The idea of “replacing employees with AI” has become a siren call in business circles. Many companies have followed it and many are now regretting it.
As a co-founder of Ascendix, a software development and consulting company that has worked with real estate and financial services firms for over three decades, I’ve seen first-hand how leaders approach AI adoption. I’d like to share a perspective from inside the industry to help decision-makers avoid common pitfalls.
The George Washington Test
Ask me five facts about George Washington, supposing I know nothing about him. As a human, I will say, “I don’t know. Let me research this.” What an LLM model could return to you is:
- George Washington was the President of the United States of America.
- George Washington’s name is George.
- George Washington’s surname is Washington.
- Washington is the capital of the United States of America.
- George Washington won the election with 59% of the electoral votes in 1787.
By the way, the last one is a lie.
This is what AI does so convincingly: It wraps common knowledge into insightful language sprinkled with some empty words. Then it looks for data on similar terms, even if they are barely related, to return as much text as possible, so you consider it a comprehensive answer. When it has nothing to add, it hallucinates, often creating made-up statistics to show you some numbers, and adding bullet points for a sense of structure.
On trivial topics, it’s harmless. But when executives lean on the same process to make strategic or operational decisions, the risks become very real.
And that brings me to two disappointing truths for those hoping to replace their teams with “AI agents”:
- Most “agentic” solutions are not autonomous per se. They can’t replace people. Most things tagged as “made by AI” are actually “made with AI”.
Remember how Builder.ai topped all the lists of the most promising startups? Investors were so excited to get rid of expensive software developers that they didn’t listen to the reasoning why it was impossible. We know what happened next.
And this isn’t an isolated case. We’ll likely see similar stories repeat: ambitious promises of full automation, when in practice, there’s still a team of people behind the scenes keeping the system running.
- AI works more like a cautious intern than a seasoned employee. It can follow instructions, but it won’t think outside the box, and it only admits mistakes when caught. The problem is that people double-check after interns and unquestioningly believe AI.
And unlike interns, AI doesn’t grow into a mid-level specialist with guidance. Training it to reach that level of judgment is far more expensive and time-consuming.
Lessons from Early Experiments
Some companies learned this the hard way.
Duolingo tried a similar approach and faced public backlash. Users now report more technical issues and language mistakes. Nonetheless, the stock price may keep rising because investors want AI, while users, the core of the app, want a human-made experience (even if AI helped to create it behind the scenes).
Another example is when Klarna tried to replace 700 customer support workers with an assistant powered by OpenAI. This was not even a specialized solution. Klarna is a buy-now-pay-later service that stores financial information. Logically, users were concerned about how this data would be protected when AI accessed it. The situation escalated quickly and led to a dramatic $40 billion loss and a decline in service quality.
Soon enough, Siemiatkowski, Klarna’s CEO, admitted, “We went too far with AI. As cost, unfortunately, seems to have been a predominant evaluation factor when organizing this, what you end up having is lower quality.” Now, they are trying to hire everyone back.
A Smarter Path Forward
AI can help streamline operations. The tricky thing is not to kill your human touch with it. Invest in people. Teach your employees how to use AI in their work. Do not force AI on them but encourage adoption.
At Ascendix, we invest in people first. Every employee has access to our private, OpenAI-powered platform, where they can build custom GPTs tailored to their roles. They’re encouraged to share those experiments during internal workshops. On top of that, we run structured training programs, curate specialized courses, and provide ongoing resources.
These are just a few examples, but the principle is simple: we help people integrate AI into their work responsibly and creatively – not as a crutch, but as a tool to enhance their expertise.
Closing Thought
Orgvue’s annual ‘State of the Nation’ perspective on AI and workforce transformation survey, published this spring, shows that 55% of UK business leaders who rushed to downsize their teams due to automation now admit they made a mistake.
And I believe the number is going to keep going up. The yellow brick road to cutting expenses on human expertise is still under construction, and it looks like the bricks are not even yellow.
AI optimism will take turns with AI regret, and eventually, we will find a way to use AI effectively, ethically, and sustainably.
Maybe I’m too skeptical. Or too optimistic.
