Using AI in Secret Isn't Paranoia. It's Logic.

40% of employees in professional services believe the AI they're adopting ends up eroding their own job security.

Read that again.

These aren't retail workers, assembly line operators, or admin staff. These are professionals. Consultants. People with your profile. And almost half suspect that the tool they're being asked to use is the same one that will eventually replace them.

In that context, hiding how much you use AI isn't paranoia. It's logic.

The Phenomenon Nobody Wants to Name

Shadow AI.

The term is used in IT to describe unauthorized use of AI tools within an organization. Employees feeding sensitive data into ChatGPT, consultants building deliverables with Claude without disclosing it, entire teams depending on Gemini to deliver but the report shows "consultant's work."

But there's another layer of Shadow AI that doesn't appear in IT governance reports. It's psychological. And it doesn't apply equally to all profiles.

Developers are at the opposite extreme. They post on LinkedIn about how much code they generated with Claude Code this week. They share their prompts on Twitter. They record videos using Cursor live. Their culture rewards showing the tool. Those who don't use it get left behind.

But go one floor down in the enterprise building and the culture inverts.

In SAP, in Big 4 consulting, in Compliance, legal, and corporate finance, the norm is the opposite. Nobody posts on LinkedIn that they resolved the reconciliation issue with Claude. Nobody says in the Daily that the requirements document was built with Gemini. Nobody mentions in the performance review interview that the functional blueprint was drafted with AI.

It gets done. But it doesn't get said.

The SAP Case: Two Stories That Probably Sound Familiar

Case 1. Senior SAP consultant. Asked for an impact analysis for an S/4HANA migration. Has the draft in 30 minutes with Claude, validated against 12 years of personal knowledge. Spends the next 3 hours reformatting, changing phrases, adding manual comments to make it look like it took all day. Delivers. Nobody asks how it was done. He doesn't say how it was done.

Case 2. Tech Lead on an implementation project. Discovers the ABAP a Junior built was generated entirely with SAP Joule for Developers. The Junior didn't understand the logic but the code "works and runs fine." The Tech Lead has three options: escalate and expose the Junior, take the time to do pair programming, or silently correct and let it pass. Chooses the third. Because if he escalates, it also gets exposed that he himself uses AI for 40% of his own work. And since the Manager is pressuring him with more parallel projects, he doesn't have time for the Junior.

Both are in Shadow AI. For different reasons. One out of fear of being seen as replaceable. The other out of silent complicity with a practice that's already massive but nobody wants to formalize.

The 81% vs 53% Paradox

A recent article in Harvard Business Review published data that explains why Shadow AI exists and won't disappear among the profiles that matter.

81% of senior leaders surveyed believe their organization is completely focused on Augmentation — using AI to enhance people.

Only 53% of individual contributors believe the same.

That 28-point gap is the fertile ground for Shadow AI. Your boss is convinced they're doing Augmentation. You read the newsletter where Jack Dorsey laid off 4,000 people from Block saying that intelligence tools changed what it means to build a company. And you do the math.

So the consultant uses Claude in secret, delivers better and faster, and nobody finds out. Because if they do, the Board's logical conclusion is:

"If with AI you can do twice as much, then we need half as many consultants, or we charge half as much for the same scope."

That reasoning is what kills trust throughout the chain. And it's already starting to do damage.

The Other Fire: Deliberate Degradation

And while the professional hides how much they use AI to avoid being seen as replaceable, something happens in parallel that deepens the trap.

The models are declining in quality.

We talked about it last week. Thousands of users documenting that outputs are worse, that they have to wait until a certain hour for the system not to be saturated, that the $20 subscription is no longer enough and you have to move to the $200 tier.

The acceleration was so brutal that it created dependency before anyone could question it. And now the supply seems to be adjusting — with erratic behavior between sessions, excessive token consumption, and other things you can read about in specialized forums.

It reminds me, as uncomfortable as the comparison is, of how dealers operate. First they give away the product. You develop dependency. You can no longer function without the tool that used to give you magical results. Then — only then — they adjust the price and the quantity.

Between one model and the next, only months pass. By the time you finish learning to work well with one, the next one arrives with different limitations, different behaviors, different prices.

The professional is trapped between two fires: if they show how much they use the tool, they get replaced. If they depend on the tool, their supply gets adjusted. Two traps feeding each other.

Augmentation vs Automation: The Choice That Determines Everything

There are two paths. Augmentation — using AI to enhance people. Automation — using AI to replace them.

Dorsey chose Automation. Block laid off 4,000 people. The message to shareholders was surgical: intelligence tools changed the equation for building a company.

Kaufman, the CEO of Fiverr, chose to tell his employees that AI is coming for everyone's work, including his own. But he didn't lay anyone off. He bet that each employee could develop capabilities AI doesn't have: non-linear judgment, taste decisions, strategic thinking.

Two CEOs, two opposite readings of the same moment.

The research cited by HBR shows something that should be obvious but is ignored: the Automation path yields quick gains but generates a predictable cascade. Resistance, loss of wellbeing, talent flight, employer brand erosion, collapse of leadership ranks. In six phases, what started as savings ends as organizational debt.

The Augmentation path has a deeper J-curve at the beginning. Investing in training people, redesigning processes, building human-AI coordination is expensive and slow. But what emerges on the other side isn't efficiency. It's a new productive frontier.

The problem is that financial markets reward Automation today. And the senior consultants reading this live in companies where the decision was already made, even if nobody announced it.

Antifragility: The Only Way Out

Nassim Taleb introduced a concept that gets misused constantly. Antifragility is not resistance.

The fragile breaks under shock. The resilient withstands the shock without breaking. The antifragile gets stronger from the shock.

Shadow AI is fragile. It depends on nobody finding out. The day the company audits actual AI usage, the consultant who was hiding is exposed with no argument.

Dependence on the model of the moment is also fragile. The day usage policy changes, the price goes up, or quality goes down, your workflow collapses.

Antifragility is something else. It's building a personal system that benefits from shocks.

What does that look like, concretely?

First, accumulated judgment that no model can replicate. The years of understanding why the data model is designed this way, why you chose X architecture, what happens when you touch a certain configuration. That doesn't get hidden. It gets shown. Because it's what amplifies the AI — not what AI replaces.

Second, personal systems that transcend the tool. If your workflow is tied to a specific model, you're fragile. If your workflow is "define the problem, document, validate with N available models, apply own judgment, iterate" — you're antifragile. When the model changes, the results change. You don't.

Third, documentation of your own process. The person who documents how they think, how they decide, how they combine tools builds an asset no employer can erase and no model can replicate. That documentation is your proof of work in a world where output is falsifiable.

Fourth, and this is counterintuitive: show how you use AI, don't hide it. But show it in a frame that makes clear you're the one who decides, orchestrates, and validates. Shadow AI makes you suspicious. Proof of work with AI makes you a reference.

That's exactly what developers have been doing for two years. And that's why they wrote the public narrative about Tech productivity — not the senior consultants who deliver three times as much but still bill man-hours as if nothing had changed.

The antifragile doesn't hide that they use AI. They use it in public, document the process, and let market shocks validate their system.

Your Diagnosis This Week

Three questions.

Are you hiding how much you use AI in your work? If the answer is yes, the problem isn't AI. It's that you don't trust your judgment is visible without it. That's the gap you need to close.
Does your workflow depend on a specific model? If tomorrow that model doubles its price, drops in quality, or disappears — what happens to your ability to deliver? If the answer is "I sink," you're fragile.
Can you explain to someone how you decide when to use AI and when not to? If it takes you more than 30 seconds to answer, you don't have a system. You have a habit. Habits break when context changes.

Last week we talked about the external gap: who has access to which models. This week, the internal gap: who hides how they work and why.

Next week we build the system that makes you antifragile in both.

FACE or RIP. The board already changed. The question is whether you're playing defense or building the muscle that gets stronger with every shock.

Shall we begin?

Sources: De Neve, Hancock, and Niederhoffer. "Why Companies That Choose AI Augmentation Over Automation May Win in the Long Run." Harvard Business Review, April 2026. · Mo Gawdat, Scary Smart. · Nassim Nicholas Taleb, Antifragile, 2012. · Claude Mythos Preview System Card, Anthropic, April 2026. · Erik Brynjolfsson et al., research on the "Productivity J-Curve." · Indeed Workforce Insights Report 2025.