Cult Products

The episode contains a handful of genuinely useful insights - the >0% error rate argument against AI-driven allow/deny decisions, and ring fencing as a pre-existing solution for agentic AI containment - but these are buried under repetitive marketing framing, personal origin stories, and company culture anecdotes. The 'deny by default' mantra is repeated so often it dilutes the episode's signal.

“anything greater than 0% is less good than default deny. So default deny basically is going to block everything. Okay, good, bad, doesn't make any difference.”

“we have product called Ring Fencing... designed five years ago, six years ago, long before agentic AI was a thing. But the cleverness about it is it's exactly the solution that's needed for these agentic AI tools.”

The deny-by-default/application whitelisting philosophy is not new to the market - it's the foundational pitch for products like Carbon Black and others - and most of the framing is polished marketing rather than first-principles thinking. The ring fencing-as-agentic-AI-guardrail connection is a genuinely fresh and timely observation, earning the episode from a low score.

“we experimented with using AI to categorize unknown websites first... it got it right somewhere in the region about 85% of the time. About 85% of the time, we got it right. 15% of the time, we got it wrong.”

“50% miscategorization of malware basically, would be catastrophic because all of our customers would be wiped out by ransomware pretty much overnight.”

Rob Allen is a legitimate CPO at a real, scaling cybersecurity vendor with 750 staff and 65,000 customers, and his 18 years in MSP work gives him genuine practitioner credibility. However, the interview is structurally a vendor promotional piece, and his insights largely stay within ThreatLocker's own product narrative rather than drawing on broader industry experience.

“I spent 18 plus years working for a MSP IT company basically in Ireland. So fairly significant proportion of those 18 years were spent trying to clean up after ransomware attacks”

“Today, I think we're about 750 staff... I think we're currently somewhere about 65,000 companies around the world are protected by Threat Locker.”

The episode delivers a reasonable number of concrete data points - malware volume, AI attribution estimates, accuracy rates, headcount, customer count - which is better than average for a vendor interview. However, claims about competitive differentiation, marketing scale, and product superiority are largely asserted without third-party evidence or customer outcome data.

“we get it right somewhere in the region about 85% of the time... 15% of the time, we got it wrong. And we'd need a human being basically to check those results”

“we do I think over a thousand events a year at current count”

The host is personable and well-prepared, and asks one sharp contextual question about AI differentiation versus large models. But most questions are open-ended prompts that function as on-ramps for marketing monologues, with no pushback on vendor claims, no probing of failure modes beyond what the guest volunteers, and no challenge to the product's limitations or competitive weaknesses.

“I suppose when you say we still move like a startup, what are some of the tangible examples of that and how is that possible at 700 people?”

“it's also brave to go with it. Also, we were caught out by it and our leading security experts were as well.”