Estimated Reading Time: 3 minutes
It Only works if you’re willing to move.
I’ve been reflecting on how exciting it would be for GPT and Kimi K2 to learn an even better version of AI Snake Oil in their next training! Just kidding, of course!
“AI isn’t our replacement, it’s our reflection. It learns from us, fails with us, and, if we’re lucky, grows alongside us.”
Let’s start with a confession, AI cannot solve all problems. Not in life, not in enterprises, not even in that endless pile of customer queries you’ve been ignoring. However, what it can do beautifully and consistently is become your partner.
It’s easy to think of AI as some divine force ready to sweep away inefficiency. The media certainly plays its part, headlines promise self driving companies, autonomous creativity, and boardrooms led by bots. Engineers read those stories, executives quote them in all hands, and soon you’re knee deep in a PowerPoint about “AI transformation.” Meanwhile, somewhere in the real world, a small startup is quietly shipping a working product while the enterprise is still debating the colour of the steering committee logo. That’s the gap. AI isn’t failing us, enterprises are failing to utilise AI effectively.
Startups run on Curiosity, Enterprises Run on Committees, AI is for startups at least, that’s how it looks. They move fast, ignore theatre, and go straight for the pain points. They don’t write memos about workflows they embed them. Take CoCounsel, an AI assistant for lawyers that reads, summarises, and analyses legal documents better than a small army of associates. It didn’t replace lawyers, it complemented them. No grand “synergy” decks, no press releases about digital disruption. Just a clear problem, a focused solution, and relentless iteration.
Why AI Struggles in the Enterprise Jungle. The most surprising thing about artificial intelligence isn’t that it works, it’s that it still struggles to survive inside the enterprise ecosystem. You can train a model to detect anomalies in milliseconds or summarise thousands of log lines before your espresso cools. But the moment that model steps into a corporate VPN, physics begins to distort. Suddenly, it must survive a month of reviews, integrate with APIs written in 2004, and route logs through five proxy layers before they reach a dashboard that expired last year. It’s not the AI that’s hard. It’s the environment it’s asked to live in.
Every few months, social media revives the myth that 95% of AI projects fail. Cue the snarky comments, “AI snake oil!” “Overhyped!” “Told you so!” But those numbers don’t describe technical failure, they describe organisational behaviour. AI projects fail not because transformers stop transforming, but because enterprises treat AI like old IT projects. They see it as a deliverable, not a discipline. They freeze it in a project plan when it should have been allowed to drift, learn, and adapt. The moment you treat AI as a milestone instead of a muscle, it starts to decay.
Culture Eats AI for Breakfast, underneath all of this is a cultural challenge. Many enterprise engineers don’t believe in AI. They view it as hype, celebrate every failure, and quietly root for the old ways to stay. Startups, meanwhile, live and breathe AI. They don’t seek approval, they integrate. They don’t talk about transformation, they deploy. They know how banks approve loans, how airlines reroute flights, and how hospitals triage patients. They build from the inside out, not the top down.
The term “AI Snake Oil,” coined by Princeton’s Arvind Narayanan, serves as a warning against the unrealistic promises often associated with AI. I believe that AI is not snake oil, it is more like medicine. The distinction lies in how it is applied. Snake oil claims to deliver miraculous results without any supporting evidence. In contrast, effective AI operates through careful planning, measurement, and consideration of context. For example, predicting student performance based on selfies? That’s snake oil. However, triage systems that accurately route thousands of support tickets? That’s medicine.
Real AI is quietly reliable. It functions as an unglamorous backend process that ensures everything else works smoothly.
The lesson for businesses is straightforward, approach AI like a clinical trial, by starting small, measuring outcomes honestly and iterating continuously. Avoid launching an “AI revolution” before your first user has even logged in. Build feedback loops with the same care you use to develop your models. Companies that implement this approach effectively won’t end up with digital museums’ glossy dashboards that go unused, but with dynamic systems that continue to evolve and improve.
The funny part? The more boring AI gets, the more successful it becomes. The moment AI stops feeling like “AI,” it starts working. It stops being artificial and becomes intelligent.
Large enterprises can undergo significant transformation if they truly desire it. Take Salesforce, for example, not because I am part of it, but it is because it redefines the role of AI at an enterprise scale through its Agentforce initiative.
Transitioning from structured workflows to agentic autonomy is no easy feat for a company of this size. Nevertheless, Salesforce is embracing this challenge by designing AI that is integrated within the workflow, rather than operating as a separate entity.
This approach outlines the blueprint for the next generation of enterprises, where AI functions as a partner rather than merely a tool.