In 1981 I was a law clerk buried in products liability discovery, and the firm had just bought a Pitney Bowes word processor. I taught it to produce the repetitive half of my job. Nobody called that legal tech. It was just a clerk who hated the tedious, literal cut-and-paste of interrogatories, using the best machine in the building.
That is the only trick I know, and I have been running it for 45 years: when tools get good enough to do the next layer of knowledge work, build the thing that puts them to work.
The receipts
By 1983 I was the lawyer, and I wrote a dBase system to track shipments to Superfund waste sites. In 1986, I built the Parallax Asbestos Case Management system, which Owens-Corning, FibreBoard, and dozens of law firms across the country used to run their dockets. Three legal systems in five years: each moved software deeper into work lawyers were supposed to do.
In 1992 I launched Books.com, the first online bookstore, before people had heard the word “browser.” Barnes & Noble eventually acquired it. Selling books was the point; proving commerce would live on the internet was the lesson.
In 1999 I founded Flashline, a software component management platform, on the bet that code you already own is an asset worth managing. Oracle ultimately acquired it, and it remains an active product at the core of its Oracle Enterprise Repository. Something I shipped a quarter century ago is still doing its job, which is the quietest proof I have.
In 2009 I launched Sideways, a digital book publishing platform. We never found traction and closed in 2011. Being early feels exactly like being wrong while it is happening, and only demand tells you which one it was. Sideways taught me not to argue with that.
In 2013 I founded the Flashstarts Accelerator, a series of small venture capital funds, each backing its own cohort of very early-stage startups. In 2015, I opened StartMart, a 35,000-square-foot coworking space in Cleveland’s landmark Terminal Tower, on a lease based entirely on profit sharing, because I had watched WeWork’s numbers and structured our lease for the case where I was wrong. Both closed in December 2019, at the end of that lease, months before Covid would have made the decision for me. Reading hundreds of pitches and watching dozens of portfolio companies live or die is where my scoring instincts stopped being opinions.
At the same time, I became a member of the North Coast Angel Fund, and for a while, served on the screening committee.
In 2026 I founded Coworkers.Global, an AI staffing agency that places managed agents into real roles.
The punchline I did not plan
I have already been in the coworker business once. The 2015 version was humans and desks; I was the landlord. The 2026 version needs no desks, and the job is not landlord but manager: provisioning, training, supervising, and standing accountable for the work.
The loop closed tighter than I expected. The law clerk who automated discovery in 1981 now ships Devorah, a managed paralegal agent. Alex, our first agent, evaluates business plans the way I learned to across five startups, an angel fund, and an accelerator. The pattern never changed. Only the tools did.
Current view, subject to change
Here is the opinion under all of this: for 45 years, capability was the scarce thing, and each wave rewarded whoever wired the new capability into real work first. I think that era just ended. The models are commodities; what is scarce now is management, someone who provisions the capability, checks its work, and answers for it. That is the bet Coworkers.Global makes. What would change my mind is evidence that unmanaged agents hold up inside real accountability structures, in front of clients, regulators, and deadlines. Nothing I have seen in 45 years of watching new tools meet real work suggests the supervision layer goes away. It just changes who does it. And that is why this business exists.
If you want to see the current tools at work, Alex reads business plans at coworkers.global/ai-business-plan-review. The read is free until July 27.
Regards,
Charles Stack