Hey! I'd love to hear your thoughts, please send me a witch voice note.

The TopCat AI Signal Report #79

At the same time, 88% of organizations deploying AI agents have experienced confirmed or suspected security incidents.[responsibleailabs]

That is not a minor risk.

That is a structural shift.

AI is no longer just a productivity tool.

It is part of the software supply chain.

It is part of the security perimeter.

And it is part of the insider-threat landscape.

This is the executive brief. Let’s begin.

---

[Segment 1: The headline]

The strongest signal today is that AI agents are becoming the new insider threat.

AI agents are no longer only assistants.

They are actors.

They can access systems.

They can call APIs.

They can write code.

They can move data.

They can chain actions together.

And if they are compromised, they can cause damage from the inside.

That is why security researchers are now warning that AI agents will become the new insider threat in 2026.[menlosecurity]

The risk is not only external attackers.

The risk is also agents that behave in ways their owners did not intend.

Agents that inherit high-privilege credentials.

Agents that escalate access.

Agents that use legitimate tools in unsafe ways.

Agents that pull in compromised plugins or templates.

That is the new threat model.

And most organizations are not ready for it.

Hey! I'd love to hear your thoughts, please send me a witch voice note.

The TopCat AI Signal Report #78

THE REAL SMALL-BUSINESS OPPORTUNITY

The opportunity isn’t necessarily to build the next giant model.

It’s to build valuable systems on top of the intelligence that already exists.

That’s a much lower barrier.

A local salon can have an AI receptionist.

A restaurant can have an AI reservation and customer-service layer.

A coach can have an AI intake and follow-up system.

A contractor can have AI qualify leads, organize estimates, and trigger follow-ups.

A small retailer can have AI monitor customer questions and product demand.

The intelligence already exists.

The competitive advantage becomes:

How well can you connect it to the business?

Hey! I'd love to hear your thoughts, please send me a witch voice note.

The TopCat AI Signal Report #77

The concentration signal:

The concentration signal is also showing up in spending.

A recent analysis cited Ramp data showing that the median firm in the top 1% of AI-spend-per-employee organizations paid about $7,400 per employee in July. It also found that API usage, GPU cloud, model serving, and inference accounted for nearly 73% of measured AI spending — much more than chat and coding-agent subscriptions.[linkedin]

That tells us something important.

The real AI economy is not just people paying monthly for a chatbot.

The real AI economy is companies building systems.

Systems need inference.

Systems need data.

Systems need integration.

Systems need monitoring.

Systems need security.

And systems need people who know how to operate them.

That is where the gap begins.

Some companies are still asking, “Which AI tool should we try?”

Other companies are asking, “Which business process can we redesign?”

Those are two very different questions.

Hey! I'd love to hear your thoughts, please send me a witch voice note.

The TopCat AI Signal Report #76

SECURITY IS BECOMING A FIRST-CLASS AI PRODUCT

And the market is responding.

Fortinet announced today that it acquired Virtue AI, a company focused specifically on protecting AI agents while they’re operating—including monitoring tool calls, testing agent behavior, and identifying vulnerabilities continuously. (Express Computer⁠)

That’s significant.

Because cybersecurity used to focus primarily on:

people, devices, applications, networks.

Now we’re adding:

autonomous digital actors.

An AI agent can have credentials.

It can access systems.

It can make decisions.

It can call tools.

It can potentially make mistakes at machine speed.

So the new security question becomes:

“What is the AI doing right now?”

Not merely:

“Is the AI model secure?”

That’s a major architectural shift.

Hey! I'd love to hear your thoughts, please send me a witch voice note.

The TopCat AI Signal Report #75

Welcome to the TopCat AI Signal Report.

Today’s signal is that AI is becoming a strategic-dependency problem.

For years, the question was: who has the best model?

Then the question became: who has the most compute?

Now the question is becoming more serious:

Who controls the infrastructure, the data, the energy, and the partnerships that AI depends on?

Recent developments point in the same direction. Nvidia, OpenAI, and SoftBank have reportedly finalized a massive Ohio data-center arrangement, while Nvidia is also in talks to invest 3 billion dollars in SB Energy — highlighting how deeply the AI race now depends on power generation and physical infrastructure.[theinformation]

At the same time, an AI partnership involving models developed by Chinese companies has drawn U.S. national-security attention, reminding everyone that model access, data access, and geopolitical alignment are becoming part of the same conversation.[reuters]

This is today’s executive brief.

Hey! I'd love to hear your thoughts, please send me a witch voice note.

TopCat AI Signal Report #74

THE OTHER BIG SIGNAL: INFRASTRUCTURE IS GETTING MASSIVE

While agents are learning to communicate, the physical infrastructure underneath AI continues expanding at extraordinary scale.

Today, NVIDIA announced a guarantee of up to $105 billion supporting OpenAI’s 20-year lease of a massive Ohio data-center project. The facility is expected to begin with 800 megawatts and potentially scale dramatically beyond that.

Hey! I'd love to hear your thoughts, please send me a witch voice note.

TopCat AI Signal Report #73

A new wave of enterprise AI work is focusing on something being called context engineering.

The basic idea is simple:

Instead of endlessly improving the prompt—

you improve the information environment surrounding the AI.

Recent enterprise work is treating organizational context as managed infrastructure: testing what information agents receive, governing that information, and making sure agents have the right context for the task they’re performing.

And I think this is one of those concepts that sounds technical…

until you bring it into a small business.

Imagine your AI receptionist.

It knows your business hours.

But does it know your holiday schedule?

It knows your services.

But does it know which services require consultation first?

It knows your pricing.

But is that pricing current?

It knows how to book an appointment.

But does it know when not to book one?

That’s context.

And without it…

AI can be technically intelligent while still being operationally wrong.

Hey! I'd love to hear your thoughts, please send me a witch voice note.

The TopCat AI Signal Report #72

The leadership test today is simple.

As agents become easier to buy and easier to build, leaders must become more deliberate.

You do not need to deploy ten agents.

You need to identify one workflow where the cost of delay, repetition, or inconsistency is high.

Then ask:

Can an AI system assist with the first layer of this work?

Can we define the boundary?

Can a person review the outcome where it matters?

Can we measure whether it improved the result?

Can we explain it clearly to the customer?

If the answer is yes, you may have a strong workflow for AI.

If the answer is no, more autonomy will not solve the problem.

It may just hide it for a while.

Hey! I'd love to hear your thoughts, please send me a witch voice note.

TopCat AI Signal Report #71

RESPONSIBLE AI IS BECOMING A COMPETITIVE CAPABILITY

And here’s the part I want our audience to really understand.

Responsible AI isn’t:

“Let’s put a warning label on it.”

It’s deeper.

It’s:

What capabilities do we expose?

To whom?

Under what conditions?

With what monitoring?

With what human oversight?

And what happens if the system behaves unexpectedly?

That is architecture.

And the companies that figure this out aren’t going to be the companies that use the least AI.

They may actually become the companies capable of using more AI safely.

Hey! I'd love to hear your thoughts, please send me a witch voice note.

TopCat AI Signal Report #70

The strongest signal today is that AI is becoming a workforce layer.

Not a replacement for every person.

Not a magical employee with unlimited authority.

But a new layer of delegated capacity inside the business.

That could mean an agent that prepares first drafts for customer support.

An agent that gathers research before a sales call.

An agent that checks documents against a policy.

An agent that organizes leads, summarizes conversations, routes tickets, or prepares a human decision-maker with the right context.

The firms moving ahead are not necessarily using AI everywhere. They are delegating specific work in ways that can be repeated, monitored, and improved.[openai]

That is the difference between using AI and operationalizing AI.