Key takeaways
I bought a new TV recently. The first thing it asked me to do when I turned it on was enable its AI feature.
Seriously? I just wanted to watch TV.
That moment is a pretty good summary of where we are right now. Every company in your tech stack is knocking on your door, demanding you turn on their AI functionality. Your ticketing platform, CRM, knowledge base, and even your TV. Every vendor has an announcement, a press release, a new SKU. And if you're a buyer trying to actually evaluate what any of this means, I have a lot of empathy for you, because it is genuinely hard to cut through.
Here’s the quiet part most people won’t say out loud: most of what's being marketed as AI isn't really AI-native at all. It's AI-added.
And that distinction matters more than anything else you'll evaluate this year.
What does “AI-native” mean?
I get asked a lot about what makes a platform truly AI-native versus one that's just tacked AI on as an afterthought. And I've landed on the simplest possible litmus test: if you turn the AI off, does the product still work?
For a legacy platform with AI features, the answer is yes. The core product exists. The AI is an add-on, an enhancement, a layer. You can disable it, and the software hums along just fine.
For a genuinely AI-native product, the answer is a resounding no. For these platforms, AI isn't a feature; it's the engine. It's woven into every part of the codebase, every workflow, every decision the system makes. Take it away, and you don't have a product anymore.
Once you start evaluating through that lens, the market clarifies real fast.
Why legacy vendors build the way they do
To be fair (and I do want to be fair here), I don't pity the incumbents. I understand the position they're in.
If you're a platform serving hundreds of thousands of customers spanning SMB to enterprise, B2C to B2B, you face an impossible customization problem. Every customer has different workflows, different data structures, and different definitions of success. So what do you build? You build for the middle, the average. You build something that mostly works for most people.
Think about the AI-generated meeting summary you got after your last call. It was probably fine. But, if you’re honest with yourself, it was probably pretty vanilla. And it was vanilla on purpose—because it was built for everybody, which means it was built for nobody in particular.
That's the fundamental tension at the core of every legacy AI feature: the promise of LLMs is that you can solve your specific business problems, not a generalized market's business problems. But if you're trying to serve a market of hundreds of thousands, specificity is the first thing you sacrifice.
It’s an integration problem at its core
Here's what I believe is the real differentiator between AI-native and AI-added platforms, and it's not what most people focus on.
It's integration.
AI is only as good as the data you can feed it. Garbage in equals garbage out, like with any tool. And in enterprise SaaS, that data lives everywhere. Sales teams work in Salesforce, CS teams live in Gainsight, Support runs in Zendesk. There isn't a single source of truth. There are six partial ones, and everyone is fighting to be the canonical one.
This creates what I'd call an innovator's dilemma for legacy vendors. Should I build deep integrations with my competitors' data? Because if I do, what happens to my core business? The answer, for most of them, is no. They stay in their lane and optimize for their platform, yet the integration gaps persist.
For AI to actually work (and I’m not just talking about a great demo, but actually work and at scale), you need to be able to pull structured signal from all of those systems simultaneously. You need a platform that was designed, from day one, to be agnostic about where data lives and obsessive about connecting to it. That's an architectural decision you can't retrofit.
The "expensive experiment" pattern
I've talked to enough support leaders in the last few years to recognize a pattern. Someone gets inspired by ChatGPT, realizes their engineering team can connect an API to a data source, and spins up a Slack bot that answers questions from a knowledge base. It takes a couple of weeks. It works in the demo.
Then they try to scale it. And that's where it falls apart.
Because there are a hundred things under the hood that a Slack bot doesn't address: Security, role-based access, volume handling, reporting and analytics, and multi-persona workflows across different interfaces
The requirements that seemed simple become an engineering project that takes quarters, not weeks. By the time something ships, the AI landscape has already moved.
What I've come to believe is that the gap isn't really about the AI itself. It's about everything around the AI. The integration infrastructure. The observability layer. The ability to customize the last mile so the output is relevant to your support org, not a generic one.
This is what we refer to as the last-mile fit, which is exactly what most internal builds and most AI-added platforms can't deliver.
The question to ask every vendor right now
If I were sitting on the buy side right now, evaluating AI for B2B support, here's the question I'd be asking every vendor:
If I turned your AI off tomorrow, what would I have left?
A good product with an AI feature is still a good product. Nothing wrong with that. But in B2B support, the edge case is the job.
You're not automating high-volume, low-complexity tickets; you're trying to get the right context to the right engineer at the right moment on a case that might take weeks to resolve. That's not a problem you can solve by adding a feature onto an existing ticketing platform.
So if you want AI to genuinely change the economics of your team, you need something where that question doesn't have an easy answer.
Because the vendors who can't answer it? They're selling you a TV with an AI button.
Josh Solomon is GM & VP of Revenue at Ask-AI, where he works with enterprise B2B support organizations to help them operationalize AI at scale.
