I’ll admit it. I binge-watched Inventing Anna on Netflix over a couple of nights, the way most of us did back in 2022. And like most viewers, I walked away simultaneously appalled and a little bit fascinated by Anna Delvey. Here was a woman with no trust fund, no inheritance, and no real business, who nonetheless convinced New York’s elite, its banks, and its hottest hotels that she was a German heiress about to open a private arts foundation. She rented private jets. She dropped hundred-dollar tips like they were nothing. She hired lawyers and architects for a foundation that never existed on paper as anything more than a mood board.

There’s a scene early in the series — Anna, cornered by her friend Neff about how she’s actually going to pay for all this — that has stuck with me:

It’s a great line, delivered with total conviction, by someone who was doing everything except the hard work required to make her fictional empire real. She wasn’t actually doing work building a foundation. She was building the appearance of one, in the hope that appearance alone would eventually conjure the funding to make it real. Spend the money first, look the part, and the substance will follow — that was the bet.

This is an exploration for me too, but the more I watch large enterprises approach AI adoption in 2026, the more I see Anna’s playbook being run at scale, with board decks instead of Birkin bags.

The Thesis, Up Front

Large enterprises today are running the Anna Delvey playbook on AI — spending big, moving fast, and performing transformation for the board and the Street, without doing the unglamorous, structural work that would make the returns real. And just like Anna, when the bill comes due and the money isn’t there to back it up, the fallout won’t be limited to a few embarrassed executives — it will show up as real technical debt, real financial debt, and real credibility debt that takes years to unwind.

I want to be careful here — I’m not saying enterprises are committing fraud. Of course, they are not. But the underlying dynamic — spend conspicuously to project transformation, without a credible mechanism to generate the return that justifies the spend — appears uncomfortably similar. Let’s walk through why, using the lens of First Principles, the same way I approach every problem that requires Systems Thinking, which this one certainly does..

Anna’s Strategy, Translated to the Boardroom

Anna’s entire operation rested on a few pillars: look the part, associate with the right people, spend visibly, and let the perception of inevitability do the fundraising for you. She wasn’t lying about wanting to build something real — the Anna Delvey Foundation was a genuine idea. Her failure was in sequencing: she tried to manifest the outcome before building the underlying capability to deliver it.

Compare that to how a lot of large, traditional enterprises are approaching AI right now:

• Announce aggressive AI initiatives in earnings calls and investor decks to signal “we get it” to Wall Street and the board.
• Reallocate budget and headcount away from revenue-generating, working projects to fund flashy AI pilots.
• Push AI into products and workflows without a clear definition of the business outcome it’s supposed to produce.
• Skip or shortcut the risk, compliance, and feasibility assessments that would normally gate a major technology investment and transformation of this size and targeted impact.

None of this is because leadership is dumb or malicious. It’s because the pressure to appear transformed is enormous, and — just like Anna discovered — appearing transformed is a lot faster and cheaper in the short term than actually being transformed. The problem, of course, is that the bill always comes due.

The Debt Accumulates Quietly, Then All at Once

In Inventing Anna, there’s a moment where Anna tells her friend, with total sincerity, “When you’re out of new ideas, make your old ideas bigger.” That’s a pretty good description of what I see happening inside a lot of AI programs today: instead of validating whether the original idea actually works, teams double down and scale it, on the theory that bigger will somehow fix the fact that it never had a foundation.

This is where I want to draw a distinction I’ve made in previous posts about DevOps and cloud adoption, because it applies directly here: there is explicit technical debt and there is implicit technical debt, and enterprises rushing AI are stacking up both.

Explicit technical debt is the debt you can see and point to — AI features bolted onto products without proper rearchitecture, agents wired into workflows with no validated eval framework, pipelines built for a demo that were never meant to survive contact with production being pushed to Prod. This is the “we’ll clean it up later” debt that every engineering leader has lived with in some form for decades. It’s ugly, but at least it’s visible on the balance sheet, so to speak.

Implicit technical debt is more dangerous, because it’s invisible until it isn’t. This is the debt you accumulate by skipping risk assessments — not evaluating whether a model’s outputs are fit for a regulated use case, not checking a use case against emerging AI regulation (the EU AI Act, sector-specific guidance, or frameworks like NIST AI 600-1), not stress-testing whether a vendor’s promised ROI has ever actually been demonstrated anywhere at enterprise scale. This debt doesn’t show up in a sprint retro. It shows up in a regulatory audit, a customer impacting incident, or a board question nobody can answer, usually 18 to 24 months after the original decision was made — right around the time someone expects to see returns.

Anna’s downfall happened at exactly this inflection point. Her plan required her business venture — the foundation — to eventually generate the capital that would retroactively justify all the spending and cover the debts. When the financing fell through and there was no revenue, no assets, and no real foundation to point to, the entire structure collapsed in on itself, fast. The hotels wanted their money. The banks wanted their money. And there was nothing behind the curtain.

Enterprises are setting up the exact same collapse condition. They are spending against a promised future return on AI that, in most cases, has not been demonstrated — not by them, and frankly not by most of the vendors selling it either.

Why the Bet Is Worse Than It Looks

Here’s where I think the analogy actually understates the risk enterprises are taking on, because Anna at least had one asset going for her: charisma and total conviction. Most enterprise AI bets don’t even have that — they have a MBB partner produced deck. And there are four structural realities that make “spend now, prove it later” an especially bad bet in AI, specifically:

• Most vendor-promised returns are still unproven. The productivity and revenue numbers in AI vendor pitches are, in most cases, extrapolated from small pilots or cherry-picked case studies, not demonstrated at the scale and complexity, or in the regulatory context of a large traditional enterprise.
• The competing models are evolving too fast to make durable big bets. Committing multi-year budgets and org redesigns around a specific frontier model’s capabilities is a bet on a snapshot of technology that could be obsolete, or simply out-priced, within a single model release cycle.
• Open-weight models are closing the gap on frontier models at a startling rate, which directly undercuts the argument for locking into expensive proprietary model contracts before you’ve proven the use case even needs that level of capability.
• Most tokens are being wasted, not invested. Untrained or undertrained users, with no tooling to manage or govern token spend, are burning through compute budgets on tasks that don’t move the needle — while nobody is tracking token economics the way we spent the past decade teaching organizations to track cloud unit economics.

Put together, this isn’t just “risky” — it’s structurally similar to building a foundation on a financing deal you haven’t closed yet. Anna assumed the money would show up because it had to. A lot of enterprise AI roadmaps are making the same assumption about ROI.

What Doing This Right Actually Looks Like

I don’t think the answer is to stop investing in AI. Far from it. I think the answer is exactly what I have argued for years about DevOps and Cloud adoption: play the long game, and build and validate the capability before you scale the spend. Boards don’t need a magic trick. They need to be educated, not dazzled.

Here’s what that looks like in practice:

• Educate the board on knowns and unknowns, not just upside. Give them the honest picture — which vendor claims are proven, which are speculative, and how you need to account for the fact that the AI evolution is occurring faster than your planning and budget cycles.
• Invest in change management before scale. AI adoption is an organizational transformation, not a tooling rollout — the same lesson DevOps taught us a decade ago, and one we keep forgetting every time a new technology wave shows up.
• Run real risk assessments, for feasibility and for compliance. Every AI use case should be evaluated against frameworks like NIST AI 600-1 and applicable regulation before it gets funded, not after it’s already in production and a board member or regulator asks an uncomfortable question.
• Build token economics discipline. Treat token spend the way we taught the industry to treat cloud spend — with visibility, governance, and accountability for waste, rather than an open tab.
• Train your people before you scale usage. Most wasted AI spend isn’t a model problem — it’s a skills problem. Untrained users generate untrained outputs, at cost.
• Fund AI from growth, not by starving what already works. Cutting headcount and revenue-generating projects to fund unproven AI bets is exactly the kind of move that leaves you with nothing to fall back on if the bet doesn’t pay off on schedule.

None of this is as exciting as announcing a bold AI transformation on an earnings call. But it is the difference between building a real foundation and building the appearance of one.

Final Thoughts

Anna Delvey almost pulled it off — and that’s precisely what makes her story such a good cautionary tale rather than a simple morality play. She was smart, she was convincing, and for a while, the bet looked like it was working. Right up until it wasn’t, and there was no capital, no business, and no plan behind the persona to absorb the fall. She was last seen on Dancing with the Stars, sporting an ankle monitor.

Enterprises rushing to look transformed by AI, without the underlying architecture, governance, and economics to back it up, are running the same play. The good news is that unlike Anna, they still have time, and the resources, to close the gap between appearance and substance — but only if they choose the long game over the impressive-sounding headline.

In my next post, I want to dig deeper into what token economics discipline actually looks like in practice — the tooling, the metrics, and the governance model enterprises need to stop burning compute budget on nothing. Watch this space.

As always, I’d love to hear how this lands with you. Do reach out on LinkedIn/X or drop a comment.