A new era of support has begun
In just a few short years, AI has gone from science fiction to everyday reality. “We now live in a world where AI is able to have ongoing conversation in natural language on general subject matter,” Shan told the audience at Mavenoid’s Product Support Summit.
It’s easy to forget how monumental this is. For decades, experts believed this kind of general intelligence was impossible without AGI. And yet, here we are, with machines capable of holding conversations that sometimes even fool humans.
“If you think about it in terms of how important it is for humanity, it’s probably bigger than the moon landing,” Shan said. “It’s definitely a new era for customer support, the most obvious application of all.”
The rise of AI Slop
But not everything is rosy in the world of AI. Shan didn’t shy away from calling out what he sees as a growing problem…AI slop.
“AI slop basically means low-quality generative content – generic stuff, unoriginal things, uninspired things, things that are not true, things that are not trustworthy,” he explained.
Recent examples from the headlines with chatbots giving out wrong prices, refund policies gone rogue, or bots that start insulting customers, are all symptoms of the same issue, teams prioritizing flashy demos over dependable systems.
“You shouldn’t optimize for how good the demo looks,” Shan warned. “You have to be able to trust the bot. You have to know when to trust it.”
The principles of reliable AI
So what does non-sloppy AI look like? Shan laid out a set of design principles that every company should follow when building AI-driven support:
- Avoid needless customer effort.
Don’t make people craft elaborate prompts. - Optimize for resolution, not conversation.
“People aren’t there to chat, they’re there to fix something.” - Don’t play fake human.
Overpromising on empathy and underdelivering on accuracy only frustrates users. - Ground everything in real data.
“The more you ground things, the less room there is for bullsh*tting.” - Enhance context.
The more you know about the user and their problem, even down to showing a photo, the better the AI can perform.
Rejecting false trade-offs
Companies often feel forced to choose between coverage or curation, autonomy or alignment, scalability or safety. Shan believes these trade-offs are false.
“Great AI support rejects all these false trade-offs,” he said. “You can have depth and dependability, autonomy and alignment, scalability and safety.”
How? By being deliberate. For instance, you can allow generative systems to cover a broad range of questions, but inject curated content where accuracy matters most.
Human enhancement in an AI world
One of Shan’s key messages was that even as AI takes on more of the heavy lifting, humans still play a vital role.
“This is AI-driven now, but it still needs human enhancement at every level – in real time and in how you design the experience.”
The best systems aren’t autonomous replacements, they’re AI-assisted ecosystems, where humans teach, guide, and correct the AI to make it smarter and more dependable over time.
