Will O'Brien, founder and CEO of Provenonce, argued that enterprise adoption of AI agents is waiting on a way to verify their work, and traced a 2014 essay on machine-to-machine commerce to the Bittensor subnet he is building to close what he calls the assurance gap.
Podcasts ·
Will O'Brien, founder and CEO of Provenonce and a co-founder of BitGo, argued on the episode AI Agents Have a Trust Problem. Can Provenonce Fix It? (3 November 2026), recorded at Exploit Summit in Montreal, that enterprise adoption of AI agents is waiting on trust infrastructure: agents' output is set to outpace anyone's ability to check it, and the missing assurance layer is what keeps enterprises from trusting them. He is building Subnet 87, a Bittensor subnet, to close that gap.
O'Brien named the problem the assurance gap. As companies adopt more agentic systems, he said, "that digital exhaust, which is the output of agentic work, is going to outpace the actual analysis and assessment of the quality of that work" (from 0:04:47). His example: after step changes in customer experience such as Zappos's free returns, a company lets agents set its customer service policy and they make the wrong call repeatedly: "Customer and consumer expectations are so high and you're losing so much money that you have a catastrophic collapse." (from 0:08:59)
He presented Provenonce as the culmination of a February 2014 essay, which said widespread adoption would not be just humans sending crypto to each other: "it's going to be machines, machines as payers, machines as what we now call agents, machines with wallets." (from 0:03:00) The stage model came from the same essay: "We needed mining before we could get to liquidity. We needed liquidity before we could get to institutionalization, institutionalization before mass adoption" (from 0:19:00) The essay put Bitcoin, after the collapse of Mt. Gox, at the start of the third stage. Bittensor, he said, has proven out its mining and validating and is now at the same point.
What is missing is a trust layer: "there's a brand new trust layer that's not yet built. The trust of whether the agent is authorized, what the agent is scoped to do, what is the evidence that's gathered during that transaction, and what's the review and replay from learning." (from 0:03:47) He traced the lineage from Bitcoin's proof of work to multisig custody to Bittensor, and put assurance next: "BitGo delivered trust infrastructure with multi signature and other companies and other organizations, other networks are delivering. BitTensor is delivering trust infrastructure with its network." (from 0:09:58) His summary is short: "I have this statement that adoption is waiting for trust infrastructure." (from 0:09:40) His forecast for what follows: "And if we solve it, that means the enterprise is going to come in droves to BitTensor." (from 0:09:58)
On Subnet 87, he described customers handing over an assurance capsule, the evidence of what the AI did and the goal it was given, and miners competing to generate an assurance differential that can be assessed. He also warned the economics can go the other way: "the economic incentives can drive either a virtuous cycle or can drive a cycle where the subnet suffers." (from 0:05:48) Asked whether systems like Bittensor guard against an Elysium-style dystopia, he granted the argument merit and added: "I will say the greater risk is ignorance and that can often come from fear." (from 0:17:18)
Provenonce's own whitepaper, Proof of Assurance: An Open Bittensor Market for Evidence-Backed Assurance of Autonomous Work, is dated 21 September 2026 at version 0.7.2 Final Candidate, by O'Brien and Chris Zacharia of bitstarter.ai. The paper carries a section titled The Assurance Gap and defines the Assurance Capsule, the Assurance Differential, and the challenge and validation roles. In the paper, validators issue hidden, class-specific challenges, and each differential declares findings, no material deviation, or insufficient evidence: Provenonce: Proof of Assurance whitepaper.
The project's public repository records the build-out O'Brien described as recent. The readme states that SN87 is live on Bittensor testnet 582, and that an attested series of 21 runs starts at block 8,152,574 on 5 October 2026. The scorer and checks are open under the MIT License, and the readme says the weighted miners are all operated by Provenonce, so the results so far are conformance evidence, not a contest between methods: sn87-provenonce-public README.
An independent on-chain analysis of subnet 87, based on a chain read of 8 September 2026, reported no repository, no testnet, 246 of 256 registered UIDs earning nothing, all miner emission going to the subnet owner's own hotkey, and the technical roadmap and testnet running about two months late. The dates reconcile it. The whitepaper is dated 21 September 2026 and the repository's first attested testnet runs are dated 5 and 6 October 2026, so the analysis describes a subnet registered and funded but not yet built: SN87_PROVENONCE_ANALYSIS.md.
This page was written from an episode in Alphaday’s podcast feed. Its record, with the audio file, the episode page and a description, is one call away. Free, no signup, no key.
Start building with the podcasts API
Track what the shows say about a project, from each episode's description.
The podcast feed as tools, for Claude, Cursor and any MCP client.
The stories the shows are reacting to.
Quotes are from machine transcripts of the episodes, with misheard names corrected, and each timestamp is where its passage starts. What a guest says is their claim; the sources linked beside it are what it was checked against.