By Stephanie Simone
</span>
<span itemprop=”>
How do you separate the signal from the noise and explore what’s truly working in AI implementations across industries, and where companies are falling short?
From the common pitfalls that derail well-intentioned efforts to the design principles behind AI initiatives that create measurable value, Ryan Frederick,principal,Transform Labs, shared the lessons that every leader needs to understand to make AI a competitive advantage rather than a costly experiment during his session, “Learnings From the Front Lines of AI,” at Data Summit 2026 in the new Data + AI Leadership Forum.
The annualData Summitconference returned to Boston, May 6-7, 2026, with pre-conference workshops on May 5.
New for 2026, theData + AI Leadership Forum is an exclusive space for business and technical leaders to explore strategy, governance, responsible AI, and value realization.
“I think most companies are asking the wrong question and going for the wrong objective,” Frederick said.
Five pitfalls Frederick sees everywhere include:
- Pushing adoption before the foundation: Adoption without foundation is just busy. Stand AI up as a capability, then drive workflow automations.
- Confusing pilots with production: If it doesn’t get operationalized it doesn’t prove much or bring value.
- Chasing tools, not building a system: A pile of tools is not an operating system.
- Running AI as projects, not an operating layer: Projects end, operating layers compound. Build the layer your company runs on.
- No one owns the outcome: When everyone is accountable, no one is. Pick one name and give them authority.
“We’re seeing conflicts inside companies where CFOs are asking what value did it bring to the business and the answer is none,” Frederick said. “This is not a technology initiative; this should be a business initiative.”
According to Fredrick, what works is:
- Anchor to a line on the P&L: Name the line and how do you expect AI to affect that.
- Go at the core, not the edges: Go where the operational friction lives.
- Build it modular, not tool-dependent
- Trust is earned twice, by two different people: Executives have to trust the output and outcomes from AI. The team has to develop trust within the system.
- One person owns strategy, execution, and outcomes
“Your team has to believe the system allows them to do their best, high-level work,” Frederick said. “AI is a leadership problem wearing a technology costume.”
AJ Meyers,principal solutions architect,Elastic, built on his keynote speech with his session, “Retrieval: The Key to Agentic Success.”
Enterprise AI agents are only as intelligent as what they retrieve—yet most organizations are investing heavily in models while ignoring the layer that determines whether those models reason accurately, act on current knowledge, and deliver distinctive results.
Retrieval is the layer that decides what context an agent receives before it reasons or responds.
Five capabilities the retrieval layer must have includes:
- Multi-model orchestration
- Real-time ingest
- Query-time access control
- Retrieval observability
- Multi-modal and AI-model ready
“You have to refactor what your core capabilities are,” Meyers said.
Many Data Summit 2026 presentations are available for review athttps://www.dbta.com/datasummit/2026/presentations.aspx.