How Rehinged got to the point where it could build this.

On almost every call lately, someone asks how a company that started as a data science shop in 2018 ended up running on agents. It's a fair question and a longer answer than a call allows, so I'm writing it down, one era at a time. Here's the first.

I started Rehinged in 2018 as a venture-funded data science company in San Diego. The first dataset we worked with was Crunchbase startup data, through a venture firm's API access, and the job was analyzing market segments and startup performance for the fund and its portfolio companies.

Working with that data wasn't hard. Enriching it with public sources was. Our first ingestion was hand-built Python scrapers, 500 lines of code that took days to build and would break. I realized quickly that wasn't scalable, and challenged the team to look globally for engineers and solutions that could acquire the datasets our clients needed at scale, cost effectively, without ingesting the entire US market.

The other problem was people. The thesis was that we'd hire senior data scientists in San Diego and supplement them from UCSD's new undergraduate data science program. A year in, the average data scientist salary had doubled what we'd projected, and Amazon and Facebook were hiring our top talent at three times what we could afford. That forced a decision about the business model.

Plugging those datasets in around the markets we were analyzing was the start of the foundation. It's what the agents work across today, for companies selling into skilled nursing and the rest of U.S. healthcare.

A head of R&D at a large pharma company told me, close to word for word: Jim, we're just not ready to trust AI analysis of a large corpus of material. It's not good enough to stake our reputation on. We want to, but we can't.

That was 2020. The models did the work, just not reliably enough. When the pandemic hit we ran our natural language processing models across the COVID-19 research corpus and pulled sentiment on specific topics at a scale that would have taken humans months. For client projects we'd ingest large sets of customer and influencer videos about a technology and present the sentiment to product and go-to-market teams.

The skeptics would look for the one result they disagreed with and point to it as proof the AI didn't work. Sometimes that was protecting a job. Sometimes they'd found a real outlier. Accuracy wasn't close to 100%, and they weren't wrong to notice. A model that performs in a demo and fails on the ninth facility is the thing we spent those two years learning to distrust.

What we could fix was the economics. I moved from a local in-house team to specialized global resources, run by an experienced AI product manager who could source and manage engineers in Asia, Eastern Europe, the US and Latin America. That aligned our costs with what clients would pay. The trust problem had to wait for the models.

Everybody in tech knows the date. December 2022, and the accuracy question that pharma R&D head had raised was answered well enough for certain use cases.

We wired the new model into our 1.0 platform within weeks. It was an API call at the end of the day's run, generating messaging and talk tracks for the healthcare entities we'd scored. It was monumental to see it written. The results were OK, not great.

But we saw where it was heading, and that the models would improve. The question had changed from whether the models could do the work to whether anyone had built the thing the models needed to work across. The data platform was the expensive part, and we had already paid for it. The finished work is what the economics made possible.

In 2023 and 2024 the engine worked and the business model still didn't, and the reason was people, ours and the client's.

Delivering our first platform to healthcare commercial teams took a lot of human work. We had human-based sales processes. We had to integrate with the workflows of everyone from the VP of commercial to the sales manager to the reps to the SDRs to the marketing lead, move the data around, refresh it. The technology scaled. The human side cost too much to.

So we took the engine to larger projects. We licensed the scoring to partners and applied it to their data in volume. One project had us scoring over 200 million objects for various audiences, for targeting and for marketing and sales messages. It worked, and even for the partner the economics didn't. The audiences that needed the intelligence most didn't have the budget to cover the infrastructure, the data scientists and the technical team required to operate it, deliver it, maintain it and improve it.

That revenue funded the rebuild of the healthcare foundation for what came next. The low point of those two years was knowing the work was right and the economics weren't, and not yet having the thing that fixed it.

At the beginning of 2025 we realized we could build agents to do the backlog of work our engineers couldn't get to, and the work clients wanted but couldn't budget for. If we didn't embrace it we wouldn't survive.

The first agent handled part of client onboarding, which until then meant multiple calls and meetings with me and our product leads before any revenue came in. When it produced two weeks of work in ten minutes at the same or better quality it was clear agents belonged in most of our processes and in our client interaction.

We looked at hiring agentic experts. They were scarce and expensive and the space was moving too fast, so I learned it myself: four to six hours a day, every day, building while learning, while the engineers got the platform ready for the agents. The only non-engineer in the company was the one building them. Painful and exhilarating in equal measure.

The team had wanted to hire five more engineers for the next version. We built the agents instead. We didn't use them to cut headcount; we used them for work we never had the bandwidth for. What people keep came out of that: the methodology, the judgment on what a signal means, the review before a client sees anything, and the conversation with the client. AI carries every part of the business it can. People hold the parts it can't.

What Rehinged is now, in my words: decades of hands-on go-to-market work with marketing and sales teams, in B2B and in healthcare, built into agents that run on our own data and check it against every outside source we can find, so a company creating a healthcare, healthtech or wellness product can get what I know, quickly and at a cost that makes sense.

What that produces: ranked accounts, reasons to call, contacts and draft outreach, battlecards, messaging and territory prep, built around what you sell. Not a data platform.

Every era in this series is still in the building. The 2018 data work is the foundation. The 2020 lessons are the review. The 2022 shift is the economics. The 2023 rebuild is why the outputs hold. The 2025 production model is how the work gets done. There's no way we could deliver this without the eight years before it.

I know some people are opposed to agents doing human work. I choose to be an optimist about it: it keeps driving costs down so more people can live better and businesses can do more, and change is still hard for most of them.

If you sell into U.S. healthcare and want to see it on your own market, tell us what you sell and we'll build your Market Draft. Free, and nobody calls you.