Everyone's obsessed with getting "good data." But here's the uncomfortable question: what happens when your AI has accurate data and still makes the wrong decision?
In this episode of The STEMINISTS podcast, host Phoebe Goh welcomes back AI data expert Darnell Fatigati for a conversation that challenges one of the biggest assumptions in AI today: that data quality is enough. Spoiler alert: it ain’t.
As AI agents move beyond answering questions and start making recommendations, triggering workflows, and taking action, the stakes get a lot higher. A data point can be technically correct and still completely miss the bigger picture. Without critical context, AI isn't making informed decisions. It's making confident guesses. And these guesses turn into mistakes that scale fast.
Phoebe and Darnell unpack why data context, trust, governance, and even organizational memory are becoming essential ingredients for successful AI. They explore how businesses can help AI understand not just what happened, but why it happened, so agents can act responsibly instead of accidentally creating expensive chaos at machine speed.
Key Takeaways:
- Why clean data doesn't automatically lead to smart AI decisions
- The difference between data quality and data integrity, and why it matters more than ever
- How context turns information into understanding and understanding into action
- Why organizational memory may be your next competitive advantage in AI
- Practical steps to make your data more AI-ready without boiling the ocean
Plus, Darnell shares the story of an AI that managed a professional baseball game surprisingly well... until one very human variable showed up and reminded everyone that context still matters. ⚾🤖
Because in the age of agentic AI, it's not enough for your data to be correct. Your AI needs to understand the assignment.
Like what you hear? Follow and share The STEMINISTS Podcast with your network. The future of AI won't be built on more data alone. It'll be built on better understanding. 🚀
Learn more about the Symbiotic relationship between AI and Data:
https://www.netapp.com/blog/symbiotic-relationship-data-and-ai/
Check out how AI is only as good as the data that fuels it:
https://www.netapp.com/blog/ai-is-only-as-good-as-the-data-that-fuels-it/
Connect with us!
https://www.linkedin.com/in/phoebegoh/
https://www.linkedin.com/in/darnellfatigati/
-----
Let's Solve IT!: Darnell Fatigati on why agentic AI needs more than clean data
Episode overview
In this episode of "Let's Solve IT!," host Phoebe Goh talks with returning guest Darnell Fatigati about what changes when AI stops answering questions and starts taking action [0:00] [0:24]. Fatigati's central argument: agentic AI raises the bar for data, because systems that recommend decisions and execute tasks need more than accurate information. They need context [1:07] [1:23].
Fatigati draws a sharp line between data quality, whether information is correct, complete, consistent, and current, and data integrity, whether that information can be trusted in the moment it's used [2:07] [2:18]. A revenue figure can be completely accurate and still mislead an AI agent if it doesn't account for a sold business unit, a discontinued product, or a mid-quarter pricing change [2:18]. Without that context, Fatigati argues, very different business situations can look identical to an AI system [3:01] [3:14].
The conversation moves from definition to consequence. Fatigati explains that agentic AI doesn't just produce poor reports when data is flawed, it takes poor actions, and those actions can scale and propagate before anyone notices [4:05] [4:24]. To ground the argument, Fatigati cites a real-world example: an independent minor league baseball team that handed a full game to an AI manager. The system got almost every call right, until the starting catcher fell ill and the AI lacked the one current fact that mattered most [11:08] [11:27] [11:31] [11:54].
Key takeaways
The following points capture the practical guidance Fatigati offered for organizations preparing data for agentic AI.
- Agentic AI raises the standard for data because systems are increasingly asked to recommend decisions and take actions, not just retrieve or summarize information [1:07] [1:23].
- Data quality (correctness, completeness, consistency, currency) is necessary but not sufficient; data integrity, whether information can be trusted in context, is what determines good outcomes [2:07] [2:18].
- Context transforms raw information into understanding, and its absence can make distinct situations look the same to an AI system [3:01] [3:14].
- Agentic AI scales the consequences of poor data, since mistakes can propagate across business processes before a human intervenes [4:05] [4:24].
- Agents need comprehension, not just access: they must know which sources are authoritative, how current the information is, what permissions apply, and when to escalate to a person [6:28] [6:45].
- Giving an agent access to every company document doesn't create expertise, the same way giving a new employee access to every shared drive doesn't create expertise [6:28].
- Organizations tend to preserve what decisions were made but lose the reasoning behind them, and that reasoning is increasingly valuable for training future AI systems [7:37] [7:48].
- The recommended starting point for AI readiness is understanding existing data, not collecting more, beginning with a small number of high-value use cases [8:48] [9:40].
- Knowledge graphs can help organizations connect distributed enterprise information and identify authoritative sources [9:16].
- When organizations get data context right, AI moves past isolated productivity gains and becomes a participant in driving business outcomes [10:13].
Topics and themes
This episode centers on what enterprise data needs to look like as AI shifts from answering questions to taking action.
- Context as the missing layer in enterprise AI: what data means, where it came from, how current it is, and what policies govern it [1:23] [2:18] [3:01] [3:14] [5:29] [6:45].
- The gap between data access and data comprehension, for AI agents and for the organizations deploying them [6:28] [8:48] [9:00].
- Trust as a prerequisite for autonomous action, grounded in visibility into agent reasoning, permissions, and controls [5:29].
- Agentic AI as a distinct risk category, where repeated, scalable actions change the consequences of bad data [4:05] [4:24].
- Organizational memory as an emerging priority: capturing the trade-offs and assumptions behind past decisions, not just the outcomes [7:37] [7:48].
- Knowledge graphs as a mechanism for connecting enterprise information and surfacing authoritative sources [9:16].
- Real-world agentic AI deployment outside the technology sector, illustrated through the minor league baseball example [11:08] [11:27] [11:31].
Notable quotes
- "We're increasingly asking AI to recommend decisions and to actually take actions. That really raises the standard for the data that's underneath it." [1:07] [1:23]
- "Context is really what transforms that information into understanding." [3:01]
- "As agents begin to recommend and execute actions, this context becomes the difference between an informed decision and just a fast, confident mistake." [3:38]
- "Access for data isn't the same as comprehension of that data. Just giving an agent access to every document in the company doesn't make it an expert any more than say giving a new employee access to every shared drive." [6:28]
- "The agent wasn't wrong, the data wasn't bad. It was reasoning very capably over information, but the information didn't reflect current reality." [11:54]
- "It's more now about having AI that can understand the business well enough to help actually drive business outcomes." [10:13]
