Most companies approach AI prompting like programming. They think there's an empathy button to press, a creativity switch to flip. When it doesn't work, they assume they need better SOPs first.

But prompting AI isn't programming—it's coaching. And just like coaching humans, it's iterative. The difference? AI has a faster feedback loop and actually remembers what you taught it yesterday.

This changes everything about deploying AI in customer support. You don't need perfect documentation before you start. You don't need to "get your house in order" the way you did before offshoring. You can iterate your way to excellence.

Here's what we’ll cover:

• Why you need to consider prompting as coaching not programming

• Realistic business expectations around AI in support

• The real opportunity that AI opens up

The edge case problem isn't new

It's human nature to assume our issue is unique and warrants a custom response from the company we are dealing with. This has been the scaling challenge in customer support for decades, long before AI entered the picture.

The customer service industry created standard operating procedures (SOPs) to handle scale, but it has always struggled with perceived, or actual, edge cases. Anyone who has ever demanded a refund from a junior support rep knows the frustration of rigid SOP-following. Bad support existed before AI—it was just delivered by under-resourced humans copy-pasting responses.

The mismatch between a brand's rigid procedures and a customer's messy, real-world needs has always been the core problem. AI didn't create this. But it does change how we solve it.

Prompting is coaching, not programming

This is where many companies get it wrong. They approach AI prompting like it's programming—like there's a button for empathy they need to press.

Prompting is actually coaching. Just like coaching humans, you give feedback and iterate based on results. The difference is that AI has a faster feedback loop and doesn't forget.

You can tell AI to "be more empathetic" the same way you'd coach a human. If it's too empathetic or not empathetic enough next time, you give more feedback. You iterate. The AI won't forget your coaching or need to be retrained after a month.

This is fundamentally different from software, where you're programming specific behaviors. With AI, you're developing capability through iteration.

You don't need perfect SOPs to start

Remember the offshore support era? The advice then was: "Get your house in order before outsourcing."

That was necessary because feedback loops were too long. Agents forgot training. Managers couldn't monitor everyone. Iteration took months.

But with AI, you can actually move forward without perfect SOPs. The entire process of creating and refining procedures has been transformed:

• Draft SOPs and deploy to your AI agent.

• Get immediate feedback from real interactions.

• Refine the SOP based on what you learn.

• Repeat.

This create-deploy-learn-iterate cycle was nearly impossible with human agents. The feedback loops were just too long with legacy structures.

Table stakes vs. actual value

Just ingesting SOPs and answering FAQs is increasingly table stakes. Every vendor can do this now.

Real value comes from AI that can be coached to handle nuance and complexity—AI that can partner with customers to find creative solutions, not just follow decision trees.

The partner matters here. You need good coaches on the vendor side who understand this iterative approach. Our best customers understand they're hiring a software agent that needs coaching, not a piece of software that just runs.

What this means for buyers

Most companies walk into AI deployments with expectations shaped by decades of traditional software. Here's what actually matters versus what doesn't.

The real opportunity

For many companies we work with, adopting AI provides a useful moment to examine whether their processes are actually best for customers. They reconsider old assumptions about what was possible: "We put that procedure in place because we couldn't offer 24/7 coverage" or "This is how we deal with seasonal spikes without increasing headcount."

The AI products you evaluate should be judged by their ability to ingest and learn your existing SOPs. That's the baseline at this stage. You should now be demanding that the products you evaluate actually make your existing support better than it was.

Customers deserve better than a low-cost replacement. They deserve personal service at scale, AI or not.

Human configurability has always been preferable because humans could be prompted in nuanced ways. That's no longer an advantage exclusive to humans. We can now coach AI to be creative in certain situations—the trick is telling it when to be creative, when to follow SOPs, and what combination to use.

That's always been the trick with human agents, too. With AI we're just faster at teaching.

Interest in how Lorikeet can help scale your best customer interactions? Book a demo.