Blend Autopilot Week 6 Update: A Redesigned Pre-Underwriting Summary, MCP Platform Expansion, and Smarter Follow-Up Creation

Subscribe to Autopilot updates

Your Pre-Underwriting Summary just got smarter.

The Pre-Underwriting Summary was one of the first things Autopilot shipped. When lenders told us how they actually use it, a pattern emerged: they skip straight to what needs attention. The old layout organized by data category. The new one organizes by action. That redesign, built directly from lender feedback, is the centerpiece of this week’s update, alongside continued expansion of the MCP platform and a smarter follow-up creation pipeline.

Pre-Underwriting Summary redesigned for action-first review

The Pre-Underwriting Summary has been one of the most popular Autopilot features since it was shared in Week 1. This week it gets its biggest redesign: a complete restructure around how lenders actually use it.

The new layout is action-first. Instead of organizing by data category, the summary opens with what matters most, focusing on what needs attention and what to do about it.

Executive brief

The summary now opens with a one-glance overview that shows traffic-light status (green, yellow, red) across every dimension of the loan: income, assets, credit, compliance, property. An underwriter can open the summary and immediately know where the loan stands before reading a single detail.

Blockers and next steps

Directly below the executive brief is a priority-ordered list of every open item and required action, sorted by severity. If there are three things blocking this loan from moving forward, they’re the first three things the underwriter sees.

AUS overview

For loans with automated underwriting results, the summary now includes a dedicated AUS section showing the recommendation, key conditions, and their current status as met, not met, or partially met. When Autopilot’s AUS findings agent is active, the analysis it performed is reflected directly in this section.

Seven structured sections

The full summary now covers: executive brief, blockers and next steps, AUS overview, income and employment, assets and reserves, credit and liabilities, compliance audit, and document log. Each section uses consistent formatting with guideline citations and actionable detail.

The PDF rendering is rebuilt with a modern HTML-based pipeline. No external font dependencies. Cleaner layout. Faster generation. Borrower-facing views automatically exclude sensitive sections like the AUS overview and compliance audit, so the same summary infrastructure serves both internal and borrower-facing use cases.

MCP platform expansion one week after launch

Last week we shipped Autopilot MCP, giving AI agents programmatic access to the full Blend lending platform through one standardized interface. One week later, the platform is already more capable and more reliable.

New MCP capabilities for AI agents

Loan search. Agents can now search across loans by filter criteria, not just operate on a single loan they already know about. This opens up portfolio-level workflows and cross-loan analysis.

Borrower communications rewritten. The email and SMS tools were rewritten to match the Lending API specification, making borrower outreach from agents more reliable and consistent.

Enriched request catalog. When agents work with borrower requests and documents, they now have richer context. Cover descriptions, entity IDs, and document types are all part of the request schema. The agent understands not just that a request exists, but what fulfilling it looks like.

MCP reliability and tooling improvements

The core infrastructure connecting the MCP server to the Lending platform now includes automatic retry on transient failures, PII masking in all log output, and structured error handling. The team also corrected field mappings across the data model (HMDA fields, REO fields, pricing domains) and fixed tools across credit, pricing, borrower invite, MISMO export, and eSign. A systematic smoke test harness now runs against every tool, catching issues before they reach production.

The MCP server shipped seven days ago. It’s being tested on real workflows by lenders building their own AI-powered integrations, and it’s maturing fast.

Follow-up, guideline, and quality improvements under the hood

The agent is getting more accurate and more rigorously tested every week:

• Smarter follow-up creation: A two-layer dedup system now checks every follow-up before creation, using fast text matching for obvious duplicates, then LLM semantic comparison for nuanced cases. Fewer duplicate requests reaching borrowers and fewer tasks for loan officers to clean up.
• Custom guideline accuracy: The agent now skips irrelevant prompt fragments in custom guideline mode, building on last week’s fix. Cleaner, more accurate reasoning for lenders running portfolio products and proprietary overlays.
• Deeper quality gates: New evaluation scenarios for gift letter workflows, follow-up resolution validation, and system-triggered follow-ups. The automated test suite that runs against every change before it ships is broader than it’s ever been.

Six weeks in, Autopilot is maturing. Not just adding new capabilities, but going back and making the core experience better. The summary that underwriters open on every loan is now faster to scan, clearer about what needs action, and structured around how lenders actually work. Every agent intelligent improvement ships automatically to every lender with Autopilot activated. No upgrade cycles. No implementation projects.

---

Blend Autopilot is currently in preview and free to activate and use during the preview period for all Blend customers. To get started, visit your Lending Config Center or contact your Blend account team.

We publish a new update every Wednesday. Subscribe to the Autopilot weekly update to make sure you stay up to date with everything we are shipping.