Agent built a multi-source KPI dashboard in 10 minutes

An artificial intelligence agent just built a working business dashboard in about 10 minutes by pulling data from advertising platforms, web analytics, customer records, payments, and a database into one live view. The demo came from Cody Schneider, whose site says he is the cofounder of Graphed, a company focused on dashboards and insights. (x.com) (codyschneider.com) Most small software companies already have the raw numbers for revenue, leads, signups, and ad spend. The problem is that those numbers usually live in separate systems like Google Analytics for traffic, HubSpot for customer relationship management, Stripe for payments, and internal databases for product events. (developers.google.com) (developers.hubspot.com) (docs.stripe.com) That split creates a simple but expensive reporting problem. A founder can open five tabs and still not know whether a $2,000 ad campaign produced paying customers or just low-quality traffic, because the click data, lead data, and payment data are stored in different places. (developers.google.com) (developers.hubspot.com) (docs.stripe.com) A dashboard is supposed to solve that by putting the key numbers on one screen. In practice, building one usually means wiring application programming interfaces, cleaning field names, joining tables, handling missing records, and deciding which source is the source of truth for each metric. (developers.google.com) (developers.hubspot.com) (docs.stripe.com) That plumbing work is what people mean by extract, transform, and load. It is the process of pulling data out of one system, reshaping it into a common format, and loading it into a place where reports can query it. (github.com) (docs.stripe.com) The demo is interesting because the agent appears to have done that plumbing automatically instead of waiting for a human analyst to map every field by hand. According to the shared description, it created data pipelines, assembled a knowledge graph, and used memory and evaluation loops to keep the dashboard coherent while it connected multiple sources. (x.com) A data pipeline is the conveyor belt in this setup. Stripe’s own documentation says its Data Pipeline can export payment data to systems like Snowflake, Amazon Redshift, Databricks, Amazon Simple Storage Service, Google Cloud Storage, and Azure Blob Storage, which shows how payment events can be moved into a shared reporting layer. (docs.stripe.com) Google Analytics offers a similar machine-readable path on the traffic side. Google says its Data Application Programming Interface can be used to build custom dashboards, automate reporting tasks, and integrate analytics data with other business applications, which is exactly the kind of feed an agent would need for sessions, page views, or campaign performance. (developers.google.com) HubSpot provides the customer relationship management layer that links leads, companies, deals, and custom objects. HubSpot’s developer documentation says its platform is built for custom customer relationship management and data-driven experiences, so an agent can treat the sales pipeline as another structured data source instead of a separate manual report. (developers.hubspot.com 1) (developers.hubspot.com 2) The knowledge graph piece is the map that tells the system how “the same thing” appears in different tools. If one source says “customer,” another says “contact,” and a third says “account_id,” the graph can help the agent understand that those labels may refer to related business entities that should be connected before a chart is drawn. (x.com) The retrieval-augmented generation part is less about flashy text generation and more about grounding answers in the right data at the right moment. Amazon Web Services describes retrieval-augmented generation as a setup where the model references an authoritative outside data source before generating a response, which is useful when the answer depends on current business records instead of old training data. (docs.aws.amazon.com) The memory and evaluation loops matter because dashboards fail in quiet ways. If an agent joins the wrong date field, double-counts refunded revenue, or mixes ad clicks with signups from a different time window, the chart can look polished while the number is wrong, so repeated checks are part of making the output usable. (x.com) (docs.stripe.com) (developers.google.com) For a small business software company, the appeal is speed more than novelty. A team that would normally spend days connecting Google Analytics, HubSpot, Stripe, and database tables could instead describe the metrics it wants in plain language and let an agent assemble the first version of the reporting stack. (developers.google.com) (developers.hubspot.com) (docs.stripe.com) (x.com) That does not mean the hard part disappears. The hard part shifts from writing every connector by hand to setting permissions, defining metric logic, checking joins, and deciding which numbers deserve executive trust. (developers.google.com) (developers.hubspot.com) (docs.stripe.com) The larger point of the demo is that product-facing analytics is starting to look less like a six-month business intelligence project and more like an orchestration problem. When an agent can retrieve data, transform it, connect entities, and render a live dashboard in one workflow, the bottleneck moves from coding the dashboard to choosing the right business questions. (x.com) (docs.aws.amazon.com)