On May 21, 2026, TD publicly detailed its first agentic AI application, built inside Layer 6 and running today across TD mortgage and TD Home Equity FlexLine applications. The model handles document scanning, income calculation, policy validation, consent checks, income verification, discrepancy search, and underwriter memo generation. PYMNTS measured the impact: the pre-adjudication queue, which used to average 15 hours before a human underwriter opened the file, now averages under 3 minutes. That is a 300x compression of one of the most reliably slow steps in Canadian mortgage origination.
Last week, we mapped Canada's three-layer regulatory stack: FINTRAC supervision under Bill C-12, FCA enforcement under Bill C-29, and OSFI prudential expectations tightening across credit risk and large exposures. This week, the competitive layer caught up. Underwriting speed and documentation discipline are now the visible operating expectations; the regulatory stack just defines the floor. For every mid-market bank, credit union, and non-bank lender in Canada, the strategic read is not "Big Six is doing AI." It is "the baseline just moved."
Per the TD Stories write-up, the agent is not a chatbot. It is a multi-step reasoning system that ingests a client's application package and produces a structured underwriting memo. Sandra Aziz, the Senior Machine Learning Engineer at Layer 6 who led the build, describes it as a system that "will summarize different information from different data points and provide the right information to a human." Crucially, every output sits inside TD's Trustworthy AI governance framework with human-in-the-loop oversight; the agent accelerates the underwriter, it does not replace them.
The mortgage launch is also a foundation, not a product. Aziz's team plans to extend the same agentic capabilities end-to-end across the mortgage process, from document submission to funding payout. Foundations from this first solution are already being reused in other areas of the bank. That is what makes the May 21 announcement structurally important: TD did not ship a feature; it shipped a platform pattern.
TD is the most visible datapoint, but it is not alone. BMO appointed its first head of digital assets and tokenization strategy on May 12. RBC, Scotiabank, CIBC, and National Bank all maintain dedicated AI research groups with public roadmaps. The Big Six are not piloting; they are integrating. Q2 earnings calls (May 27-28) will surface additional production milestones.
For the Big Six, the question is no longer "will agentic AI work in lending." It is "how fast can we scale the same pattern across HELOCs, lines of credit, small business, and credit cards." That answer is measured in quarters, not years.
Most mid-market Canadian banks and credit unions operate underwriting stacks built around rules engines, document OCR, separate fraud tools, and human review queues. They have ML models in specific areas (fraud, propensity, attrition), but no end-to-end, agentic orchestration of an application. They also do not have Layer 6.
That last point matters more than the first. The Big Six have 100 to 200 person AI research labs. Replicating that internally is not a viable strategy for an institution with $5B to $50B in assets. The realistic path is platform: buy or partner for the agentic underwriting capability, retain the policy, judgment, and customer relationship. Institutions that try to build a half-scale Layer 6 in 2026 will spend three years catching up while their borrowers route to faster lenders. The institutions that adopt platform-delivered agentic underwriting in 2026 will be at parity with the Big Six on the visible client experience within two renewal cycles.
Private lenders, MICs, and alternative mortgage originators sit on the most acute version of this challenge. Many operate underwriting today through spreadsheets, email threads, and broker-managed application packages. The 2026-27 OSFI Annual Risk Outlook named non-bank financial institutions the second-largest systemic threat behind real-estate-secured lending; the opacity argument is central. When borrower expectations are anchored to a 3-minute pre-adjudication at TD, the operational gap within the non-bank channel becomes visible to brokers, borrowers, and regulators simultaneously.
This is the segment FundMore built Fathom for. A freemium platform that ships AI underwriting, KYC, and identity verification (with FINTRAC obligations out of the box), document gathering and classification, fraud detection, OCR, property valuation, title search, compliance insurance, and customer engagement together; no upfront cost, public beta this summer. The thesis is direct: bank-grade tooling used to be the price of being federally regulated. It is now the price of operating.
Three readings worth taking forward.
Speed is the new disclosure. If a Big Six bank can produce a structured, auditable underwriting memo in under three minutes, regulators and brokers will increasingly treat the absence of one as a documentation gap. Documented decisioning is the new examination axis; the same shift the three-layer stack already pointed toward, now visible in production output.
The 12 to 18 month window is the operative timeline. Historical Canadian banking adoption cycles for major Big Six operational moves (mobile cheque deposit, e-statements, biometric authentication) ran 12 to 18 months before becoming table stakes across the rest of the market. The competitive grace period for "we have not gotten to AI yet" ends within that window.
Buy beats build for everyone outside the Big Six. The economics are not close. A platform that delivers TD-class agentic underwriting outcomes via configuration, with KYC and compliance baked in, costs orders of magnitude less than the internal build, and ships faster. The institutions that internalize this in 2026 will spend 2027 competing on policy and judgment; the ones that do not will spend it staffing AI teams.
1. Did TD actually replace its underwriters? No. The agentic AI accelerates pre-adjudication and generates a memo for the human underwriter; the underwriter still decides. TD's Trustworthy AI framework explicitly mandates human-in-the-loop oversight on every AI solution.
2. How much of mortgage underwriting does the model actually touch? Per TD's account, the current scope covers document scanning, income calculation, policy validation, consent checks, income verification, and discrepancy search, plus auto-generation of the underwriter memo. The roadmap extends it across the full mortgage lifecycle to funding.
3. Is this only for prime mortgages? TD specifically named both mortgages and the TD Home Equity FlexLine HELOC as live use cases. The same agentic pattern is being extended into other lending products at TD and is portable to alternative and private channels through platforms like Fathom.
4. What does this mean for brokers? Brokers route deals to whoever returns a credible decision fastest. A pre-adjudication that goes from 15 hours to under three minutes resets broker expectations across the channel; lenders without a comparable response time will see deal flow shift.
5. How does this interact with the new regulatory stack? Documented decisioning is now both a regulatory expectation (per FINTRAC, the FCA, and OSFI's credit risk direction) and a competitive expectation (per TD's operational benchmark). The two reinforce each other: a system that automatically produces a structured underwriter memo also produces an audit trail.
6. What should a mid-market or non-bank lender do this quarter? Map current underwriting decision time end-to-end, identify which steps are agentic-eligible (document parsing, income, consent, policy, discrepancy), and evaluate platform options. The Fathom waitlist is the fastest path to a bank-grade compliance and AI underwriting baseline without an internal build cycle.