The case against the AI agency
Artificial intelligence has created an entirely new category of digital agencies almost overnight.
Browse LinkedIn for a few minutes and you'll find hundreds of companies promising AI transformation, AI consulting, AI automation and AI strategy. The technology is real. The demand is real. The marketing, however, often tells a different story.
Somewhere along the way, "AI" became the product instead of the means to build a better one.
That's a mistake.
Businesses rarely wake up wanting artificial intelligence. They want fewer repetitive tasks, faster customer support, better forecasting or more efficient operations. AI may happen to be the right way to achieve those outcomes, but it's only one of many possible approaches.
When technology becomes the headline, the problem it's supposed to solve quietly disappears.
We've seen proposals where "AI integration" is listed as a deliverable without explaining what it actually improves. Chatbots are added to websites where customers never asked for conversational support. Image generation appears inside products where search would have solved the problem more effectively. Automation is introduced for workflows that happen once a month.
The result is software that feels modern without becoming useful.
Good engineering has never been about using the newest technology. It's about choosing the appropriate one.
There are situations where AI genuinely changes what's possible. Document analysis that once required hours can happen in minutes. Customer enquiries can be categorised automatically. Large knowledge bases become searchable through natural language instead of rigid navigation. Internal workflows become dramatically faster when repetitive decisions are delegated to intelligent systems.
Those are meaningful improvements.
But they begin with understanding the workflow, not selecting the model.
The most successful technology projects we've encountered share a common trait. Nobody talks about the technology after launch.
Nobody praises the database.
Nobody compliments the API architecture.
Nobody asks which cloud provider was used.
People simply notice that the product works.
Artificial intelligence should aspire to the same level of invisibility.
When every button is labelled "AI-powered," the interface starts competing with itself. Users don't need constant reminders that a machine learning model exists somewhere behind the scenes. They care about outcomes.
Can I finish this task faster?
Can I trust the result?
Can I correct it if it's wrong?
Those questions matter far more than the underlying implementation.
The companies likely to benefit most from AI over the next decade won't necessarily be the ones talking about it the loudest. They'll be the ones quietly redesigning workflows, removing friction and making complicated systems feel surprisingly simple.
That's a far less exciting marketing message.
It's also far closer to reality.