One evening this month, after wrapping a round of keynotes in Singapore, I found myself nursing a jet-lagged coffee in a hotel lobby. A chief investment officer from a world-renowned family office approached me:
“Barry, I was pitched a company using AI for predictive maintenance in commercial real estate. The demo was amazing, the data insights outstanding, yet every time I ask to start pilots across our portfolio, they say, ‘we’re still training the model on proprietary data.’ Does that sound strange? This investment is starting to haunt me.”
I smiled. I’ve heard that script several times in the past few weeks.
In London, a VC told me she backed a “generative coding AI,” only to discover it was essentially a glorified Loveable prototype and costlier than maybe half of her enterprise SaaS startups to scale because of excessive platform token calls.
In New York, a startup CTO showed me a deck for “autonomous retail agents” that turned out to be rule-based bots. And the highlight (I won’t say where) was a founder pitching a “neurosymbolic AI platform.” Catchy words, no working product or customers, seeking a $2M seed at a $20M valuation.
Across continents, the question people keep asking me is the same:
How do you tell whether a so-called AI project is alive or already dead?
Over the last few months, traveling across North America, Europe, and Asia, more executives, VCs, and founders are asking exactly that. The hype around AI is real. It’s overhyped in some corners, underhyped in others. What’s missing is a practical pattern language to detect Zombie, Ghost, or Ghoul AI projects, ventures that promise the stars but deliver shadows.
So, in the spirit of Halloween and innovation, here’s how to see the ghosts in the machine. And, more importantly, how to spot the living AI ventures like Evalify.ai that are trying to do the work, deliver real value, and build sustainable models.
The Ghosts of Tech Past
Back in March 2018, I attended one of the first major blockchain conferences in San Francisco, Token Fest. Every booth I visited was rebuilding existing products “on blockchain.” There was “the Airbnb of blockchain,” “the Uber of blockchain,” and even “the Facebook of blockchain.”
When I asked, “Why put Airbnb on a blockchain?” the responses were invariably similar and candidly, weak: “So people can’t change the record of bookings,” or “so you can own your own data.”
While they were interesting ideas, none felt like a burning customer pain point. None of them really leveraged the unique characteristics of the technology to innovate business models or customer behavior. Two years later, almost all those companies had folded. Technology was a decoration, not a differentiator. Zombies from day one.
Today, we’re seeing a similar déjà vu with AI. Why? Because the majority of investment is flowing into “AI.”
In Q1 2025, for example, AI commanded roughly 71% of total VC deal value according to PitchBook. Founders feel pressure to AI-wash their decks. The result is vaporware, issues masked by hype, and systems that are more promise than product.
At Nobody Studios, we try to take the opposite approach. We find boring, critical problems, ones people actually pay to solve, then layer in AI as the engine that learns and improves as usage accumulates. That helps us build living ventures, not haunted ones. And yes, we built and launched a few AI monsters in our early days. We have unlearned, and hope to help you avoid the same mistakes.
So to help you distinguish the living from the walking dead, here’s a field guide: 21 Signs of Zombie, Ghost, or Ghoul AI Projects, plus the counterpoints that show when AI is truly alive. Then I’ll we’ll walk through Evalify.ai as a living, iterating example.
21 Signs of a Zombie, Ghost, or Ghoul AI Project
Use this toolkit for due diligence, whether you’re investing in, joining, or buying from an AI startup.
Zombie Projects: They look alive, but the pulse is gone.
1. Traction is all talk. White papers, “research partnerships,” or pilot claims, but no paying, repeat users.
2. Endless pilot mode. “Testing” for 12–24 months without scaling or commercialization.
3. No learning loop. The model doesn’t improve with usage, it just automates static rules.
4. Overengineered core, under-engineered delivery. Fancy architectures solving trivial, infrequent problems.
5. Investor deck over working product. Slides sparkle, the product is rough, buggy, or missing.
6. Vanity metrics. “100,000 API calls a day,” but no retention, Customer Acquisitions Costs, or gross margins.
7. Technology costs don’t scale. Compute, storage, or data-labeling costs balloon with no path to profitability. (👋 HELLO LOVEABLE, and yes, all the “I built a billion-dollar company in a day” blog posters!)
Ghost Projects: You hear the name everywhere, but never see substance.
8. No live demo or sandbox. Everything is “coming soon” or buried under NDAs.
9. Hidden data sources. Vague “proprietary datasets,” no provenance.
10. Invisible customers. “Enterprise partners” that are anonymous or unverifiable.
11. Overstated capabilities. “AGI-level” or “full autonomy” for narrow domains.
12. Soft social proof. Testimonials from advisors or influencers, not paying users with recognizable logos.
13. Sci-fi roadmaps. “Quantum AI,” “neurosymbolic AGI,” “conscious agents,” all on the next slide.
14. Buzzword salad. Multimodal, federated, decentralized, causal inference, all at once. If your mum or mine can’t understand it, that’s your sign. Boohoo, bye-bye.
Ghoul Projects: They feed on fear and your funding.
15. AI as costume. Under the hood, it’s a standard SaaS or marketplace. “AI” is just a fundraising outfit.
16. FOMO baiting. Constant name-dropping of OpenAI or Anthropic to borrow credibility. Those firms have vast resources, sales, and marketing budgets. The startup you’re evaluating likely does not.
17. Secret sauce is a prompt. The “proprietary algorithm” is a prompt template on a general LLM.
18. Stage time over ship time. Founders win panels and podcasts but don’t deliver to users.
19. No ethics or compliance. Shrugs at privacy, bias, copyright, or hallucinations.
20. No ROI clarity. “Transformation” with no link to revenue, savings, or risk reduction.
21. Fatigue without growth. Team burnout, feature churn, stagnant core metrics.
Counterpoints: What Healthy Human AI Companies Do
- Start with a real, recurring pain people already pay to solve.
- Roll out minimal viable solutions, then gradually layer in intelligence.
- Show a learning curve with performance improving as data and usage grow.
- Demo from day one with working sandboxes and proof points.
- Actively manage cost, drift, bias, privacy, and ethics.
- Make bounded, measurable claims, not grandiose pronouncements.
- Articulate a business model and ROI map with a path to scale.
- Obsess over customer feedback and iterate early and often.
Evalify.ai: A Living Case in Contrast
I’m not saying everything I’ve done, or our Studio has done, or any of the portfolio companies has been perfect. Far from it. Most of what we’ve learned came through hard-won lessons and the occasional monster we created and had to kill ourselves.
I’m not a data scientist, an LLM modeler, or an entrepreneurial oracle. I’m a builder who has seen enough patterns to tell life from the undead.
So let me draw you into a living example from Nobody Studios: Evalify.ai. It’s the kind of company you want people to call you about, and it was recently highlighted by Inc. Magazine as one to watch because it tackles a “boring” but essential legal problem with AI to make it faster, cheaper, and easier to understand.
The Problem They Solve
Investors, incubators, corporate innovation teams, and founders routinely face intellectual property (IP) risk when vetting new ideas. Startups fail or get litigated because they didn’t clear prior art or patent-infringement risk before going to market.
Evalify automates part of that due diligence. It scans global patent data to generate a Freedom-to-Operate (FTO) score, surfaces risks, maps claim overlap, and accelerates decisions. As Inc. put it, it’s “the AI-powered patent check that is reducing risk for startups and their lawyers.”.
Instead of waiting months and spending tens of thousands on manual reviews, Evalify provides a preliminary view in minutes for a fraction of the cost, helping investors and startups avoid nasty IP traps.
It’s a PBSD problem:
- Painful (litigation and blockade are existential)
- Boring (not sexy, but crucial)
- Scalable (global patents, repeating patterns), and
- Deep (domain expertise, processing, and scale required).
What They Actually Do
Public information from Evalify shows hallmarks of a living AI business:
- Analysis across over 200 million patent records in over 170 jurisdictions for preliminary FTO assessments.
- A proprietary risk score (roughly 250–900) indicating the depth of due diligence required.
- Almost 250 pilots reported, with high user satisfaction and strong intent to convert to paid.
- A SaaS subscription model to stabilize revenue.
- Transparent methodology in blogs and feature updates, such as “Infringement by Usage.”
- Backing from Nobody Studios for talent access, governance, and network.
Evalify isn’t hiding behind hype. It focuses on technical depth, transparency, iteration, and customer feedback. As users engage, Evalify enhances the product and pivots thoughtfully based on real-world use.
Why Evalify Isn’t a Zombie, Ghost, or Ghoul
Run Evalify through the 21 haunted hints and you’ll see;
- Real pilots and users, not just white papers.
- Clear methodology showing how claims are interpreted and overlaps detected.
- Data scope and jurisdictions disclosed.
- Iterative feature growth, such as Infringement by Usage.
- ROI focused on reducing legal risk and costly missteps.
- No magical AGI claims.
- Broad accessibility for VCs, accelerators, and founders.
- Signals of conversion from pilots to paid.
Evalify tries to set a standard for what it means to be alive in AI ventures. It’s a company that resists haunting, and is hunting real business value and customer delight.
Lessons, Recommendations, and Playbooks for Zombie Hunting
From the contrast between haunted hype and living ventures, here’s what I’d urge any executive, founder, or investor to do:
1. Start with a real, painful problem, not an “AI canvas.” AI is a tool, not a destination.
2. Ship incremental intelligence. Begin with rules or supervised approaches and add learning layers over time.
3. Demand learning curves. If performance doesn’t improve with usage, feedback, or new data, it’s likely hollow.
4. Insist on demoable proof. No sandbox, no MVP, endless NDAs, red flag.
5. Model cost and scalability. Push on compute, labeling, and data acquisition. If they can’t justify scale, be wary.
6. Ask for bounded, measurable claims. “Tripled throughput in X” beats “revolutionizing industry.”
7. Vet the data story. If they can’t explain provenance and permissions, that’s a ghost trap.
8. Push for transparency. Good teams show their work, their methods, and their trade-offs.
9. Map ROI for customers. If they can’t articulate savings, revenue, or risk reduction, it’s hype.
10. Favor builders and studios. At Nobody Studios, accountability, governance, and product discipline help ventures escape the haunted realm.
Closing the Graveyard
When you scan your pipeline, whether you’re joining an exciting AI team, investing in a startup, or evaluating a product, use this lens. Ask:
- Is there a pulse? (traction, usage, feedback)
- Is it learning and adapting?
- Are the claims grounded in reality?
- Is there transparency in data, method, and metrics?
- Can it map to ROI or risk reduction?
Here’s the trick. True AI value compounds. The power isn’t in version one, it’s in how the system improves as more data, feedback, and edge cases fold in.
The winners are learning systems, not one-shot demos.
Evalify is an example of that. A venture built to survive, iterate, reveal, and deliver. It’s not magic. It’s engineering, transparency, iteration, and discipline, and there’s still a long way to go to unlock the ultimate value for customers.
This Halloween, when someone hands you an AI pitch deck full of promise, pull out your list. Don’t be spooked. Hunt for the undead. And place your bets on ventures that breathe.
Build what matters. Build something that outlives you.