Five vendors gave 10-minute live demos at Data Science Connect's Agentic AI in Action showcase. Informatica, PromptQL, Teradata, Precisely and Lumi AI showed agents that find, clean and enrich data, answer business questions and keep humans in the loop.
Key takeaways
- Informatica automates data prep. CLAIRE agents discovered data, checked permissions, built a pipeline and cleaned a file from natural-language prompts.
- PromptQL targets confident wrong answers. It states its assumptions and builds a wiki-style semantic layer from team conversations.
- Teradata's analyst agent lives in Gemini Enterprise. It ran a multi-step hospital readmission analysis inside Vantage with per-user access controls.
- Precisely adds location context. Its Geo assistant verified addresses, added a Precisely ID and suggested enrichment data for a pipeline.
- Lumi AI turns alerts into actions. Agents traced excess inventory to specific items and recommended which stores should absorb it.
Five vendors showed how AI agents handle real enterprise data work in "Agentic AI in Action: 10 Minute Demo Spotlights from the Frontier," a Data Science Connect virtual showcase hosted by co-founder and president Brian Mink. The session was sponsored by Informatica and PromptQL as headline sponsors, with Teradata, Precisely and Lumi AI. Each presenter had about 10 minutes for a live demo. Here's what each one showed.
How does Informatica's CLAIRE agent find and prepare enterprise data?
Sandeep Kumar Karkera of Informatica played a marketing analyst at a fictional company who needed the top countries by order volume. Working in natural language, he asked Informatica's CLAIRE agents to find retail customer datasets. A discovery agent searched the enterprise catalog, showed its plan and reasoning trace, and asked for confirmation when it found several copies.
He then pulled the asset's data quality score, glossary terms and lineage. Before returning a data sample from Snowflake, the agent checked whether he had permission to see it. "Of course, it is end-to-end autonomous, but human is in the loop and human is in the driving seat," he said.
A single prompt then had a data integration agent build a mapping, Informatica's term for a data pipeline, that could run every month. A data quality agent cleaned a messy customer file, for example standardizing "MSTR" to MasterCard, and produced a reusable component for future pipelines.
How does PromptQL avoid confidently wrong AI answers?
Rob Dominguez, engineering manager at PromptQL, opened with a question: "Which of these is worse? An AI tool that says, 'I don't know,' or one that gives you the wrong answer?"
His demo, on a synthetic retail dataset, showed a collaborative interface with public and private threads, where teammates can see and join each other's conversations with the agent. Before answering, PromptQL lists the wiki pages it drew on and states the assumptions it is making.
Those wiki pages form a semantic layer built mostly through conversation, Dominguez said. When the agent finds something undocumented, such as an unexplained column, users can add it to the wiki. Users can also tag a colleague for expertise. Users can also edit pages directly and review version-by-version diffs to see who changed what.
"The majority of AI initiatives that are failing out there are failing because of people getting these confidently wrong systems," he said.
What did Teradata's analyst agent show?
Trinath Yarlagadda, principal solution architect in Teradata's consulting services, presented an analyst agent built with Google Cloud and deployed inside Gemini Enterprise. Computation runs inside the customer's Teradata Vantage environment, and an orchestrator built on Gemini models routes each part of a request to specialist agents for SQL and Python work. Access is enforced per user, he said, so the depth of an answer depends on who asks.
Playing a hospital administrator with no database knowledge, he asked why readmission rates for COPD, a chronic lung condition, stayed high. The agent reported 23.9% for COPD patients versus 15.6% for everyone else. It also ruled out medication adherence and access to a primary care doctor, and pointed to smoking status combined with insurance coverage.
Follow-up questions led it to transportation barriers among patients eligible for both Medicare and Medicaid, then to cost estimates for interventions and a pilot design.
How does Precisely make location data ready for AI agents?
Ruslan Sultanov, senior product marketing manager at Precisely, said "the agents are only as good as the data that they operate on." He described what Precisely calls the agentic AI data integrity gap, between the data enterprises have and what autonomous systems require.
In the Precisely Data Integrity Suite, its assistant Geo helped a telecom user find a table of serviceable addresses in Snowflake and build a pipeline in natural language. Geo suggested an address verification step, added a Precisely ID as a consistent join key and recommended enrichment datasets for network infrastructure risk. Sultanov called it a semi-autonomous agent that keeps a human in the loop.
"A lot of the time it's just the context that we're missing," he said.
How does Lumi AI use agents for inventory analytics?
Lumi AI's founder and CEO, Ibrahim Bashar, started from an email alert flagging rising excess inventory. The insight showed its prompt, assumptions and code. "There is no black box here," he said.
Drilling into one outlier store, he asked which items drove the excess and where each could be moved. Mid-analysis the query hit an error, and a built-in diagnostic agent retried and fixed it on a dataset he put at about 100 million records. The result was a transfer list showing units and receiving stores.
Bashar said context "in this age of agents is essentially infrastructure." Lumi's knowledge base lets admins choose tables and fields, define joins and business terms, and commit user feedback to memory.
What's next for Agentic AI Showcases
Agentic AI Showcases are one of four series in Data Science Connect's 24 free live webinars on the 2027 schedule, alongside the AI × Enterprise Panel Series, AI Demo Days and the Plug & Play Series. The AI Enterprise Conference returns on Tuesday, August 31, 2027, at Pier Sixty in New York City.
Frequently asked questions
What does an agentic AI data management demo look like?
In Informatica's demo, agents discovered datasets, showed lineage and quality scores, checked access rights, built a pipeline and cleaned a file, all from natural-language prompts with a human confirming steps.
How can AI agents avoid giving confidently wrong answers?
PromptQL's approach is to state assumptions, show which wiki pages it relied on, and let teammates correct and add context that is saved for next time.
Can business users run data analysis with AI agents without SQL?
Teradata and Lumi AI both showed agents answering plain-English business questions by planning multi-step queries and returning charts and recommendations.
Why does context matter for AI agents?
Several presenters said agents are only as good as the data and context they get, from location attributes to business definitions and feedback stored in a knowledge base.
Watch: Agentic AI in Action: 10 Minute Demo Spotlights from the Frontier on YouTube
Event: Webinars · More coverage: Webinars insights
