Dustin Smith wants publishers to know who’s reading

Dustin Smith has a pleasantly inconvenient lesson for founders: an interesting product still needs somebody who wants it. Hum originally pursued professional associations. Publishers proved more receptive, using the software, finding value in it, and offering ideas for what should come next. Smith’s verdict on the experience was brisk: pivot sooner. There is an entire business bookshelf that could have saved itself the paper.

That willingness to revise the plan runs through his work. Today, Smith is Hum’s co-founder and CEO, based in Charlottesville, Virginia. His subject is publishing, although the questions he keeps returning to extend beyond journals: how do you understand the people arriving at your door, and what do you offer once they get there?

He calls himself a philosopher turned technologist. The combination has its uses. Philosophy encourages questions about meaning; publishing technology supplies an alarming number of databases in which meaning has been mislaid.

The reader has been divided into departments

Consider the reader who also attends a conference, reviews manuscripts, and takes a professional course. Within an organization, those activities can belong to separate systems and separate teams. Each department has a perfectly respectable fragment of a person. Putting the fragments together is the difficulty.

Smith addressed that problem in an October 2023 article written with Ann Michael, Laura Harvey, and Andrew Smeall. His contribution described societies with substantial information about events, membership, learning, and journals, much of it marooned in disconnected platforms. Having the data was only the beginning.

The practical possibilities were specific: recommend a relevant event to a journal reader, shape a newsletter around actual interests, or notice subjects that attract attention but have little published material. In this view, better information should change an encounter. Another dashboard is a rather expensive ornament if nobody does anything differently afterward.

An optimist who checks the instructions

Smith’s enthusiasm for AI comes with an interest in how experiments are conducted. In August 2023, he challenged a published test in which ChatGPT had supplied invented research papers. He argued that asking an offline model for recent literature beyond its knowledge cutoff made failure considerably more likely.

He tried alternatives with internet access and revised the instructions. His proposed prompt specified the role, context, task, and desired output. The point was methodological: a verdict about a tool depends on which tool was used, what it could access, and what it was asked to do.

“Give the robots a fighting chance!” he wrote. The phrase catches his temperament rather well: encouraging, slightly mischievous, and interested in the machinery. His argument concerned those particular 2023 experiments. It offers a useful portrait of his habits without turning an old test into a promise about every answer a model might produce today.

“Give the robots a fighting chance!”

Dustin Smith, 2023

The quieter business of understanding

By January 2024, Smith was making a related argument about the kind of AI attracting attention. Generating prose had become the conspicuous party trick. Understanding a publisher’s existing material deserved a place at the table, too.

His essay on BERT distinguished generative systems from interpretive models used to represent and understand text. He explained embeddings as numerical representations that make it possible to compare material by meaning. For a publisher, that can support recommendations or help retrieve relevant content before a generative model produces an answer.

The distinction matters to the portrait of Smith as a builder. He was arguing for infrastructure underneath the demonstration: organize the material, make it searchable in useful ways, and choose models for their jobs. The charming answer on the screen arrives last. Considerable housekeeping precedes its entrance.

He also discussed the costs and practical limits of running large collections through generative models. Those calculations belonged to the technology available when he wrote them. The continuing question in his work is how a promising capability becomes a usable publishing service.

Letting Alchemist out of the kitchen

Hum introduced its Alchemist AI engine in June 2023. At the time, Smith was the company’s co-founder and president. The launch described an engine working within Hum’s customer data platform, using language models and other techniques to interpret content and audience information.

In March 2025, Smith explained a broader product direction. Capabilities developed inside that platform would become available as separate tools. The proposed range covered organizing content, understanding audience activity, searching collections, and helping editors assess manuscripts.

He chose a kitchen analogy to explain the underlying work: gather the ingredients, clean and sort them, understand the combinations, prepare something useful, and learn from the response. It is an unusually domestic description of an AI system, and rather more helpful than announcing that the future has arrived wearing a lanyard.

The expansion also made his ambition easier to examine. A publisher could consider a particular workflow rather than treating an entire platform as the only entry point. His imagined future remained expansive, but the products gave individual publishing teams concrete tasks to discuss.

A catalogue with a practical problem

A taxonomy can sound like something designed to punish a dinner guest. In publishing, it is the system of topics and labels that helps a collection behave like a collection rather than a cupboard. Smith’s work becomes more tangible when that cupboard contains hundreds of thousands of articles.

In June 2025, Hum announced that the American Society for Microbiology had selected Alchemist Taxonomy for its sixteen journals. The organization used more than 260,000 articles to develop a custom taxonomy; Hum reported that the taxonomy build took about two weeks.

The announcement also described applying tags across the back catalogue and connecting the process with existing publishing systems. The operational detail is revealing. A tool has to enter an organization’s working life, where other software is already installed and people already have responsibilities.

Those figures describe that deployment. They make a particular piece of work legible, without requiring a prediction about every publisher or every archive. For Smith’s wider argument, the example supplies a useful answer to the inevitable question: what does this actually do?

The title catches up with the work

On August 6, 2025, Hum announced Smith’s appointment as CEO. He succeeded Tim Barton, who stepped away from day-to-day operations and continued as an adviser. Smith had already been guiding product development and the company’s expansion into content and audience intelligence.

The change put a broader remit around familiar work. Andrea Hoffman, Hum’s chief operating officer, highlighted his product instincts and attention to customers. Smith said he intended to remain closely involved with customers and the team while expanding the platform.

There is a recognizably practical tension here. A CEO has to make room for strategy, growth, and the organization around the product. Smith’s stated intention was to keep contact with the people using it. The market pivot had already given him a reason to value those conversations.

The answer arrives. Who keeps the relationship?

By November 2025, Smith was writing about a different front door to publishing. Readers increasingly begin with a question and expect an answer, rather than a list of links and the afternoon free to investigate them.

His argument was that publishers need material that answer systems can find, interpret, and attribute. Clean metadata, citations, and connections between ideas matter. So does the experience on the publisher’s own platform: a useful answer can become the beginning of a relationship if the reader has somewhere worthwhile to continue.

That concern joins his earlier interest in audience data to the changing mechanics of discovery. A publisher might reach someone through an outside answer engine and still learn very little about that person. Smith wants publishers to think about what happens after the answer, including how readers explore related work and return.

It is a commercial proposition, certainly. It is also a revealing preoccupation for someone building publishing software: the encounter between a collection and an individual deserves as much attention as the collection itself.

A little less solemnity, please

Smith’s public biography makes room for trail running and being a father to daughters. His Hum team-page aside is more occupational: “When I’m not humming I’m probably thinking about Hum.” An efficient answer, though perhaps a warning to anyone hoping to change the subject over dinner.

The humor appears elsewhere. When Hum colleagues experimented with fabricated research papers during a team exercise, including a taxonomy of cheesecake, Smith responded on LinkedIn with a joke about what happened while he took paternal leave. The exercise concerned a serious publishing problem, but the team had found a wonderfully unserious specimen.

His public work in October 2026 brings him back to company-building itself: he is listed as guest co-host for a conversation about scholarly communication startups, alongside Lou Peck and three founders. Pivots and institutional friction are on the agenda. Having learned to change direction at Hum, Smith has something more useful to bring than a tidy origin myth. He has a revised plan, and an interest in what people actually use.