Last month, Bret Taylor (Chairman of OpenAI and CEO of Sierra) said he wants customer support AI agents to be “paid on commission” for closing tickets. If the agent resolves your issue, it gets paid. If it hands you to a human, it’s free. He called it “great incentive alignment” and “right from first principles.”
On the basics, Bret and I probably agree: most issues, possibly up to 90%, should be handled quickly and automatically by AI. Think: simple knowledge questions, password resets, obvious refunds.
Where we part ways is what happens with the rest. Incentives should create great experiences, not push bots to avoid humans. The problem with having your AI be “paid on commission” is that you’re making your AI into a used car salesman: the AI is going to try to handle your issue as fast as possible, avoid connecting you to a human, and not prioritize giving you a good experience. It’s about the speed of closing the ticket.
But is this what consumers actually want? At this point in our emerging relationship with AI technology, I continue to hear stories underscoring the fact that humans want other humans in the loop when it comes to support. They care disproportionately about how a company handles them.
I recently tried to return final-sale running shoes. The chatbot stonewalled me, sent me in circles for a half hour, then finally handed me off. A full three days later, an agent appeared and restarted the process from scratch. It was frustrating. And this is the point: moments like this shape how you feel about a brand. Optimizing only for “cases closed by AI” misses where loyalty is won.
That’s why the more important question is how companies choose to design support itself. Before you can think about automation, you need to understand what agents actually do and what customers really value in those moments. In this way, the debate shifts from simply cost savings to what actually builds brand loyalty.
A case for responding to support tickets
Before you can expect to understand how to automate the role of customer support, you should have a clear understanding for the job. This means sitting with support teams, BPO visits, reviewing tickets teams have actually worked on, and even resolving some yourself. I think having this viewpoint is critical not only for shaping supportive products but also for your worldview on the true needs of support roles.
In fact, at Assembled, everyone from our interns to our CEO has taken turns on support ticket rotations. We have been in the queue, in the escalation channel, and on the hook for outcomes. After almost a decade in this space, you learn there are very few silver bullets.
That belief comes from our time at Stripe, where the founding team got its start. Before we even had a formal support team, everyone from engineers to operations took turns in the queue. The Collison brothers would host support rotations at their apartment in San Francisco, where you’d fix customer issues side by side with the founders. Those moments made it clear how much support shapes a company’s culture, and how hard it is to preserve that same level of care as you grow.
The future of support roles
The conversation around AI in customer support tends to spark fear about job losses, but that misses the bigger picture. The support agents who thrive in the AI era won’t be the ones doing password resets. They’ll be the ones handling the complex, emotionally charged moments where human judgment matters most. This means we need to invest in upskilling our teams, not replacing them.
Based on our work with customers, we’re seeing AI reconfigure support work rather than eliminate it. New roles are emerging: Knowledge Curation Specialists who maintain the knowledge bases AI relies on. AI Deployment Strategists who manage technical setup and integration. Process Mapping Experts who document how products actually work. Quality teams are shifting from manually reviewing individual calls to system-level oversight, focused on improving model performance rather than checking tickets one by one.
Traditional support roles are being elevated too. Customer service reps are becoming account managers and support engineers, getting deeper into critical thinking tasks while AI handles the routine work that used to consume their days. The pattern is clear: support professionals need to be comfortable with changing scope and ambiguity, combining technical AI fluency with domain expertise and human judgment.
The best support organizations are teaching their agents to work alongside AI, not compete with it. When agents see AI handling the work that used to burn them out, they get to spend their days solving real problems and building genuine connections with customers, which is why most people got into support in the first place.
Training the next generation of entry-level support teams in the AI era
Getting there requires rethinking how we develop talent from the ground up. If AI is handling the straightforward tickets, we can’t rely on the old model of “start with password resets, work your way up over two years.” New agents need to be ready for complexity and ambiguity from day one.
That means building onboarding programs that focus on judgment, empathy, and critical thinking, not just product knowledge. New hires need to learn how to effectively prompt AI systems, when to trust outputs versus when to verify, and how to catch plausible but incorrect answers. They need deep enough product expertise to know when exceptions apply, because AI can provide quick answers but can’t always judge nuance.
Organizations should create environments where AI learning happens through practice. Some companies run weekly sessions with protected time for exploration and demos. Others make AI fluency part of new hire training itself, using AI agents to answer onboarding questions so employees learn the tool while learning the job. The key is making experimentation safe, starting with low-risk tasks that build confidence.
And it means rethinking career paths. If the repetitive rungs of the ladder disappear, we need new ways for people to build expertise and advance. Skills-based development matters more than job titles now, with organizations training for specific functions and hybrid capabilities rather than traditional role definitions.
The right incentives
All of this brings us back to the original question: what incentives actually create great customer experiences? Not “paid on commission” for closing tickets, but designing systems where speed, quality, and trust all matter.
Interacting with brands should feel easy and human. That means:
- Automate the obvious. Resolve the majority of issues quickly and cleanly.
- Know when a human should intervene. Spot context, urgency, and who you are and then choose bot or human accordingly.
- Measure what matters. Not “bot first,” but “best outcome with best experience,” including time-to-relief, quality of handoff, and downstream retention.
Unlike Bret’s thesis, the goal of AI shouldn’t be full automation for automation’s sake. It’s choosing the right point on the curve between cost, speed and customer trust. Get that balance wrong, and you optimize for closed tickets while losing customer loyalty. Get it right, and AI becomes what it should be: a way to give both customers and agents a better experience.