10 Questions for AI leaders: Paul Weston

AI is done being a party trick. The real story now is what happens when it moves into the plumbing of how companies actually work.

Paul Weston, VP of Product for HubSpot’s Service Hub, takes a look at AI’s next chapter - where the hard parts aren’t prompts or pilots, but systems, accountability, and what happens to customers long after the demo is over.

I’m excited that we’re finally moving past the "AI as magic" phase into "AI as a core tool." 

I love watching companies realize they need to rethink entire systems, not just bolt AI onto broken processes. This forces necessary conversations about what work actually matters and about ideal customer experiences.

The accountability gap. A lot of companies are automating decisions faster than they’re building systems to monitor outcomes. Everyone's racing to deploy, but who's measuring whether the AI actually improved the customer's life six months later?

AI should make humans more accountable for outcomes, not less.

I used to think AI's job was to "augment humans." Now I realize that's too vague. The real shift is that AI forces us to be explicit about what we're optimizing for. You can't deploy AI without defining success metrics, edge cases, and feedback loops. Systems thinking is what will actually transform organisations.

My simplified framework:

The real question isn't "can AI do this?" but "do we understand this process well enough to encode it?" If you can't draw the decision tree, you're not ready to automate.

Systems thinking, hands down.

We're about to have an explosion of point-solution AIs that each work beautifully in isolation but create chaos when combined. The people who can map entire journeys, identify handoff points, and design for system-level outcomes are going to be invaluable. We need more "AI orchestrators" and fewer "AI enthusiasts."

Yes and no.

The technology is real and transformative, that's not the bubble. The bubble is in expectations about how fast and how easily organizations can capture value.

Everyone's funding AI pilots. Very few are funding the unglamorous work of data cleaning, process redesign, change management, and outcome measurement. In a year or two, we'll see a wave of "AI didn't work" articles. There are actually a lot of these examples today. But it won't be because AI itself failed. It'll be because organizations treated it like a product you buy, not a capability you build.

The correction will be healthy. It'll separate companies doing real transformation from those doing AI theatre.

Trust is absolutely critical, but ultimately I think reliability is what matters.

Of course, you want customers to trust every single aspect of your business. For a SaaS company, is there any feature you’d feel comfortable being low-trust for your customers? 

Most importantly, I think customers need to trust that when something goes wrong someone (a human) will fix it. Trust can be built by:

The companies that win will stop asking "how do we make customers trust our AI?" and start asking "how do we prove we're accountable for what our AI does?"

Starting AI projects with a “pre-mortem” or a "failure mode workshop."

"What are all the ways this could go wrong? What would we see? How would we know?"

This tiny ritual forces teams to think in systems, not features. And it creates a paper trail of considered risks, which is helpful when something inevitably breaks.

Where AI wins: Proactive pattern recognition. "We noticed you've called three times about billing. Let me connect you directly to someone who can fix the underlying issue." AI can spot patterns across millions of interactions that no human could see.

Where humans win: When someone says "this is really frustrating" and means "I feel disrespected." AI can detect sentiment, but it can't understand that the customer doesn't actually want a discount. Emotional repair requires a human who can go off-script.

The realization that LLMs are better at following complex instructions than humans are at writing them.

We've spent years simplifying processes so humans could execute them consistently. Now we have AI that can handle complexity, but we've discovered most organizations don't actually understand their own processes well enough to explain them.

This flips everything. The bottleneck isn't AI capability, it's institutional knowledge. The companies that will win are those investing in process documentation, edge case libraries, and outcome measurement. Boring stuff. But that's where the actual leverage is.

"More accountable for what actually matters"

The grunt work will be automated. The excuse of "I didn't have time" largely disappears. What's left is the hard part: judgment, prioritization, and taking responsibility for outcomes.

Some people will find this liberating. Finally, they can focus on work that requires their unique human judgment. Others will find it terrifying because it’s a big shift from business as usual.

The question isn't whether AI will take your job. It’s about evolving your skills and mindset with technology to deliver value in increasingly impactful ways. AI is about to make that distinction very clear.

Paul Weston is VP of Product for HubSpot’s Service Hub product line, powered by HubSpot's Breeze AI tools.