If there was one underlying tension driving this conversation, it was this:
AI is moving fast. Organizations are not.
At this MTLC Tech Hot Topic at Capital One, senior operations leaders dug into what it actually takes to redesign workflows for AI—not in theory, but inside real organizations with messy data, legacy processes, and human constraints.
And the takeaway wasn’t that companies need more AI.
It’s that most aren’t structurally ready for it.
The Shift: From AI Adoption → Operational Reckoning
Doug Levin (Harvard & Stanford) set the tone early with two ideas that stuck with people:
- The concept of AI “half-life”—where tools, models, and strategies are decaying faster than organizations can adapt
- And the “jagged frontier”—where some teams are operating at a completely different level of AI capability than others inside the same company
That second point sparked a bigger realization throughout the morning:
AI isn’t scaling evenly—it’s fragmenting organizations.
The result? Internal “cognitive debt” where advanced teams are pulling ahead, while others fall further behind.
Where It Gets Real: What Practitioners Actually Said
The practitioner panel—featuring Urvashi Batra (Prioriwise), Nicholai Mitchko (InterSystems), Sanjay Yadav (Parlance), Paul Baier (GAI Insights), and Brian McMahon (Capital One)—moved quickly past surface-level takes.
Here’s where things got interesting:
“The biggest mistake? Bolting AI onto broken workflows.”
Multiple speakers pushed back on the idea that AI is a simple efficiency layer.
“If you don’t understand the workflow, you shouldn’t be automating it.”
The real work isn’t picking tools—it’s:
- mapping how work actually happens
- identifying where decisions break
- redesigning flows before automation ever happens
This was reinforced again and again:
AI exposes bad process faster than it fixes it.
Everyone is experimenting. Almost no one is truly “AI-native.”
One of the most honest distinctions that came up:
- Using AI: copilots, chat tools, productivity boosts
- Being powered by AI: workflows that run through AI
Most companies are still firmly in the first category.
And the blocker isn’t technical.
It’s:
- unclear ownership
- siloed teams
- no shared definition of “what good looks like”
As one speaker put it, “people are optimizing inside their box—and hitting a wall the moment they have to collaborate.”
Organizational readiness is now the real bottleneck
This showed up everywhere—and not in abstract ways.
Examples that landed:
- Teams want to automate—but don’t trust the data
- Leaders want transformation—but default to incremental tweaks
- Employees have access to AI—but no clarity on when or how to use it
One comment that stuck:
“It’s like wanting to get healthy—you know what to do, but are you actually willing to do it?”
That gap between intent and execution is where most organizations are stuck.
The token debate revealed something deeper
What started as a tactical question (“Should we measure token usage?”) turned into a philosophical divide.
Some leaders are already dealing with real cost overruns—unexpected spikes in AI usage.
But the stronger takeaway was this:
Optimizing for cost too early can kill innovation.
Teams that are over-managing usage:
- explore less
- experiment less
- ultimately learn slower
The more mature view emerging:
- Early stage = invest in learning
- Later stage = optimize for efficiency
Not everything should be AI—and that’s becoming a real issue
This was one of the sharper moments in the discussion.
As enthusiasm grows, so does overuse.
“Don’t use probabilistic AI for problems that require exact answers.”
Examples:
- finance and accounting workflows
- high-risk compliance decisions
- areas requiring full explainability
The implication:
Part of AI maturity is knowing when not to use it.
AI agents aren’t the problem. Definitions are.
A surprisingly lively thread emerged around one simple question:
What actually counts as an “agent”?
Answers varied:
- autonomous systems making decisions
- prompt + tools
- workflow-level automation
- product features
That lack of alignment matters.
Because without shared definitions:
- teams build different things under the same name
- expectations misalign
- governance becomes harder
This is still a moving target—and leaders know it.
One of the most practical takeaways: start with bottlenecks
Across company sizes—from startups to enterprises—the most effective approach wasn’t broad transformation.
It was targeted:
- find a bottleneck
- deeply understand it
- automate that
From customer success workflows to engineering pipelines, the pattern was consistent:
“Where there’s friction, there’s opportunity.”
Lightning Talk: A Shift in How to Think About the Problem
The closing lightning talk from Anand Swamy and Poornima Ramaswamy (PivotX Advisors) pushed the conversation one level higher.
Their message reframed the entire discussion:
This isn’t just about deploying AI.
It’s about redesigning how organizations operate.
Including:
- how decisions are made
- how work moves across teams
- how systems adapt over time
The takeaway:
AI isn’t a layer—it’s a forcing function for operating model change.
What Leaders Are Leaving With
Across the morning, a few clear truths emerged:
- The technology is ahead of the organization
- Workflow design is now a core leadership capability
- Learning speed is a competitive advantage
- And most companies are earlier in this journey than they think
Or in simpler terms:
The hard part of AI isn’t the AI.
It’s everything around it.
Thank You to Our Speakers
- Doug Levin, Harvard & Stanford
- Urvashi Batra, Prioriwise
- Nicholai Mitchko, InterSystems
- Sanjay Yadav, Parlance
- Anand Swamy, PivotX Advisors
- Poornima Ramaswamy, PivotX Advisors
- Paul Baier, GAI Insights
- Brian McMahon, Capital One
Thank You to Our Sponsors
Jalasoft • MIT Professional Education • PivotX Advisors
Join us again on July 22nd at SharkNinja for Product Development That Adapts to Rapidly Changing Customer Needs
