Agentic AI, Vertical AI, and the Gap Between Hype and Reality

Over the last year, it has become fashionable to declare that agentic AI – autonomous, goal-directed systems capable of acting with minimal human supervision -represents the next great platform shift. In many corners of tech, the tone is breathless: agents will replace roles, automate workflows end-to-end, and compress entire organizations into a handful of people and models.

In my conversations, sports organizations feel the pressure too. From owners to executive leaders, headlines about AI are raising expectations – promising efficiency and revenue gains that once felt out of reach. Teams are now being asked to “use AI” as if it were a magic switch that instantly changes the game.

There is real signal behind this excitement. But there is also a widening gap between how AI is discussed and how it’s actually deployed inside real businesses – especially in the sports world.

Here’s my take at reconciling those two worlds.

The Optimistic Thesis: From Systems of Record to Systems of Action

Across startup ecosystems and investment circles – particularly among firms like Y Combinator, Bessemer Venture Partners, and Insight Partners – there is broad agreement on one thing:

Software is shifting from systems of record to systems of action.

In this framing, AI doesn’t just surface insights or generate content. It acts. It makes decisions, executes workflows, and closes loops. Vertical AI, in particular, is expected to win by embedding deeply into domain-specific workflows, compounding proprietary data advantages, and delivering outcomes rather than dashboards.

This thesis is directionally right – and important.

Where hype creeps in is the assumption that intelligence alone is enough.

What the Data Now Shows

Recent industry research – including the State of AI in Business 2025 report discussed widely in YC circles – adds an important dose of realism to the narrative.

The findings are consistent across industries:
• Executive optimism about AI is extremely high
• Actual deployment remains early, fragile, and uneven
• Most companies are still operating in copilot mode, not autonomous mode
• The clearest, most reliable ROI today comes from productivity and cost leverage, not transformation or net-new revenue

Perhaps most importantly, the data highlights a widening gap between bottom-up experimentation and top-down governance. Teams are adopting AI tools quickly, while leadership recognizes – often too late – that integration, security, brand risk, and accountability have not kept pace.

The constraint is no longer model capability. It’s organizational readiness.

Where Reality Pushes Back

1. Autonomy Is a Trust Problem, Not a Model Problem

In most organizations, autonomy is earned gradually. Revenue-affecting decisions, brand communications, pricing, and fan engagement are not places where leaders are willing to “let the agent decide” without guardrails.

The near-term value of AI comes from graduated autonomy:
• observe
• recommend
• simulate
• approve
• execute (within policy)

Full autonomy is the end of the journey, not the starting point.

2. Data Readiness Is the True Bottleneck

Agentic systems assume clean, unified, real-time data. Most sports organizations operate in the opposite reality: fragmented systems, inconsistent identity, permissioned data, and contractual constraints.

AI doesn’t fix bad data. It amplifies it.

Vertical AI only compounds advantage when it’s grounded in:
• trustworthy identity resolution
• clear data rights
• consistent schemas
• well-defined ownership

Without this foundation, agents are confident – and wrong.

3. Protocols Don’t Replace Governance

Tooling, APIs, and agent frameworks help systems talk to one another. They do not answer harder questions:
• Who owns the decision?
• Who is accountable when something breaks?
• When does a human override the system?
• How are actions audited and rolled back?

In multi-stakeholder environments – especially sports – governance is not overhead. It is the product.

4. Revenue Is Fragile Without Measurement

“Personalized experiences unlock new revenue” is true in theory. In practice, leaders want to know:
• Incremental versus cannibalized?
• Short-term lift versus long-term trust?
• What displaced what?

Research and operator experience increasingly show that early AI ROI appears first in productivity and cost leverage. Revenue impact takes longer – and requires deeper integration, cleaner data, rigorous experimentation, and far more trust.

AI that cannot prove incrementality eventually loses credibility.

5. Change Management Is the Silent Killer

AI fails less often because models underperform and more often because organizations don’t change how work gets done.

If incentives, workflows, and accountability remain the same, AI becomes:
• another dashboard
• another pilot
• another tool teams route around

Agentic AI is a work redesign problem before it’s a technology problem.

What Market Leaders Are Actually Doing

This tension between hype and reality is visible in how leading customer data and engagement platforms are talking about AI.

Companies like GrowthLoop, Hightouch, and Iterable are leaning into AI to reduce friction, accelerate workflows, and improve decision quality – but they stop well short of full autonomy. AI recommends, assists, and optimizes within defined guardrails. Humans remain accountable.

Even in commerce – where AI-driven personalization and optimization are most mature and directly tied to revenue – leading platforms emphasize constrained optimization, explainability, and human oversight. That restraint reflects not a lack of ambition, but a deep understanding of how trust is earned in revenue-critical environments.

This posture is not conservatism. It’s scar tissue.

What Vertical AI Gets Right – and Underestimates

To their credit, investors and operators increasingly acknowledge these realities. Vertical AI playbooks emphasize:
• deep workflow integration
• proprietary data
• outcome-driven systems
• engineering discipline over model novelty

What is often under-emphasized is that the very things that make vertical AI defensible – domain complexity, data rights, governance – also make it harder and slower to deploy.

That is not a weakness. It’s the source of durability.

Sports as a Stress Test for AI

If AI can work in sports – where data is permissioned, brands are sacred, stakeholders are fragmented, and mistakes are public – it can work anywhere.

Sports organizations don’t want magic. They want:
• confidence
• control
• proof
• then automation

This makes sports a proving ground for what responsible, enterprise-grade AI actually looks like.

A More Honest Framing

Right now, there’s a growing belief that agentic AI will replace headcount. My view is the opposite: without the right people in place, AI is ineffective. The real impact comes from changing how people work – once trust is earned and the right foundations are in place.

That’s what determines longevity. Not sporadic, one-off use cases like copy generation or ad hoc email writing, but systems that integrate into real workflows and compound over time.

In that sense, AI isn’t a race to deploy first. The real winners will be teams and organizations that deploy AI responsibly, measurably, and in alignment with how the sports business actually operates.

Final Thought

Agentic and vertical AI represent a real platform shift. But the constraint is no longer intelligence – it’s organizational readiness.

The next generation of enduring software companies will not be defined by who ships the flashiest agent, but by who builds systems of action that customers trust with their data, their brand, and their business.

That’s not slower thinking.
It’s the thinking required to make AI last.

Vince

More fans. Better engagement.

Talk with us