“An imitation of life”
"We want to use Agentic AI to replace our SaaS. And we want autonomous agents to replace some people. And we'd like to do that right now. Like, today" our client said.
The eagle-eyed film buffs among you will know that the title of this blog post is a quote from the science-fiction film “I, Robot”, which in turn was based on the book of the same name by Isaac Asimov.
But what has that got to do with SPG Intelligence?
Well, we’re building a proof of concept for a client at the moment. The outcomes are pretty clearly defined; the "big hairy audacious goal" is to get rid of as much SaaS bloat as possible. Arguably, given less steps in the workflow and more automation, it's not inconceivable that headcount reductions may be possible in the longer term.
"Everyone's talking about agents" they said. "How quickly can we have them?".
Well, unfortunately, when you start to bring concepts into the real-world, things have a habit of becoming more complex. Of slowing down. Of maybe not delivering the dream you'd read about online.
For AI Agents to actually work in real-world enterprise environments, they need more than a smart model. They need infrastructure. Integration. Context. Control.
And that’s the hard bit.
Most workflows aren’t clean. They’re fragmented, legacy-ridden, and stitched together with brittle logic, clunky permissions, and weird data structures. Dropping a model on top of that and expecting magic is wishful thinking.
Just like the getting internet to as many households as possible, the last mile is always the hardest. You need software to bridge the gap between the AI and the messy operational reality:
You need some sort of software to help plug-in systems that were never designed to be friends.
You probably need software to pull or reference the right data at the right time.
From a security perspective, it's crucial to honour roles, permissions, SLAs and edge cases. That'll need software.
And you'll probably want the ability to enable human override where it counts
This is all before you get to what the customer actually wants; support, liability, integration partners, tailored pricing, category-specific onboarding.
Every industry - and even some horizontal workflows - need this layer built with domain-level precision. The closer you are to the problem, the more durable your value becomes.
Yes, models and agents will improve. That’s not the threat - it's actually the opportunity if you start now.
But expecting the full-on "I, Robot" experience today, in 2025 is probably unrealistic.
Today is more about foundation than revolution.