Everyone keeps telling me that AI is going to revolutionize productivity. I’m reading about tech companies promising it, consulting firms selling it, and governments planning roundtables about it. For instance, next week, Australia’s federal government is hosting one of these productivity love-fests where AI takes center stage.

But I found myself pondering one night if this stuff actually works. So I dove into the research, talked to a few companies, and some public servants who actually use this technology, and tried to separate the marketing hype from reality.

And what I found was messier than the truth coming from Grok.

---

More rants after the messages:

Join me and Eliska Went on September 5th when we show you how to create your own automated AI tool to find, write and apply for jobs. Register here. Here, I mean: here

1. Comment, or share the article; that will really help spread the word 🙌
2. Connect with me on Linkedin 🙏
3. Subscribe to TechTonic Shifts to get your daily dose of tech 📰
4. Visit TechTonic Shifts blog, full of slop, I know you will like !

---

The promise vs the reality

The numbers sound impressive at first glance. A company Nielsen Norman Group found a 66% average productivity boost across their studies, and Microsoft says that 75% of knowledge workers are using AI, with 90% saying it saves them time. McKinsey waves around a $4.4 trillion productivity growth figure like it’s confetti.

In software development, GitHub Copilot apparently makes coders 55% faster. Manufacturing sees 5-20% productivity gains and 15% less downtime when machines stop having their mechanical issues. Healthcare workers report 30-50% productivity increases, and financial services averages a 20% bump across all AI applications.

Those are some pretty sexy statistics.

But then reality crashes the party.

A Danish study that is tracking 25,000 workers found that although AI adoption rates hit 47-83%, the actual economic impact was – and I quote – “precisely estimated zeros”. Now, that is academic lingo for “zilch, nada, nothing to see here.”

Another study discovered that only 3-7% of these magical productivity improvements actually translate into higher worker earnings. The rest just vanishes into the corporate thin air.

---

Managers are spilling the tea

For this piece, I have talked to a few of senior tech managers, the folks who actually have to make this AI stuff work in the real world. These are the people buying, using, and managing AI tools, and not the ones selling them with PowerPoint slides full of hockey stick graphs.

Their first revelation was that introducing AI to existing workflows works about as smooth as me trying to do some roller skating after having a drink or two. It is slow, expensive, and requires a lot of of organizational gymnastics.

One of my contacts described it as “like driving a Ferrari on a smaller budget. Sometimes those solutions aren’t fit for purpose for those smaller operations, but they’re bloody expensive to run, they’re hard to support”.

The well-funded organizations can afford to play around with different AI toys for “proof of concept” testing, but smaller operations are stuck trying to figure out how to pay for these tools without having to sell a kidney.

---

The devil is in the data

Off-the-shelf AI tools like Copilot and ChatGPT can handle simple tasks like summarizing meetings, extracting information from documents, creating an email, writing you a letter, or turning your rambling voice notes into something coherent.

That’s the easy stuff.

But when organizations want AI to do some real work, like running call center chatbots or building internal information systems, they need to feed the AI their own data, and this is where most companies discover that they’ve been data hoarders, collecting data like it’s vintage vinyl but never organizing any of it.

As one tech manager told me, “data is the hard work”. Without clean, well-structured data, these AI tools do not perform well. Companies realize they need to invest serious money in data profiling, data governance, and cleanup before their AI dreams can come true.

---

Privacy paranoia and security nightmares

Using AI means that your organizational data goes on a magical mystery tour through servers owned by tech giants inside and outside your country. These companies pinky-promise they’ll keep your data safe and won’t use it to train their AI systems. But the tech managers from government organizations I talked to were about as trusting as my Weiner near a vacuum cleaner.

People might innocently paste sensitive information into ChatGPT, not realizing they’re essentially broadcasting their company or their client’s secrets to the internet, or AI vendors could add new features that create new data flows without anyone noticing, like a digital Trojan horse delivering privacy violations. Read: Google was showing the ChatGPT chats you shared with others | LinkedIn

The research I did backs up these concerns.

Organizations that are handling sensitive information need constant monitoring to make sure the AI tools comply with legislation and internal governance rules.

---

Productivity measurement

Measuring AI productivity is somewhat trickier than nailing jelly to a wall. Managers rely on feedback from their most AI-savvy workers to give them feedback on the operation with AI, but those are the ones who already love technology and would probably find ways to be productive with a pocket knife and some string.

One guy I talked to was refreshingly honest about this measurement problem: “I’m going to use the word ‘research’ very loosely here, but Microsoft did its own research about the productivity gains organizations can achieve by using Copilot, and I was a little surprised by how high those numbers came back”.

Luckily this means that even the people buying this stuff are side-eyeing the vendor claims.

Companies want AI to either improve efficiency, increase output or improve the relationship with their clients, but they are not measuring whether their products and services actually get better.

And they are also ignoring how the workplace changes for the humans who have to babysit these AI systems all day.

Organizations are also using AI to monitor the AI.

They are implementing enhanced workplace surveillance to make sure their people aren’t using AI inappropriately, for security and privacy reasons and to prevent shadow AI. They are basically hiring security guards to watch the security guards.

Now that’s really fucked up. A recent survey found that these surveillance methods might actually harm workers. So we’re boosting productivity by making people more miserable.

Wow.

That’s truly some next-level business thinking right there.

---

The cognitive load

The research also shows that AI works best for what everyone politely calls “low-skill” tasks, like taking meeting notes, basic customer service, and work typically done by junior employees. The peeps at Microsoft found that 79% of Copilot (again) users report reduced cognitive load, and 87% of developers say that AI preserves their mental energy during repetitive tasks.

This sounds great but when you think about the fact that the people who need AI help the most, you know, those with less experience and skill, they are also the least equipped to double-check AI output for accuracy.

Also, when jobs become primarily about watching an AI system work, people report feeling alienated and less satisfied with their work.

Duh.

Just imagine spending your day being a hall monitor and poop scoop for a farkin robot – not exactly the career satisfaction most people dream about.

---

The numbers game

If you put all these complications aside, there’s a whole lot of research out there that consistently shows that AI does increase productivity in controlled conditions.

Healthcare sees 30-50% gains in nursing and 17% time savings for doctors. I am working on a real-time translation project including transcription and for each hour spent with clients, there’s 30 minutes time shaved off.

Now that is impressive.

Manufacturing has always benefited from predictive maintenance and quality control improvements, and financial services automate risk assessments and fraud detection with impressive results – but truth be told – there projects have been done with battle tested Machine Learning models, and not with hallucinating Neural Nets.

The Nielsen Norman Group’s 66% productivity increase wasn’t a fluke though.

It was measured across multiple controlled studies. Software developers really are 55% faster with GitHub Copilot. These aren’t marketing fairy tales.

But controlled studies aren’t the real world.

In labs, everything is optimized, data is clean, and people are motivated to make the technology work, but actual workplaces are messy. You’ve got legacy systems, messy data, skeptical employees, and budgets tighter than my skinny Chinos after a Christmas dinner.

The bottom line is that AI does boost productivity, but it’s not the simple story the tech evangelists want you to believe. The gains are real but unevenly distributed. The costs are higher than advertised, and most of the times, the implementation is harder than expected.

And most of the benefits flow to the companies selling AI tools rather than the workers using them.

The Danish study’s finding of “precisely estimated zeros” in economic impact isn’t because the AI doesn’t work, it is because the gains get absorbed by increased costs, and complexity, and the need for human oversight.

Or maybe I’m overthinking this, and we should just enjoy watching robots try to figure out why humans make such delightfully irrational decisions all day.

Either way, the productivity revolution is here, but it’s just a bit messier, more expensive, and more human than anyone expected.

Signing off,

Marco

---

I build AI by day and warn about it by night. It’s job security. Big Tech keeps inflating its promises, and I just bring the pins.

---

Think a friend would enjoy this too? Share the newsletter and let them join the conversation. Google and LinkedIn appreciates your likes by making my articles available to more readers.

To keep you doomscrolling 👇

1. The chicken that refused to die | LinkedIn
2. Vibe Coding is gonna spawn the most braindead software generation ever | LinkedIn
3. The funniest comments ever left in source code | LinkedIn
4. Today’s tech circus is a 1920s Wall Street reboot | LinkedIn
5. The Sloppiverse is here, and what are the consequences for writing and speaking? | LinkedIn
6. Why we’re all writing like shit on purpose now | LinkedIn
7. Creating a hero’s journey for your AI-strategy | LinkedIn
8. How corporations fell in love with AI, woke up with ashes, and called It ‘innovation’ | LinkedIn
9. The corporate body snatchers | LinkedIn
10. Screw your AI witch hunt, you illiterate gits | LinkedIn
11. AI perverts make millions undressing your daughters for fun and profit | LinkedIn
12. The state of tech-jobs 2025 | LinkedIn
13. Meet the sloppers who can’t function without asking AI everything | LinkedIn