LP. “The potential is huge, but along with it may come issues related to bandwidth, memory, scaling, governance, and controls.”
Policies and procedures for integrating personal agents into the workplace could become increasingly important, Rowan says.
For example: “When you chat with your personal agent at work, does that knowledge become the property of your employer—or is it yours?” he says. “Work products aside, you and your agent will learn and build memory wherever you use it. What happens when you leave the company—what becomes of that IP?”
Indeed, “who owns your knowledge graph and digital identity?” Kearns-Manolatos adds.
‘It Could Get Messy’
A separate but related question is what can be shared with colleagues, Van Buren notes.
“How do you determine on a given user’s personal agent what they can keep as local knowledge versus knowledge for their work context, and how much of their knowledge base and memories can be shared with others?” he says.
As an example, take relationship management. “Say I’ve learned one stakeholder’s preferences for meetings, including how they like to be briefed, how much detail they prefer, and so forth,” Van Buren says. “That’s no big deal when it’s recorded in a paper notebook, but when all of a sudden it becomes accessible through an agent and potentially shareable, it could get messy.”
Potentially even messier: “When I’m working at home, I’ll likely mention my address at some point,” Rowan adds. “Or maybe a religious affiliation. There are varying degrees of sensitivity, and they call for a thoughtful policy conversation.”
‘Bring Your Bot to Work’
For enterprise leaders, a first step is to begin learning about personal agent tools.
“Gone is the time for talking in theoreticals,” Van Buren says. “Leaders should be familiar with this technology because there are significant technical and business questions to be addressed.”
Toward that end, there’s much to be said for rolling up your sleeves and trying it out yourself. “This is a tinkerer-hobbyist moment,” Rowan says. “Half the joy is the struggle of getting it to work the way you want it to.”
“With access to AI coding tools, you don’t have to be a developer to join the community,” Van Buren adds. “Functional people coupled with AI coding tools are coming up with some great examples.”
Within the enterprise, it’s important to give people a safe environment in which to try the technology without fear of causing harm. “If you’re not giving your employees a place to play with this, they’ll likely find somewhere else, and your data could be at risk,” Rowan says.
To foster innovation, events such as a “Bring Your Bot to Work” day can give employees a way to demonstrate their creations and inspire others to do the same, Van Buren says.
‘Daily Use Can Be Critical’
When asked which capabilities they most need to develop over the next two years, 44% of tech leaders cite deepening AI and data literacy as their biggest focus area, according to the Deloitte U.S. CIO Program’s Global Technology Leadership Study.
Executives should try to use AI for a minimum of 20 minutes each day, Rowan suggests. “If you’re not trying these new things out personally, you may start falling behind.”
Indeed, in many past technology waves, “experimentation looked different,” Kearns-Manolatos agrees. “With the pace of change today, daily use really can be critical.”
For personal learning, Rowan suggests that leaders seek out focused communities at work or elsewhere.
“Getting involved in that knowledge flow can be really valuable, both for you and for helping your agents self-improve with a set of commands,” he says. “It could be possible to point your agents to an external chat, for example, and say: ‘These people are talking about really cool ideas. Implement three of them, and tell me how it works.’”
—by Katherine Noyes, senior writer, Executive Perspectives in The Wall Street Journal, Deloitte Services LP