# AI Won't Fix Your Company by Itself

Author: Norbert Biedrzycki
Published: 2026-09-15
Language: en
URL: https://norbertbiedrzycki.pl/en/ai-wont-fix-your-company-by-itself/
Summary: Over the years I've watched dozens of tech fads come and go. Big data, blockchain, "digital transformation." The ritual was always the same.

> *"We buy AI like a shiny new calculator, then wonder why it doesn't add up the whole company for us."*

## A calculator, not a colleague

Over the years I've watched dozens of tech fads come and go. Big data, blockchain, "digital transformation." The ritual was always the same. A board buys licenses, announces a transformation, and the organization quietly slides back into its old habits.

Today I see the exact same pattern with AI. Most companies treat it like a shiny new calculator. They buy access to a chat tool, hand people one more window to open, and expect productivity to jump on its own. What they usually get is the opposite: information chaos, scattered data, and a team worn out from copying prompts between systems.

## An example worth sitting with

In February 2024, Klarna announced a result that made headlines across the business world. Its AI assistant, built on OpenAI's models, handled 2.3 million customer conversations in its first month, doing work roughly equivalent to 700 human agents. Average resolution time dropped from eleven minutes to under two, and the company estimated annual financial benefits of about 40 million dollars.

A little over a year later, though, Klarna's CEO admitted the company had leaned too hard into automation. AI had genuinely improved efficiency and was still handling a large share of customer contacts, but the heavy focus on cutting costs came at the expense of service quality. In response, the company went back to hiring human agents and made clear that customers should always be able to reach a person.

So what actually happened? Klarna didn't make a technical mistake. The model worked exactly as it was designed to. The mistake was assuming that deploying AI would automatically improve the whole process around it. What didn't get redesigned properly were the escalation paths, the exception handling, who owns which decision, and the situations that need broader context and human judgment.

## The real cost isn't where we're looking

The real cost inside a company isn't the time people spend thinking. It's the time spent connecting the dots: making sure CRM data matches what's happening in production, that invoices line up with contracts, that a decision made in the boardroom actually reaches execution further down the chain. This is slow, invisible work, and it eats more hours than anything else in every organization I've worked with.

A chat tool, even a great one, won't do that work. It will answer a question if someone asks it. It won't notice a mismatch in the data unless someone thinks to check.

![An AI operator connecting CRM data, invoices and board decisions](https://norbertbiedrzycki.pl/images/li/fix-inline0.jpg)

## A shift most boards haven't caught up with yet

That's why I believe AI shouldn't be one more browser tab where a human types in questions. The future is autonomous agents, digital colleagues that work in the background, carry out real operational tasks, and connect those dots themselves instead of waiting to be told where to look. Klarna's story isn't proof that this path is wrong. It's proof that automation without redesigning the process and without clear decision boundaries ends in a costly course correction.

## Sophona

At Sophona we didn't want to build another bot for writing emails. We're building AI Operators, and specifically the kind we would happily put to work in the boardrooms and operational teams of the companies we've worked with: with a clearly defined scope of decisions, human checkpoints wherever the stakes are high, and a memory of context that a regular chat tool simply doesn't have.

**What's holding your teams back today from adopting AI at a truly operational level, not just the "experimental" one?**
