East Coast CIO Forum – September 2026 Summary
On September 2, 2026, the East Coast CIO Forum met for a candid discussion entitled:
“The Ugly Truth of AI [That No One Is Talking About]”
Thank you to Paolo Pelizzoli for leading a provocative and refreshingly blunt conversation. Thank you to Herb Mischel for his presentation and excellent recommendations. And most of all, thank you to everyone who participated in the discussion.
The theme of the evening was simple:
AI is real. The opportunity is real. But so are the cost, hype, bad data, organizational dysfunction and workslop.
Here are the major takeaways.
1. Enterprise AI Failure Usually Isn’t an AI Problem
The technology works.
Organizations often don’t.
Most AI pilots never reach production scale. Employees frequently report that poorly implemented AI actually increases workload. Initiatives stall because of internal resistance, unclear ownership, integration debt and change fatigue.
The recurring problems are familiar:
- Poor ownership
- Dirty and fragmented data
- Undefined ROI
- Integration and technical debt
- Change fatigue
- AI teams without enough organizational authority to change anything
The fantasy is:
“We have an AI strategy.”
The reality is often:
“We have a deck and a vendor.”
2. You Don’t Have AI. You Have a Subscription.
One of Herb’s bluntest points was also one of the best.
With most enterprise generative AI solutions:
- You own your prompts.
- You own your documents.
- You don’t own the model.
You are renting intelligence from somebody else.
And when you rent, you don’t control the meter.
Token consumption can be enormous. Models change. Vendors change products. Compute, GPU, RAM and infrastructure costs matter. What looks inexpensive in a proof-of-concept can look very different at enterprise scale.
The discussion drew an important distinction:
Rent for language. Build for decisions.
Drafting, summarizing and searching are natural uses for commercial LLMs.
For high-value decisioning, regulated data or tightly controlled processes, organizations should at least consider purpose-built models, local models or air-gapped environments where cost, data and model lifecycle can be controlled.
3. AI Amplifies What Already Exists
You can’t add AI to a system already on fire.
AI doesn’t magically repair a dysfunctional organization.
It can make the dysfunction move faster.
Bad workflow + AI = faster bad workflow.
Burned-out team + another AI initiative = more burnout.
Technical debt is becoming molasses underneath technology organizations. AI may eventually help eliminate some of that debt, but today that same debt can prevent companies from implementing AI effectively.
There is also a limit to organizational change bandwidth.
Teams are being asked to absorb new tools, new processes, new security requirements and new AI capabilities simultaneously.
Eventually, people simply stop absorbing.
4. Workslop Is Real
AI can create extraordinarily polished garbage.
Poorly understood models, weak instructions, inappropriate use cases and inadequate human review produce workslop: output that looks impressive, arrives quickly and still has to be checked or rebuilt by somebody who knows what they’re doing.
That remains an important role for humans.
AI can generate an answer.
A human still has to understand whether:
- The question was right.
- The answer makes sense.
- The data supports it.
- Anyone should act on it.
5. The Model May Not Be the Differentiator
An interesting theme in the discussion was that the advantage of Anthropic and other AI companies may not always be the underlying model.
Increasingly, differentiation may come from the harness around the model:
- User experience
- Context
- Workflow
- Tools
- Agents
- Guardrails
- Access to the right information at the right moment
The model matters.
But how effectively people can use the model may matter just as much.
6. AI Is Creating a Productivity Barbell
AI does not necessarily make everybody equally productive.
It may do the opposite.
The strongest people can become multiples more productive because they understand the subject, know what good looks like, know what to ask for and can quickly recognize bad output.
At the other end, weaker performers can use AI as a mechanism to think less, learn less and produce more mediocre work.
That creates an uncomfortable possibility:
Your best people get dramatically better. Your weakest people may fall further behind.
7. Organizations May Get Much Thinner
Nearly every corporate function can potentially be augmented by AI.
Legal.
Compliance.
Technology.
Operations.
Program management.
Management reporting.
That raises a harder question than simply:
“Will AI eliminate jobs?”
The better question may be:
Do organizations still need today’s organizational layers?
You may still need a CTO.
You may still need a Head of Legal.
You may still need a Head of Compliance.
But will those leaders need teams of the same size underneath them?
Potentially not.
AI could hollow out parts of the organizational pyramid while leaving smaller numbers of senior people responsible for directing systems, exercising judgment and accepting accountability.
And accountability is important.
A human still needs to own the decision.
8. AI Isn’t Deterministic Enough for Accountability to Disappear
Large language models are probabilistic systems.
Models change.
Providers update them.
Behaviors change.
Outputs change.
Something thoroughly tested today may behave differently after the underlying model changes.
That becomes particularly important when AI moves beyond creating content and begins participating in consequential business decisions.
Human-in-the-loop shouldn’t just be a comforting phrase.
For important decisions, somebody still has to validate the answer and ultimately say:
Yes or no.
9. Your Data Is Probably Less AI-Ready Than You Think
Organizations like to say:
“Our data is our biggest asset.”
That doesn’t make it true.
Having collected twenty years of data does not make it good data.
Putting it into a data warehouse does not make it AI-ready.
Having a data team does not create data ownership.
AI requires trustworthy data, clear ownership and governance.
And there is another increasingly important layer:
semantics.
The underlying data model matters.
Relationships matter.
Properly named fields matter.
Business definitions matter.
The AI needs enough semantic context to understand what the organization’s data actually means.
Otherwise, it can misinterpret the underlying information and surface a beautiful, articulate and completely convincing piece of workslop.
10. Technology Is Moving Faster Than Organizations Can Absorb It
Perhaps the broadest conclusion of the evening was that technology in general—and AI in particular—is advancing faster than most organizations can absorb, integrate and operationalize it.
The constraint is increasingly not access to technology.
It is the organization’s ability to:
- Redesign workflows
- Clean up data
- Train people
- Reduce technical debt
- Establish governance
- Measure outcomes
- Decide what should actually change
That doesn’t mean sitting this out.
Quite the opposite.
Lean into the technology. Use it. Experiment. Learn where it works and where it doesn’t.
But don’t confuse buying AI with transforming the business.
Closing Thoughts
AI isn’t going away, and waiting for everything to settle down isn’t a strategy.
But neither is attaching an AI tool to every existing process.
The organizations that win will be the ones that understand where AI actually adds value, clean up the foundations underneath it, empower their strongest people, control consequential decisions and remain willing to redesign how work gets done.
Or, in simpler terms:
Use AI aggressively. Trust it selectively. Own the outcome.
And one final piece of advice from the evening:
Be nice to the AI.
Someday it may remember who was polite.
-malka
Malka Treuhaft
Executive Director East Coast CIO Forum &
President
Truision Inc.
