Former eBay and Thomson Reuters CEO Builds AI Platform That’s Actually Working for Publishers.
Devin Wenig and John Stokes watched the internet hollow out publishing. They’re not letting AI repeat that pattern.
Most AI initiatives at publishers aren’t working. Every media company has a head of AI running teams of good people. They’re experimenting with tools, celebrating small wins, and producing interesting demos. But they’re not unlocking their organisations. There are zero examples of true transformation so far.
Devin Wenig and John Stokes think they know why.
Wenig ran Thomson Reuters out of Canary Wharf in London and, most recently, was President and CEO of eBay, based in their California HQ. Stokes founded Ars Technica, which he sold to Condé Nast, and is a former editor of WIRED. They both watched the internet revolution hollow out media companies’ profit pools.
“We both watched the internet revolution. I was inside Thomson Reuters. John had started his company, Ars Technica. That technology revolution was not friendly to the publishing industry,” Wenig said. “It hollowed out the profit pools from a lot of companies, a lot of media companies, a lot of news companies.”
They’re not repeating that pattern with their new AI service, Symbolic.ai, which was built for news organisations, corporate comms teams, and PR functions. The early customers span all three sectors, each with different needs but a shared problem: fact-checking content accurately while getting it out fas
The platform thesis
Symbolic brings native AI capabilities to professional publishers in media journalism, corporate communications, and public relations. The binding thread across those sectors: mission-critical, short-form, fact-based communication.
“We’re not really in the book business. We’re not in fiction,” Wenig explained. “What we’re doing is critical. Must be right every time. Short form communication for professionals.”
The platform automates production from inception through completion. But it’s not meant to replace professionals. It’s meant to unleash them. Wenig estimates that when he ran Reuters, roughly half the time of their 3,500 journalists went to creating actual journalism - thesis development, source development, interviews, angles. The other half went to production: writing, editing, applying style guides, and fact-checking.
“If we eliminate the 50%, you could save some money, but I think more importantly, you can apply that 50% back to generating more and better content and getting back to growth.”
Devin Wenig, Co-Founder of Symbolic.ai
That distinction matters. Most publishers approach AI as a cost-cutting tool. Symbolic positions it as a revenue growth engine.
The company has landed News Corp. as a customer, along with other large corporate and media organisations. Wenig thinks people will be surprised at how many of the world’s best companies have bet on the platform.
News Corp isn’t an investor. They’re a strategic customer who had other options. They chose to partner with Symbolic because it delivered results.
The fact-checking architecture
Most AI tools generate content and leave verification to humans. Symbolic.ai disaggregates every claim in an article and checks each one against designated sources.
During our conversation, Wenig demonstrated the system using an article about UK Prime Minister Sir Keir Starmer. Symbolic pulled out every statement, identified every claim within each statement, and then flagged each as supported or refuted. The system checks against research materials added by the user, trusted sites designated in advance, and the web.
One claim: Starmer suggested he should provide evidence to a US committee. Symbolic marked it supported and provided the Telegraph article that backed it up. Another claim: Starmer spoke to reporters during the G20 Summit. Symbolic marked it refuted, noting that several sources specified it was on the way to or on the flight to the summit, not during.
That level of precision matters in newsrooms, where one incorrect fact can mean getting fired.
“The ultimate decision of whether something is supported or not is the writer’s. It’s the editor’s. It’s not the machine’s,” Wenig said. “The machine is helping you make that decision. It’s not making that decision.”
The fact-checking process works differently for original content. Corporate communications departments use Symbolic to verify press releases and earnings scripts against internal data files. They add the raw materials - complex data files, earnings information - and Symbolic verifies the finished content against those sources.
“If their numbers are wrong, their numbers are wrong. We’re not verifying their sales,” Wenig said. “Once they add original materials, large, complex data files, then we will verify a press release and an earnings script against that.”
Use case of Symbolic.ai
Smart model routing
Symbolic uses five different AI models, both public and private. The company has disaggregated professional communication workflows into 26 discrete tasks, from research through publication. They test every available model against those 26 tasks.
No single model performs best on even a majority of tasks. So Symbolic routes each task to whichever model handles it best. The system knows what you’re doing and sends the request directly to the optimal model.
“We want the best. We want to route each task to the model that handles it the best way to get the best output,” Wenig explained.
The side benefit: publishers never become dependent on one model company. If OpenAI triples prices tomorrow, Symbolic turns that model off and routes to the second-best option. That’s diversity most publishers don’t have when they build directly on a single platform.
The smart routing also ensures continuous service. If one model becomes unavailable, the system automatically switches to another. Publishers can’t afford downtime when news breaks.
Voice and format separation
Symbolic separates format from voice. That’s a technical distinction that produces practical results.
Format is structure: long paragraphs versus short, newsletter style versus feature articles, and section organisation. Voice is the writer’s particular style: word choice, tone, and pacing. Most AI platforms munch those together. Symbolic found that it doesn’t get close enough to actual writing styles.
The system generates voice and format guidelines from exemplars - past writing examples. Organisations can upload style guides directly, but most don’t have formal guides. So they provide examples of their writing and Symbolic extracts the patterns.
The typical workflow involves iteration. Symbolic generates initial instructions based on exemplars. Writers review them, keep what works, change what doesn’t, and save the improvements. The entire organisation can iterate on saved instructions, creating what Wenig calls “AI institutional learning.”
“It’s not only one person using ChatGPT today, then it’s gone, then somebody else is doing their own experiment over there. This is one of the reasons people aren’t getting a lot of productivity,” he said. “The organisation itself can make those stored instructions better, constantly making them better, saving them so they’re available to everyone.”
If you have a colleague with a different format and voice, they apply their preset. Each writer maintains their style while benefiting from organisational improvements.
The corporate communications use case
A practical example: a CEO gives a speech. The small internal communications department can’t cover everything. Under traditional workflows, the speech might not get covered at all.
With Symbolic, the team records the speech on an iPhone and dumps the media file into the platform. Symbolic does high-fidelity transcription in seconds. The team has templates and voices for internal communication (how they talk to employees) and external communication (how they talk to investors).
They immediately generate both versions from the transcript. The whole process takes three minutes. They make 10-20% changes to their liking. From something that might not have gotten covered, they’ve produced two high-quality articles about the CEO’s speech, fact-checked against what the CEO actually said.
Why AI initiatives aren’t succeeding
Publishers are trying to bolt AI onto existing processes. They’re doing discrete little things. They celebrate small wins here and there, but they haven’t unlocked their organisations.
“They’re not succeeding because they, in general, are trying to bolt AI onto existing processes,” Wenig said. “They’re doing discrete little things. And they’ll say, “ Hey, we have a win over here, and maybe they sort of have a small win, but they haven’t unlocked their organisations. There are zero examples of it so far.”
The problem: you can’t unlock an organisation with discrete experiments and cool demos. You need a platform that holds context, eliminates layers, and changes the way you work. AI is going to change these organisations. The question is whether they take control of that change, or if it’s done to them.
“This is not a path to deliver 60, 70% productivity. That’s what has to happen. Not 5%, not cool demos, not sometimes it works, not like we’re doing this one little thing on the side with AI.”
Symbolic’s customers had large teams experimenting with AI before adopting the platform. They realised they weren’t getting very far. What they needed was a comprehensive solution that understood publishing workflows end to end.
Growth without sales or marketing
Symbolic has no marketing function. No sales team. The company is growing through Wenig and Stokes’s network and word of mouth. All money goes into engineering and product excellence.
“I want to build a great product. I want really happy customers. I want to grow by word of mouth,” Wenig said. “No hype, no nonsense, no slop. I want satisfied customers. And we’ll grow. If we can do that, the word will get out.”
The company has customers beyond News Corp that haven’t been announced yet. The company is expanding internationally. They have international customers now and believe this is a global platform, not just US-focused. The roadmap for this year includes features driven directly by customer requests.
Whether that approach scales remains to be seen. But early customer results suggest Symbolic has found something most AI publishers haven’t: a platform that actually delivers on the promise of transforming how professional communicators work.
The question for publishers isn’t whether AI will change their organisations. It’s whether they’ll control that change themselves, or watch it happen to them as it did with the internet revolution.
Wenig and Stokes are betting that this time, publishers can come out ahead.
