The most meaningful AI wins often start with the most ordinary work.
This was the topic of episode 21 of the Box AI-First Podcast, where I engaged with Jesse Henning, Library and Information Management Manager at Argonne National Laboratory,
Jesse gave me a vivid example. Argonne has a couple hundred pages of travel requirements. A scientist traveling for work might have a basic question about how to get a meal reimbursed or what kind of snacks can be expensed. To find the answer, they once had to dig through the entire travel policy manual.
Now, Argonne is helping employees find answers a lot quicker, with AI. That may sound small. It isn’t.
When your mission is scientific discovery, every hour spent hunting through policy documents, meeting notes, or technical reports is an hour not spent advancing research. Thoughtful AI adoption should be practical, grounded, and relentlessly focused on helping experts spend more time on the meaningful work they were hired for.
For CIOs and enterprise technology leaders, here’s the TL;DR: The first real AI wins often come from governed, practical use cases that save experts time.
Key takeaways:
- There’s an emerging best practice to getting started with AI: Begin with trusted content, clear guardrails, and a problem employees actually need solved
- Enterprise AI needs governance and context: AI works best when content is secure, organized, and trusted
- Early AI value comes from reducing routine work: Helping with repetitive tasks like policy questions and document review
- AI should support experts, not replace them: Its role is to surface relevant information faster so people can apply judgment
Accelerating discovery without exposing sensitive work
We all know AI can move faster, but can it move faster without breaking the rules that matter most?
The tension of capitalizing on AI without exposing content to risk was one of the first things I asked Jesse about, and his answer was immediate and deeply pragmatic. At Argonne, the issue isn’t abstract. Scientists are working on novel research, and being first matters. As Jesse put it, they need to make sure their findings and ideas aren’t exposed in ways that allow someone else to “scoop” their work before publication.
We all know AI can move faster, but can it move faster without breaking the rules that matter most?
That concern goes beyond academic competition to privacy issues, experimental subjects, and anything that could touch national security. In Jesse’s words, the goal is to make sure that when researchers interact with AI, “that information isn’t leaking outside the enterprise.”
The environment must be scoped, governed, and intentional from the start. Jesse says, “We really pride ourselves on making sure we have the right answer, not the right enough answer.”
That line stayed with me because it captures a standard more organizations should adopt: In high-stakes environments, speed matters. But trust matters more.
AI augments, doesn’t replace, researchers
Without context, even powerful systems drift.
Jesse shared another striking AI example from Argonne’s library and research support efforts. Researchers had 300 technical reports totaling 10,000 pages, and they needed to find specific information buried somewhere inside. No one was going to read all of that linearly. So the team used AI to surface relevant sections and papers, then paired the findings with librarian expertise to further refine the results.
Jesse described this as a feedback loop:
- AI helps identify promising material
- Researchers react to what they see
- Librarians refine the search
- The process repeats until they get to the exact insight they need
This type of AI-enabled workflow design reinforces that the best AI systems don’t eliminate human judgment but improve the path to it. “AI doesn’t thrive without context,” Jesse said. AI helps move people closer to “the right-enough answer” so trained staff can then dig in and find “the true, specific real piece of content.”
That distinction matters. AI can accelerate orientation. It can reduce prep time. It can help people speak the language of a domain faster. But the final standard still belongs to humans.
Where AI delivers value first: Repetitive, time-consuming work
Jesse insists that some of the most valuable AI use cases aren’t glamorous at all, but operational and procedural. They’re the kinds of things people quietly dread — like travel expense reports.
That same pattern shows up in collaboration. Research teams spread across the globe generate meeting notes, interim materials, and constant updates. With AI, instead of spending the first 30 minutes of a meeting catching people up, teams can ask what changed between dates, what happened last week, or who’s working on what.
What if you had a really, really smart, super obedient, but very literal intern? What would you offload to that intern?
The pattern also shows up in document review. At Argonne, scientists submit journal articles, conference abstracts, and slide decks that must be reviewed for rigor, export control, national security concerns, and other risks. Sometimes those deadlines are tight. Sometimes the document is 200 pages long. Jesse’s team is exploring AI-assisted review not to make final decisions, but to help experts focus on the sections most likely to matter.
The role of AI isn’t to replace the reviewer. It’s to amplify the signal, reduce the noise, and get the right human to the right passage faster. If the task is repetitive, rules-based, and time-consuming, it may be exactly where AI can help first.
What separates AI experiments from production
Durable AI solutions are the ones that are scoped and contained and that operate within governance frameworks already in place. That’s what helps them move from interesting pilot to repeatable production.
Jesse offered a warning that applies far beyond research environments: Don’t throw a thousand unrelated documents into one place and expect consistent answers. If the data spans different time periods, authors, and subject areas without clear structure, the outputs will get messy fast.
Instead, he argued for curation. Better metadata. Better descriptions. Better rules. Better expectations for output. In other words, the quality of the AI experience depends heavily on the quality of the information environment around it.
Metadata, categorization, and context aren’t back-office chores. They’re what make intelligent systems reliable.
I was not surprised to hear Jesse’s emphasis on curated, organized content. He is, after all, a library and information manager. But his philosophy on curated content for AI is right in line with the Box ethos on Intelligent Content Management — one I see confirmed by every customer we serve.
Metadata, categorization, and context aren’t back-office chores. They’re what make intelligent systems reliable. Those who take the time to organize and govern their content are naturally set up to benefit from AI.
Jesse also pointed to a practical adoption model I think many organizations will recognize: both top-down and bottom-up. Senior leaders are heavy users of AI for synthesizing large volumes of text. At the same time, some of the most successful use cases emerge from teams closest to the work. His travel chatbot example didn’t come from the top. It came from the travel group itself, which saw a recurring problem and realized AI could help communicate answers already documented in their manual.
That’s often how real transformation happens. Leadership creates the conditions. Practitioners discover the use cases. Then the best ideas get refined and scaled.
What should leaders take away from this?
Jesse isn’t arguing for AI as a magic pill but as leverage. Used well, it helps people get oriented faster, find the right sources sooner, and spend more time on work that actually requires judgment, creativity, and care.
This is true whether you’re supporting scientists, reviewing sensitive documents, managing institutional knowledge, or simply trying to save employees from losing an hour a day to avoidable friction. Jesse said the pilots he’d keep are the ones that help “save an hour of their day every day.” I think that’s a better north star than most grand transformation slogans.
When you remove drudgery, you create room for better work. And when you pair AI with curation, governance, and human review, you create the conditions for trust.
If there’s a broader lesson here, it’s that the future of AI at work may look less like a dramatic handoff from humans to machines and more like a disciplined redesign of how knowledge gets found, reviewed, and used.
For more, I invite you to read the Argonne customer story on Box Context or watch the full podcast episode below.
