For Real: What People Really Do with AI at Work
Practical Wins, Surprising Uses, and How Teams Are Actually Leveraging AI Today
Everyday AI use at work is scaling up quickly.
According to Gallup polling from June, frequent AI workplace use has nearly doubled since 2023, and daily workplace use has doubled in the last year. It’s still more common among leaders, and in white collar roles, and it’s mostly bottom-up.
And, as I’ve written many times, most AI adoption in companies still doesn’t come with much guidance and often no guardrails.
But that is a story for another time. Today we’re looking at a pair of intriguing reports on real-world AI use right now, in 2025.
One is research data from Marc Zao-Sanders, an author and Co-Founder of Filtered dot com, that’s been released as both a long-form study on Filtered and as a Harvard Business Review AI insights article, How People Are Really Using Gen AI in 2025.
The other is from The New York Times’s The Upshot (21 Ways People Are Using A.I. at Work by Larry Buchanan and Francesca Paris), and it details specific examples of AI use going on in workplaces right now.
Taken together, these show that AI use isn’t always following the LinkedIn marketing pushes or what big AI companies are demonstrating on their conference stages.
Instead, it’s often simple things that help make work easier, faster, and more complete.
Let’s take a look.
The Top Real AI Success Stories—in Life
Before we narrow our focus to workplace use cases, let’s look at overall use.
For his research, Marc Zao-Sanders analyzed thousands of real conversations across a variety of platforms like Quora and Reddit, using a rigorous, expert-driven qualitative analysis of public discourse.
He's done this over multiple years, and he finds the main uses in 2025 are much changed from last year. And most don’t align with the use cases that get the most talk in marketing pushes.
In looking at the volume of use, he noted a general shift from primarily technological and productivity-oriented uses to more personal ones.
His top five by frequency from 2025 thus far:
Therapy and companionship. This was #2 in his research in 2024, but this year it takes the top spot. He found people are really using AI to talk though their problems because it’s always available, non-judgmental, and more accessible than professional therapy. (It’s important to note that GenAI is also not made for this and lacks the expertise and consistency necessary to counsel people responsibly.)
Organizing their lives. This use case is not surprising for me, as it’s also one of the most frequent workplace AI implementations that yields real fruit. People are using AI broadly to manage calendars, build routines, and create more structure.
Finding purpose. When AI first broke on the scene, many pundits feared it would eliminate human purpose. Instead, many people today are using AI to help them find more meaning. They’re using it to set goals and keep tabs on what matters most to them.
Enhancing learning. Another strong workplace use case, AI is improving our intelligence by helping us fill gaps in our knowledge, understand complex topics, and serve as study partners. For companies, this use takes the form of upskilling and reskilling as well as onboarding and interview preparation.
Professional code generation. Up from 47th in 2024, he found coding is the top way workers use AI in 2025. It’s helping people automate repetitive programming tasks and become more efficient.
As noted, Zao-Sanders’s research has seen a shift from pure technical and productive use to emotional and personal use, alongside education and coding.
Real World AI Use Cases at Work
Now let’s pivot and add in some of the New York Times AI workplace stories, combined with successes I’m seeing myself.
Tech: Coding AI Transforming Workflows
Of all the hype that’s out there, this use case may be the most accurate.
Simply put: AI is really helping people code. It’s already a bread-and-butter use of AI, and the Times piece profiles how DraftPilot, a legal AI company, uses Claude Code in this way.
“I can give it tasks and just walk away,” Chris O’Sullivan, CTO and Co-Founder says. “It writes the code itself.”
Of course, AI-generated code still needs checking, and humans in the loop. But if you’re repeating tasks, as in software testing you can very effectively rope in AI, for example, for highly successful automation.
Healthcare: Notation without Fatigue and Literature Review
The low-hanging fruit is where AI excels in the real-world, and for clinicians this is often taking and keeping effective notes.
AI scribing solutions listen to the patient and make structured notes inside the system. The clinician then can devote their attention elsewhere (such as to the patient directly) and only needs to review and sign off.
This not only helps avoid burn-out, but it also catches details that might otherwise get lost.
Another effective use here: narrowing down potential research sources.
For researchers who have to read medical literature every day, AI is also a great aid for pointing them in the right direction. The researchers ask specific questions and get back references to the specific literature they need to review more carefully.
Using ChatGPT, Perplexity, or Undermind proves an enormous time saver at finding the right places to go for the details, though the actual AI summaries are not yet trusted as the final word.
Nevertheless, by helping pinpoint where to go to do your work, it’s saving an enormous amount of time in the real world.
In the Public Sector: Calling AI for Business Productivity
The California Department of Tax and Fee Administration (CDTFA) gets hundreds of thousands of calls a year and is responsible for billions in revenue.
Today, they’re using a Claude system trained on their own data that uses live call transcripts to help agents sort through mountains of data.
This AI-assisted method gives their agent options on the fly and accelerates the communication process.
Other calling methods are frequently recognized as core boosters in studies like MIT NANDA’s The GenAI Divide: State of AI in Business 2025. While that report made news for noting that 95% of their surveyed enterprise AI pilots failed to generate ROI, it also pointed out areas where AI IS giving companies enormous boosts, namely:
Voice AI systems with call summarization and routing
Document automation
Code generation for repetitive engineering tasks
AI systems are not only succeeding in making sales calls but also acting in front-line customer service roles and for recruiting tasks.
Science Operations: Identification without Wasting Expert Time
The Missouri Botanical Garden has eight million dried plant specimens. And while experts can recognize what each one is relatively quickly, at their scale it means a number of expertise-heavy hours get burned on repetitive identifications.
Instead, they’re building an AI model where leaves are labeled, scanned and used as training data, based on the light each plant reflects.
Now, whenever they put new plants through the same process, this AI system quickly recognizes each one.
The experts are then only needed to go through the handful that the system fails to recognize, saving them hours and hours spent going through boxes that are easily recognized.
This institution is also collaborating with others to ensure their tool and its data are widely usable.
Again, as with in medical research examples, AI cuts down on volume and allows experts to focus their time far more effectively.
For Urban Infrastructure: Pattern Identification Extends What’s Possible
AI can be very good at identifying patterns, but in order to be useful it has to capable of learning and improving.
This example is successfully demonstrated by a company called Digital Water Solutions. By using autonomous machine learning, they have built a system that can take sensor data from individual fire hydrants, learn what’s normal, and in a few weeks start providing information on irregularities in flow might point to system leaks.
With small water systems (serving less than 10,000 people) making up the majority of systems in the US, many have small budgets, and this kind of a solution can be a life saver.
By finding potential leaks before they become big, the company is able to both turn a profit and also save these municipalities time and money.
In Design and Marketing: Improvements across the Board
It’s easy to look at how AI can whip up artwork, video, and audio and think it can do the job of a marketing team itself. But the devil’s in the details, and so far this isn’t true.
What it can do very effectively, however, is greatly improve work, and extend capabilities that weren’t there before.
I wrote last time out how this use case is helping companies place their products in varying environments and real-world demonstrations, for example, where they couldn’t otherwise afford it.
It’s also a wonder for doing things like removing glare, cleaning up objects, providing new versions and variations.
You still need taste and a keen eye on outcomes for your brand, but AI is removing entire phases of the workflow and bringing out higher quality results at a far lower price.
For Legal and Compliance: Successful AI Workplace Examples of Error Checking
One of my favorite of the examples in the New York Times piece details a custom LLM put in use by the district attorney’s office of the nation’s third largest jurisdiction—Harris County in Houston, Texas.
This system provides a series of checks on police reports, looking for things that have been prior issues (and which slow down the process or even worse, come up later in court as a problem).
A colleague of Chris Handley, their director of operations and chief of innovation, said their current version reduced her work time by 50%.
Note that another version of this tool was trained on case law and failed its test for making up facts about a case.
But this tool is succeeding by taking advantage of what Handley calls “low-hanging fruit.”
It automatically adds an additional check for common mistakes that’s been otherwise lacking, giving them a significant win in a common and effective use case for AI.
Knowledge Chores: AI Productivity Tools for Employees That Work
At this time, many of the least flashy uses of AI are proving the most effective.
Examples here in offices include summarizing documents, transcription, re-formatting (such as pasting in PDFs for form-specific breakdowns), or adding more politeness and softening the edges of communications.
Communication/companionship, organization, coding, and finding purpose are among the top overall use cases for a reason—and it’s because current GenAI systems have great capacity in these areas.
And these same capacities bring greater efficiency and consistency to workplace tasks.
Communication assistance like scheduling is one example.
Human beings must waste a significant amount of time across workplaces just coordinating appointments, scheduling, rescheduling, confirming, and reminding.
AI systems can take this task on with an extremely high degree of effectiveness, removing a work task that few human workers miss.
And the icing on the cake for employers: the results are usually more consistent, available 24/7, and result in less no-shows and last-minute cancelations.
This is a genuine, hype-free, win-win.
Conclusion: Practical AI Applications in the Workplace Steal the Show
AI is in many ways a unique innovation.
Its complexity and often baffling capacities can dazzle, but I think this also results in many organizations feeling it should be taking on high-profile, complex asks right out of the gate.
What’s perhaps counter-intuitive is that current GenAI systems, with all their capacity for conversation and vast access to information (both trained and accessible via additional resources in real-time), are now often best at tackling tedious, repetitive, back-of-shop chores—or otherwise working as an initial point of contact.
Time and again, I see the companies that succeed most with AI begin in such use cases, and often at a small scale initially.
From there you not only ensure greater comfort for employees and customers alike, but you also get to see what it does well, where it struggles, and how much you can tune and adjust it for your needs.
And once you’re succeeding in these ways, it’s far easier to scale up from there.