Let me explain what I mean.
At Omilia, we’ve been building AI solutions for 23 years. Long before “generative AI” became a dinner table conversation, long before every company started slapping “.AI” on their name and calling themselves a tech company.
(I read an article recently about a shoe company that gave up on shoes entirely to rebrand as an AI company. I’m still not sure what to make of that.)
The point is, AI has been around and doing real work for a long time. What’s changed is how we apply it, how we scale it, and how we actually measure whether it’s working.
And right now, a lot of organizations are hitting a wall they didn’t fully see coming…
The Sisyphus problem in AI automation
About a decade ago, there was a widespread belief – shared by Nuance, Omilia, and plenty of others – that AI was going to solve every customer inquiry.
Organizations invested heavily. They built centers of excellence. They started automating conversations and making self-service genuinely easier for customers.
And for a while, it worked beautifully.
Most companies got to 50, 60 percent containment. Strong start. Real results.
Then something interesting happened. The progress slowed. The journey from 60 to 85 percent turned out to be nothing like the journey from zero to 60. If you’re familiar with the myth of Sisyphus – the harder you push the rock up the hill, the harder it gets – you’ll recognize this pattern immediately.
For expert advice like this straight to your inbox every other Friday, sign up for Pro+ membership.
You’ll also get access to 300+ hours of exclusive video content, a complimentary Summit ticket, and so much more.
So, what are you waiting for?
Why does it get harder? A few reasons that tend to compound on each other.
Operational costs increase as you push further. Token costs – fractions of a cent, sure, but they add up fast at scale – start demanding their own measurement infrastructure.
Average handle time with agents actually goes up, not down, because everything that’s left after automation is the hard stuff. The complex conversations. The emotionally loaded ones. The edge cases that don’t fit neatly into any workflow.
And resolution rates plateau. The APIs aren’t broken. Your team isn’t doing anything wrong. It’s more fundamental than that. It comes down to the real-world complexity of human communication, and whether your AI strategy is genuinely built to handle it.
