Ryan Hanley, Founder and CEO of TASKD, joins AI InterConnect to explain why the next phase of enterprise AI depends on proof, not just answers. TASKD turns policies, regulations, contracts and operating procedures into deterministic AI systems that can explain every decision they make. As Ryan puts it, when an auditor asks how a decision was reached, the answer can't just be, I asked the LLM.
Lessons from regulated crypto markets
Before TASKD, Ryan spent years in increasingly regulated corners of the blockchain industry. At TokenMarket, one of the largest token issuers in 2017, his team built the infrastructure to distribute tokens to tens of thousands of buyers, working with the Gibraltar government and running KYC, anti-money-laundering checks and whitelisting on sales that could take in millions within hours.
He then moved into AML analytics at Coinfirm and institutional trading infrastructure at Elwood. Across those roles he sat in compliance meetings with officers whose licences depended on getting decisions right, and saw how much of the work was manual: gathering data from many sources, then writing reports to justify decisions already made.
That process was rarely consistent or deterministic. Ryan realized that as AI multiplies the volume of data and work, dependable, rules-based systems are the only way to run AI agents safely in regulated environments, especially as rules such as the EU AI Act require high-risk firms to show how they reach decisions.
A logic layer between data and agents
In simple terms, TASKD builds a decision layer. It sits between a company's messy data, spread across emails, spreadsheets, documents and tribal knowledge in people's heads, and the AI agents that act on it.
TASKD extracts the facts and the rules of how the business operates. Before an agent acts, that layer supplies the logic; afterwards, it checks the agent's output against the same rules. Because anything that passes through a language model becomes probabilistic, the check matters as much as the input.
Why today's approaches fall short
Ryan described three common approaches and their limits. Vector databases work well in a pilot with a few dozen documents, but at thousands of documents, full of contradictions, accuracy drops and costs rise.
The second approach is throwing agents at the problem, which often means paying for the same answers again and again instead of storing them once in a reusable structure. The third is hiring a firm to build a knowledge graph for you, which few companies can afford and which leaves the ontology and rules in someone else's hands.
TASKD focuses on structuring data deterministically and automating the rules, so that every answer has a traceable chain of reasoning that can be shown to an auditor.
Two days instead of years
Ryan walked through TASKD's work on the Alberta government's AI hackathon. The team combined records from 14 public databases into a knowledge graph, then automatically generated rules to trace relationships between people and the charities receiving public funds.
Rules such as the number of directorships one person holds, or the share of funds paid to directors, surfaced patterns worth a closer look. Ryan noted that this kind of analysis would usually take years, and TASKD completed it in two days using public data, with every finding traceable to its source record.
AI can create more work, not less
Ryan calls one side effect of AI the monkey's paw effect. Top performers become dramatically more productive, but less experienced people can produce work that looks right and isn't, and experts end up verifying far more.
In insurance, for example, AI has made filing a detailed claim almost free, so the people who must assess claims fairly now face several times the volume. Without rules and logic built in, faster output simply moves the workload to the people responsible for accuracy.
Flatter organizations and domain experts
Ryan expects organizations to become flatter as agents take on much of the checking and coordination that middle managers do today. He sees retraining those skilled people as a major gap that companies and educators have yet to address.
TASKD's own goal is to work alongside domain experts, starting in maritime, where knowledge is complex and largely unwritten. The aim is to capture how experts make judgement calls, so the company keeps that advantage, and to free those experts from manual research for the work only people can do.
Watch the full conversation
Watch Ryan Hanley explain why trust, explainability and deterministic decisions will become requirements, not options, for enterprise AI.
AI InterConnect hosts conversations with founders, investors and operators building with AI. Explore more episodes on the AI InterConnect podcast page.