AI for the Sake of AI: Welcome to POC Hell

Alright, let’s talk about something I call "POC Hell"

Here’s how it goes:

Management across the tech world has finally started understanding that AI is the future. Keep that sentence in mind; it’s important.

Now there are two sides to this story:

Side A, The company that needs AI services (let’s call it Company A):

Top management here has realised “AI is the future,” so now they decide, "We need to be AI Ready" or "There should be AI adoption". Cool. The budget is allocated. Big buzzwords flying.

Now the departments are like, okay, we have this budget, but what the hell do we do with it? So they start bringing in SMEs and domain experts and cook up some use case.

One thing to note here: if you pull these SMEs aside and ask them off the record if AI is actually needed here, they’ll whisper “No.”

But anyway, they package this use case, throw in some dashboards, buzzwords, and take it to management. Management sees the deck and gets an orgasm. They love it. It’s disruptive. It’s transformative. It’s AI. So now, they start looking for vendors to build this thing.

Side B, The company that offers AI services (let’s call it Company B):

Top management here has also realised that “AI is the future.” So what do they do? They start hiring AI engineers, building GenAI teams, and setting up CoEs.

Now these teams are sitting around waiting for work. So Company B goes hunting for AI projects. They find Company A and hear their use case.

Their inner voice goes: “What a bullshit use case. Do we really need AI here?” Inner voice answers: “No.”

But hey, there’s a budget, and Company B has a team to keep busy. So, they say yes.

Now, Company A says, “Before we proceed, we want a POC to see if you guys are capable.” Fair. And POCs still bring in revenue.

So Company B agrees. At least the team isn’t on the bench. Maybe it will become a full-fledged project.

Now here’s where it gets messy.

Nobody, literally nobody, really knows how to handle these projects. And it’s not their fault. This is new territory. I’m not talking about traditional ML projects. I mean LLMs, GenAI, orchestration, agents, all the bleeding edge stuff.

Anyway, the POC goes on for a couple of months. Company B delivers it. The team pulls it off.

Meanwhile, Company A has been sitting with this use case, hearing buzzwords, and slowly realising: “Wait... do we really need this?”

So they take the POC and start using it. Hand it off to people on the ground, the ones who will actually use this on a day-to-day basis. And two things happen:

1. The project doesn’t bring the magical impact that management had imagined. Not because Company B did a bad job, but because the use case was bullshit to begin with.

2. Or, Company A’s in-house engineers figure out they can just replicate this internally, no need to pay Company B.

   
   

So the project gets cancelled.

There’s also a third scenario: the use case was so absurd that it literally can’t be solved, but Company B still takes it. Because hey, some revenue is better than no revenue.

And this loop continues.

Until one fine day, management at Company B wakes up and says:

“We’ve been playing the POC game for too long… we need a win for the team.”

And that, folks, is the story of POC Hell.