Piers Fawkes spent 20 years running PSFK, a trends and innovation research firm, until ChatGPT arrived and effectively replaced his own job description. Rather than fight it, he built Foda, a knowledge graph platform that turns proprietary research into an MCP connector any LLM can query. He joined Deconstructing Data to talk about what happens to startups, data, and research once the hyperscalers can do almost everything themselves.
Piers Fawkes has a blunt way of putting it: ChatGPT launched in December 2022, and it took his job. Not metaphorically. He ran a trends and innovation research company for two decades, and suddenly a chatbot could answer “what’s the next trend” with a convincing-enough answer for free.
Instead of treating that as a threat, Fawkes spent the next three years figuring out where a human research business still adds value once the model itself is commoditized. His answer became Foda, and the conversation is a useful read for anyone building on top of LLMs right now, because the same commoditization pressure is coming for a lot of software, not just research firms.
Why are AI wrapper startups so vulnerable right now?
Fawkes built several products in the last few years, including a PDF-chat directory and an “Event Mind” tool that mined conference talks for insights, before running into the same wall repeatedly: whatever he built, OpenAI or Anthropic would eventually ship a native feature that did the same thing.
“I had to continue always think about how could I change, how can I evolve and how can I add something maybe onto the off-ramp or the on-ramp to the LLM system rather than try to kind of create my own highway.
— Piers Fawkes, Founder, PSFK and Foda
That’s a sharp way to frame the whole “wrapper startup” problem. You’re not competing with the hyperscalers. You’re building on their highway, and if your product is just the highway with a coat of paint, they’ll eventually pave right over you.
“If you are a wrapper, and you’re good, you’re going to be spied on… you’re going to be either acquired or they’re going to come and take your product.
— Piers Fawkes, Founder, PSFK and Foda
His conclusion, and it’s one worth sitting with if you’re building an AI product today: durable value has to come from something the model itself can’t generate on its own. For Fawkes, and honestly for BDEX too, that’s proprietary data.
What are knowledge graphs and why are they replacing vector databases?
Fawkes describes himself as new to knowledge graphs, half-joking that he feels like “the teenager who thinks they’ve invented sex” when in fact the technology has existed for decades. What’s new is applying it to the AI experience directly.
“Everybody could scrape the web and create knowledge graphs around any topic, but I used the 20 years of experience to help kind of filter as a lens to create these knowledge graphs.
— Piers Fawkes, Founder, PSFK and Foda
His working definition: knowledge graphs are databases structured specifically to help AI find answers faster and spot patterns between data points faster than the vector databases most of us have been using for retrieval-augmented generation over the past year or two. He’s building hybrids of both rather than picking one camp.
The bigger idea he’s chasing is that LLMs don’t need to know everything to be useful.
“We don’t need LLMs to know everything… we’ll have these machines which might get dumber and lighter, but have access to external data sources, including knowledge graphs.
— Piers Fawkes, Founder, PSFK and Foda
That’s a real inversion of how most people think about model progress. Instead of chasing bigger models trained on more of the internet, the more interesting frontier might be smaller, cheaper models paired with curated, proprietary, constantly updated external data. It’s the same logic driving the shift toward small language models trained on narrow domains instead of general-purpose giants trained on everything.
How do you make proprietary research usable inside an LLM?
This is the part of the conversation with the most immediate, practical application. Fawkes built Foda so that a marketing executive working inside Claude, ChatGPT, or another LLM can pull in Foda’s research library directly through an MCP connector, no separate dashboard, no separate login, just a key pasted into the tool they’re already using.
“Connectors and MCP is just like it’s changed my world… it’s going to democratize access to a whole bunch of services within the LLM experience.
— Piers Fawkes, Founder, PSFK and Foda
Before connectors existed, Fawkes assumed he’d have to sell into a chief data officer or an IT team and get his product formally installed. Now any user, a consultant, a corporate marketer, a head of IT, can add a single MCP URL to Claude in a couple of clicks and start querying. That’s a meaningfully lower sales barrier for anyone building a data product right now, and it’s part of why BDEX has been building out its own MCP server so users can query BDEX data directly from the LLM they already work in.
Fawkes is also building Foda as a marketplace rather than trying to cover every vertical himself. He’s inviting former competitors, journalists, and report publishers to convert their own research into knowledge graphs that plug into the same system, supplemented with roughly 30 to 40 API integrations from institutional sources like the World Trade Organization.
What’s the actual workflow for turning raw research into usable trend data?
Fawkes’ practical advice for teams doing trends or competitive research is refreshingly low-tech at the entry point. Start with a shared bookmarking tool. He uses Feedly, with a team account that captures interesting finds along with algorithmic searches across Google and Reddit for relevant keywords, all flowing into one central database the team scans weekly.
“I first I now do agentic coding 100%, but I only learned really about the process through no code and low code.
— Piers Fawkes, Founder, PSFK and Foda
From there, he suggests connecting no-code tools like Zapier to route that bookmarked content through an LLM for analysis and summarization, and only moving to agentic coding once you understand the steps you’re actually trying to automate. It’s good, grounded advice: understand the manual workflow before you try to have an AI agent run it for you.
The bigger picture
What I found most interesting about this conversation is how directly Piers experienced the thing a lot of us are still theorizing about. His entire business model got replaced by a chatbot, in real time, and instead of arguing that AI wasn’t “really” doing his job, he went and built the thing that makes AI better at his job.
The throughline for Foda, and honestly for knowledge graphs more broadly, is that the value isn’t in having access to a model anymore. Everyone has access to the same models. The value is in what proprietary, structured, well-curated data you can connect to that model. That’s true whether you’re a 20-year research veteran building a knowledge graph marketplace, or a data company connecting identity data to whatever platform your customers are already using. The models are the commodity now. The data is the moat.
To learn more about Piers Fawkes, PSFK, and Foda, visit foda.ai or psfk.com.
For companies that need clean identity data to power smarter decisions, visit bdex.com and click “Talk to an Expert.”
Watch the Episode
AI Trends, Knowledge Graphs, and Startup Data
This article was adapted from an episode of Deconstructing Data, BDEX’s weekly podcast on data-driven marketing. Tune in live every Thursday at 4:15 PM Eastern on LinkedIn.
About BDEX: For companies that need clean identity data to power their products, BDEX offers unmatched quality and execution. Visit bdex.com and click “Talk to an Expert” to get started.
