How Heurist Finance built an AI-native investment workbench on Amazon Bedrock AgentCore
Amazon Web Services JW Wang
Heurist built a retail investing app on Amazon Bedrock AgentCore that buys premium data only when needed. It’s trying to make institutional-style research and portfolio tools work with per-query spending and audit trails.
Based on reporting by Amazon Web Services, JW Wang — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Heurist is using Amazon Bedrock AgentCore to turn a chatbot into something closer to an investment workbench. Its Heurist Finance product pulls in market data, filings, news, research, portfolio construction, stress tests, and position monitoring, then folds the result into one conversation that’s tailored to the user’s holdings and preferences.
The twist is economic as much as technical. Much of the data Heurist wants lives behind paywalls or bespoke APIs, and signing broad enterprise contracts before a product has scale is hard to justify. So the company buys premium data one query at a time. That sounds neat until the bill payer becomes an agent acting on behalf of a user, which means spending limits, custody rules, and audit requirements all have to be enforced in the flow itself.
AgentCore is the scaffold around that. Heurist runs its agents with Strands and Anthropic Claude on Amazon Bedrock, while AgentCore services handle identity, memory, code execution, observability, and payments. Portfolio data sits in Amazon Aurora PostgreSQL. Analysis artifacts go to Amazon S3. Traces land in Amazon CloudWatch. Credentials stay in AWS Secrets Manager. And the paid-data path uses AgentCore payments to tie requests to USDC settlement on the Base blockchain network.
The execution path is fairly strict. A merchant can respond with HTTP 402 and x402 payment terms; AgentCore payments checks the charge against a Payment Session’s maxSpendAmount, then signs the payment through a scoped embedded wallet if it fits. If it doesn’t, Heurist tells the user and suggests alternatives. Code Interpreter runs the analysis in an isolated sandbox with no arbitrary network egress, then tears down when it’s done.
Heurist says the payoff is a lot less plumbing. Roughly 80% less agent-system engineering than building its own orchestration stack, plus more predictable per-user marginal costs. More importantly, the company says compliance questions can be traced back through one audit trail, which is the sort of thing that sounds boring until it saves a team from building its own small bureaucracy in code.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of AI product: less demo theater, more plumbing with a purpose. Retail finance is already a mess of paywalls, permissions, and half-trusted tools, so pretending an agent can wing it is comedy. Managed infrastructure won’t make the finance world less paranoid, but it does stop every startup from rebuilding the same fragile control stack and calling it innovation.
Read more about this at: Amazon Web Services