Clearly, Bernard Meyer, the head of AI initiatives at the email and SMS marketing platform Omnisend, has given a lot of thought to why AI adoption fails. His conclusion? “Most AI work is bullshit, and vibecoding is a trap.”

That was the name of Meyer’s keynote at a recent LOGIN tech conference in Vilnius, Lithuania. Quite obviously, Meyer has plenty of ideas on how to fix it all, doesn’t he?

We’ve written a bunch of stories about how the so-called AI visionaries are selling snake oil to the world and are now having to ward off concerned executives who don’t really see any productivity gains when AI usage is adopted across companies.

What they see are higher costs. Told by their bosses to dive into the AI pool, employees are burning through myriads of tokens – and cash – but the output simply isn’t meeting expectations.

Uber president and COO Andrew Macdonald recently told the Rapid Response podcast that the company cannot yet draw a clear connection between rising Claude Code token consumption and the additional useful consumer features it has shipped.

The rideshare giant soon told Bloomberg it was limiting all employees to $1,500 in monthly token spending per AI coding tool.

Naturally, I was expecting Bernard to almost gleefully say that the bubble might be bursting and that human professionals will still need to be in charge, also because AI is, quite frankly, very far from helping businesses make any money or even break even.

Well, soon after sitting down to chat with Bernard soon after he spoke on stage, I realized that wasn’t actually his message at all. In fact, there’s nothing wrong with AI – we just have to learn how to use it.

5 stages, 5 bullshits

Meyer argues that most companies are still in the experimentation phase of AI adoption, mistaking a flurry of activity for actual progress.

He has mapped AI adoption into five stages:

• Awareness
• Experimentation
• Institutionalization
• Institutionalization-plus
• Full transformation

According to Meyer, the vast majority of organizations worldwide haven’t yet reached the middle stage.

At Omnisend, with its 52 teams, AI has become so embedded into daily operations that people don’t even say the word “AI” anymore.

“It just runs in the background: autonomous agents pushing morning briefings to Slack, custom-coded tools shaving minutes off repetitive tasks, engineers managing four concurrent AI-generated pull requests the way they once managed junior developers,” Meyer told Cybernews.

But that progress came with hard lessons, which Meyer distilled into what he calls “the 5 bullshits” of AI adoption.

• There’s the vague mandate from leadership (just use AI, won’t you?)
• The productivity theater of individual workers claiming to save 25 hours a week without any baseline data to support it
• The addiction to shiny new projects that never get properly adopted,
• The assumption that every employee needs to build AI solutions rather than simply manage outcomes
• “Finally, the most uncomfortable one is the belief that everyone in an organization will make it through the transition. Some people won’t. Some people don't want to change or can't change. So we need to prepare for that in advance,” says Meyer.

Is he not changing his mind about AI usage after most AI giants have restricted token budgets and shifted toward outcome-based pricing? Not at all – Meyer states that even giants such as Uber should just stop and think about what exactly they want to achieve with AI.

“A lot of companies are using open source models, and they’re very successful. But when it comes to enterprise organizations, they use AI for more reasons than actual work stuff,” says Meyer.

“With Uber, at some point, they could have said, ‘we’re using 60% of our budget – let’s look at what we’re using it for.’ They were using the fanciest model for every single thing. Whereas the smart way is to have a strong thinking model that understands what needs to be done and routes it to lower models that can handle it with specific instructions.”

“The answer is very clear”

As a skeptic, I still asked Meyer whether the use of AI across Omnisend has actually resulted in productivity gains or made the teams faster.

Meyer admits that Omnisend had a stage of measuring hours saved by using AI. Soon, though, they concluded it was all “productivity theater.”

“I can say I now write my email 30 minutes faster. I write ten emails per day, so that’s five hours a day, 25 hours a week. But this is bullshit because I never measured how long it took me to write an email before,” he says.

“And I don’t measure how long it takes now. It's all assumptions and estimations. I have no benchmarks. So you can’t talk about productivity if you never measured time tracking before AI.”

Meyer is more keen to talk about tracking outcomes. For instance, Omnisend’s influencer marketing team once needed to hire someone but decided to try something else – AI – instead.

“I custom-coded them a solution, and eventually we realized we don’t need to hire that full-time person at all. That's a real outcome. But attributing it to AI specifically? That’s hard. If a team believes in what they’re doing, they'll move faster anyway. There’s no AI involved in that,” Meyer told Cybernews.

Still, he thinks that even though there’s no concrete proof just yet, the answer – if you ask senior managers or engineers whether AI makes work faster – is very clear.

“The other thing with AI is that it’s very lagging. Adoption time is going to be 6, 9, 12 months. AI-powered work is going to happen in 6 months, 9 months. And then, the results are going to come after 18 months,” says Meyer.

---

Unlock more exclusive Cybernews content on YouTube.