Blend adds AI 'Autopilot' agents
Blend rolled out a major Autopilot update that uses AI agents to reach across the lending stack — pulling in credit, pricing and compliance logic to automate decisioning and operations. The change is a direct signal that platform competitors are embedding agentic automation into origination flows, affecting how lenders source speed and controls. (x.com)
A mortgage application still moves like a relay race passed by hand: one person checks pay stubs, another checks rules, another asks for missing documents, and the borrower waits in between. Blend said on April 8 that its new Autopilot update now lets artificial intelligence agents reach into the full lending stack instead of stopping at document review. (blend.com) That stack includes credit reports, pricing quotes, automated underwriting, compliance checks, disclosures, electronic signatures, title orders, loan conditions, and borrower messages. Blend said the new server is live in beta and gives those agents one standardized connection to all of it. (blend.com) Blend first launched Autopilot on March 3 as a narrower tool that reviewed borrower documents, checked them against lender rules, updated application fields, and generated follow-ups in as little as 15 seconds. The company said that first version was a non-decisioning tool, which means human underwriters and automated underwriting systems still made the credit call. (blend.com) The bottleneck Blend is chasing is simple: mortgage origination still takes 30 to 60 days on average and costs $11,000 or more per loan, according to the company’s March launch announcement. Blend also said 53% of borrower interactions happen outside business hours, which means a borrower who uploads a bank statement at 9 p.m. often gets no response until the next day. (blend.com) Before this week’s update, Autopilot mostly worked on documents, income, and follow-ups. After the update, Blend says the same agent can query the real pricing engine, pull the real credit report, run against the real compliance rules, and move work across the same platform lenders already use. (blend.com) That changes what “automation” means in lending. A rules bot can send a canned reminder email, but an execution agent can notice a deposit on a bank statement, connect it to a specific condition, ask for the exact missing explanation, update the file, and push the loan to the next milestone. (blend.com, housingwire.com) Blend says every document review is tied to agency, overlay, or custom guidelines with citations, so the system shows which rule triggered which finding. The company also says customers can turn Autopilot on in under one hour with a single toggle during the preview period, instead of rebuilding their process in a new portal. (blend.com) The competitive signal is that Blend is no longer pitching artificial intelligence as a sidecar that summarizes files. It is pitching a platform where agents can act across origination, and Blend said lenders and partners are already building their own workflows on top of the beta interface. (blend.com) This is landing in a market that is already looking for volume without adding headcount. Blend cited a National Mortgage News survey saying 57% of mortgage professionals expect artificial intelligence-driven underwriting to change the industry in 2026, while more than three-quarters expect origination volume growth this year. (blend.com) Blend has also tied the product directly to its own business story. In its March 10 preliminary 2025 results, the company said Autopilot was already live with large customers and framed it as a way to attack that $11,000 cost-to-originate number “head-on.” (morningstar.com) The practical question now is not whether lenders will use artificial intelligence in origination. It is whether they trust one vendor’s agent to touch credit, pricing, compliance, disclosures, and borrower communications inside the live loan file without slowing the controls that keep mortgages saleable. (blend.com, blend.com)
