At Data Science Connect's Agentic AI in Action Q2 webinar, Lumi AI, Precisely, SAS and Domino Data Lab each gave a 10-minute demo. They showed agents that surface inventory insights, enrich addresses through MCP, turn decision flows into tools and keep regulated research auditable.
Key takeaways
- Lumi AI runs analytics in the background. Its agents alerted an inventory planner to overstock and recommended where to transfer excess units.
- Precisely enriches data in one prompt. Through its MCP server, an agent validated a mistyped address and pulled property, business and neighborhood context.
- SAS turns decisions into agent tools. A restocking flow published through Viya MCP servers let Claude return an auditable recommendation.
- Domino keeps agent work traceable. Coding agents work under the same access controls and audit trail as human data scientists.
- Context and controls are the common thread. Every demo paired agents with knowledge bases, trusted data or governance.
Four vendors showed how AI agents are moving from chat windows into analytics, data enrichment, decisioning and regulated research at Agentic AI in Action Q2, a Data Science Connect webinar of 10-minute demos brought to you by Lumi AI, Precisely, Domino Data Lab and SAS. Presenters were Ibrahim Ashqar, Co-Founder and CEO of Lumi AI; Carmen Adame, Senior Manager, Product Marketing at Precisely; Marinela Profi, Global Market Strategy Lead for AI and Agentic AI at SAS; and Nick Jablonski, Field CTO at Domino Data Lab.
A common thread ran through all four: agents are only as useful as the context, data and controls around them. Here's what each demonstrated.
Lumi AI: How can an agent surface analytics insights proactively?
Ashqar described Lumi as an agentic system that connects to ERPs such as SAP and Oracle or warehouses such as BigQuery, Redshift and Snowflake, then runs in the background against objectives users set in plain language. "The question is no longer whether analytics should be agentic, it's how quickly you can operationalize that," he said.
His demo started with an email alert to an inventory planner about excess inventory at one store. The insight showed the full prompt, assumptions and generated code, which Ashqar said is meant to build trust. He then asked which items were driving the overstock, and followed with a harder request: for each item, find the best location to absorb the excess. Running across what he described as roughly 100 million records, Lumi returned a list of items, unit counts and destination stores. Ashqar said that kind of rebalancing analysis would normally take an analyst days to weeks.
Behind the answers is a knowledge base that holds tables, joins, business terms, KPIs and tribal knowledge, plus "memories" built from users' thumbs-up and thumbs-down feedback. A "magic wand" agent can build the semantic model automatically in 15 to 20 minutes, he said. Lumi starts customers with ROI-based pilots because, Ashqar said, "we want choosing Lumi to be a math decision, not a leap of faith."
Precisely: What does agent-driven data enrichment look like?
Adame framed the problem as an agentic AI data integrity gap. Agents are getting better at reasoning, but reasoning on siloed, incomplete or inconsistent data "produces really confident wrong answers," she said.
Her demo, run through a trial of Precisely's MCP server, took a mistyped merchant address at a Denver mall and a single prompt asking for validation, property and business data, and nearby points of interest for card transaction enrichment. The agent corrected and geocoded the address with what Adame said was a 98% confidence score, recognized it as an alias, and attached a Precisely ID to carry downstream.
It then pulled parcel and building data, aerial imagery from partner GeoX, the 39 businesses at the location, 30-day foot traffic, neighborhood demographics, crime context and tax jurisdiction, along with the vintage of each dataset. Adame said this kind of context helps financial services teams with fraud categorization without manual lookups.
SAS Viya: How do agents combine LLMs with deterministic decisions?
Profi said the shift underway is from "an AI that only informs to an AI that acts," which makes accountability, transparency and monitoring essential before humans step back as the quality-control layer. She showed three SAS Viya capabilities: Viya Copilot, Viya MCP servers that expose SAS capabilities as tools for outside agents, and the SAS Agentic AI Accelerator, which she described as a free, open-source framework for building agents.
Using grocery store data, Copilot imported CSV files, generated Python to add store latitude and longitude, joined tables, and built a dashboard with new variables for daily sales rate and days of supply, plus a sales forecast. Profi said Copilot runs inside the customer's Viya environment and doesn't expose private data to an LLM.
She then published a restocking decision flow from SAS Intelligent Decisioning as an MCP tool. A store manager asked Claude what to do about a restock alert for limes at a Dallas store. After asking permission, Claude called the SAS flow and recommended restocking about 1,040 limes at medium urgency, tying the demand bump to a local margarita festival. It also produced an audit-ready explanation of which rules and events drove the decision. Profi cited deployments at National Bank of Greece, handling customer service requests in Greek, and at a European ministry of health supporting incident response.
Domino Data Lab: How do you keep agentic work auditable in regulated research?
Jablonski said Domino has long been a code-first platform for expert data scientists building mission-critical models, and it now extends the same reproducibility and traceability to agents. "We're not the keeper of a single sort of Domino way of doing things," he said, describing the platform as an orchestration layer for tools like VS Code, RStudio and Jupyter.
His demo used a pharmaceutical real-world evidence study, where results face heavy regulatory scrutiny. Inside a Domino workspace, a coding agent such as Claude, Codex or Copilot can use Domino-specific skills to access data, record results and publish apps, under the same access controls and audit trail as the human user. Each job captures the code commit, inputs, hardware and tables accessed.
The coding agent also built a React app for stakeholders that logs decision rationale, separates agent-written from human-written decisions, and links results back to their data, job and code. An AI data explorer let users ask questions that the agent answered with SQL against the output table. Jablonski said the same pattern applies to insurance pricing, fraud detection and defense use cases.
See more agentic AI demos in 2027
Agentic AI Showcases are one of four series on Data Science Connect's 2027 calendar of 24 free live webinars, alongside AI Demo Days, the AI × Enterprise Panel Series and the Plug & Play Series. Browse upcoming sessions on the webinars page.
Frequently asked questions
What is agentic analytics?
As Lumi AI's Ibrahim Ashqar described it, an agentic system connected to company data that understands its data model and business terms and proactively surfaces issues and opportunities in plain language.
How do MCP servers help AI agents use enterprise data?
In the demos, Precisely's MCP server let an agent validate and enrich addresses, and SAS Viya MCP servers exposed analytics and decision flows as tools any agent can call.
How can AI agents be audited in regulated industries?
Domino Data Lab showed agents working under the same access controls as users, with every job's code, inputs and data captured, and agent-written decisions logged separately from human ones.
How does SAS combine LLMs with deterministic models?
Marinela Profi showed Claude calling a SAS Intelligent Decisioning flow of business rules and analytics, then explaining which rules and events drove the decision.
Watch: Agentic AI in Action Q2: Demos from Lumi AI, Precisely, Domino Data Lab & SAS | Webinar Replay on YouTube
Event: Webinars · More coverage: Webinars insights
