Agents in action
LoanPro's AI servicing assistant cuts account summaries from 3–5 minutes to under 15 seconds
In an AWS blog post co-written with LoanPro, the lending platform describes a multi-agent assistant built with AllCloud. It runs on Amazon Bedrock AgentCore and the Strands Agents SDK and helps servicing agents get up to speed on a borrower's account during live calls.
What the agent did
- Pulls borrower data from 8 API endpoints and condenses around 1,400 fields per account into a narrative summary.
- Writes that summary in under 15 seconds, versus 3–5 minutes of manual screen navigation before.
- Replaces 5–7 clicks of orientation with one click, saving 1 to nearly 2 minutes per call.
- A composer agent sends each request to a specialist agent, either for summarization or for choosing the next action.
- Met a 99% accuracy target on a benchmark of 150 accounts labeled by LoanPro's subject-matter experts.
What AgentGid data shows
- Strands Agents: 8,913,553 weekly package downloads as of 2026-11-17, down 2.1% over 30 days.
- Strands Agents: 9,233 GitHub stars, #10 by Gid Score among agent frameworks and SDKs (#42 overall).
- Strands Agents is open source (Apache-2.0) and free. You pay only for the models and infrastructure it runs on.
What it means for you
- Trying the framework costs nothing. Model calls and hosting, such as Bedrock, are where the costs come from.
- Other open-source frameworks in the same category to compare: LangChain, LangGraph, OpenAI Agents SDK, Mastra and Agent Development Kit (ADK).
- In a regulated setting, plan for what LoanPro added around the agents: role-based access checks, audit logging and an expert-labeled evaluation set.
Why it matters
Agents spent 18–20% of every call just getting oriented, so cutting that step affects every call, not only the hard ones. The design splits work across small specialist agents, so each one can be tested and tuned separately. That may make expanding into other workflows easier, and LoanPro says it plans to move into collections and origination next. Accuracy was checked against expert-labeled accounts, and every AI action is logged. That is the kind of control lending compliance tends to require.
The results are as reported by LoanPro and AllCloud in AWS's own blog post, the call-time cut is an 'up to' figure, and AgentGid has not tested any of these numbers.
Source: AWS Machine Learning Blog
Numbers about what happened come from the source; numbers in "What AgentGid data shows" come from our tracking (methodology). Figures a vendor reports about its own product are as reported, not tested by us.
Comments
- Tried something like this? Share how it went for you.
Agents in this story
| # | Agent | Gid Score | Downloads 7d | 7d | 30d | Stars | Latest release | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| 10 | 55 | 8.9M | up 0.8% | up 0.2% | 9,240+12/day | 1.65.05d ago | |||||
More agents in action
Agents in actionGitHub's agents ported the Copilot runtime to 800,000 lines of Rust for about $120,000 in tokens
Agents gone wrongAn OpenAI model broke into a government statistics site for one bed count and took 24 years of data
Agents in actionGrab's agents ran about 150 unattended experiments and made a dispatch simulator step 24× faster