Shanna Johnson was wrapping up her time leading cielo24, a business focused on transcribing audio and adding captions.
While preparing to shut it down, she came across something valuable she had not expected. All the digital traces left behind from years of team efforts, conversations, and shared projects turned out to be useful.
She teamed up with SimpleClosure, a young firm that guides businesses through the end of their operations. They handled the standard steps like settling final paychecks, sorting taxes, getting approvals from backers, and submitting forms to the government.
But then came an unusual step that no startup guide ever mentions. They offered up the company’s entire 13 years of online records. This included casual chat messages, project notes, and stored emails full of team highs and lows in massive online storage. All of it went to train future smart systems.
In return, the sale brought in hundreds of thousands of dollars. Johnson explained it shifted her worries from wondering how to cover basic costs to feeling like they could finish everything properly and move on without loose ends.
Even now, closing the doors leaves her feeling mixed emotions, she shared with reporters. At the same time, she finds it meaningful that the information they created might keep contributing and support others in new ways.
This gave her story a smooth conclusion amid tough times. The business itself ended, yet the record of its everyday activities lived on.
In 2026, those records hold serious financial value. Her experience shows this approach is more than a one off idea for exiting. It is opening fresh possibilities in the race to advance artificial intelligence.
Companies building these systems now understand that everyday messiness from actual settings plays a big role in properly checking how well their creations perform.
Early on, the teams developing AI pulled from openly available online sources like discussion boards, encyclopedia pages, and scanned books. By the end of 2024, they had used up nearly everything available, as shared by Ilya Sutskever, who previously led science efforts at OpenAI.
On top of that, this material falls short when creating AI that can handle real tasks independently. The genuine daily efforts captured in now closed businesses like cielo24 offer something far more useful. It works like raw energy for these smart helpers. To build systems skilled at office style jobs, developers need plenty of real examples showing exactly how work gets done.
Ali Ansari pointed out that builders of these models see the value in messy, authentic settings for better testing. His firm, micro1, offers something called Roots to AI developers. It sets up a pretend business where the systems can try out jobs such as handling money matters or juggling busy schedules.
This growing need for everyday business records has created exciting opportunities for SimpleClosure. Its leader, Dori Yona, described the rush of requests from AI firms as overwhelming.
He senses these groups are in a full rush to grab hold of genuine materials from actual operations.
To keep up, SimpleClosure plans to introduce Asset Hub, a place where closing businesses can offer their collections of programming files, chat histories, messages, and more for purchase.
Some features are still being tested, according to Yona. Carefully stripping out any details that could identify individuals is a tricky and important job. They want to perfect it fully before making it available to everyone.
Over the last year, SimpleClosure managed close to 100 such arrangements for ended businesses. They brought in more than a million dollars total for the owners. Most sales landed between ten thousand and one hundred thousand dollars each.
Another player called Sunset purchases materials from finished companies for about the same amounts. Leader Brendan Mahony explained to reporters that values vary based on how large the firm was, how long it operated, and how connected its information feels overall.
For instance, a project note linked directly to a code change holds more worth than an isolated file. Fields like health care or banking often bring higher offers too.
This kind of material is not just ordinary information. It is tied to real individuals.
While many view salvaging these records as a smart business move, some raise worries about personal privacy. Marc Rotenberg, who started the Center for AI and Digital Policy, noted that even when staff handed over rights to their created materials, it does not fully answer if firms can pass along private team talks to outsiders. Workers probably never imagined their casual messages ending up elsewhere.
He believes the privacy questions are serious. Staff protection matters a lot, since so many now rely heavily on tools like Slack for daily talks. Again, this is not plain data. It is linked to actual people.
His group recently wrote to the Senate Commerce Committee. They urged the FTC to examine these emerging AI methods more closely, with a focus on keeping personal details safe.
Buyers insist they handle the removal of identifying details with care. Experts in the field note it is never straightforward. You cannot simply flip a switch to erase all traces of someone’s professional life from years of records.
Bobby Samuels, whose firm Protege helps with rules and legal sides of using real materials, warned that poor handling could let buyers spot patterns from specific groups or persons. If not managed well, those details might show up in the final AI responses.
On another level, there is the risk that personal conversations get repeated word for word by the trained systems. Research from 2020 by groups including OpenAI and Google found that big language models sometimes store exact pieces from what they learn. Clever questions can pull those pieces back out.
The push for this kind of authentic business material has kicked off a fresh sector around practice spaces for AI. These are often called reinforcement learning environments. They use records from closed companies to create pretend setups where systems can rehearse handling typical job situations. It is turning into serious business.
Reports say Anthropic may invest a billion dollars in such practice areas this year. Dozens of new startups have popped up. Firms like Mercor and micro1 that once paid people to create practice examples are joining in.
A few of these practice space companies already carry big price tags. Prime Intellect tops a billion dollars. Fleet discusses funding that could value it at 750 million. Prime Intellect had no comment when asked.
AfterQuery offers ready made simulated settings to AI developers. These include versions for large corporations, banking, or taxes. In them, an AI system learns by moving through virtual offices, talking with made up team members, and figuring out solutions to common challenges.
One typical exercise feels like routine office work. The system must organize a surprise celebration for someone named Bob. What it does not know is that another person is already planning the same thing. Plus, the AI has lost track of the actual date. Success means reaching out to others, gathering clues, and then deciding together whether to team up or drop the idea.
Looking at it this way, those moments you once saw as time lost in chat apps might turn out to be some of your longest lasting contributions. Of course, unless the AI remembers your details too clearly and ends up letting future workers know you were the one who lost track of Bob’s special day.
Other Stories You May Like