AI Reality Check: What AI Can Do

Introduction:

Artificial Intelligence is coming whether it’s welcome or not. What can artificial intelligence really do for insurance agencies selling benefits, senior, and life policies? That’s a discussion NextAgency Co-Founder Alan Katz had with ChatGPT and Claude, two of the most popular AIs available to consumers and small businesses.

The Discussion:

Alan to ChatGPT:

We’ve all been told AI is changing the world. It’s the biggest innovation since the Internet. Do you agree? Focusing on consumer-facing platforms like you and Claude, what can they do today to make life better for people?

ChatGPT’s Response:

Yes, AI is a big deal—but how big depends on what we’re measuring. Here’s what today’s AI actually does well for everyday users:

• Writing and communication: It helps people draft emails, revise resumes, write social posts, or even apologize more clearly. This is where tools like ChatGPT, Claude, and Copilot shine.

• Summarizing and explaining: Need a contract boiled down? A news article simplified? A medical study explained in plain English? AI’s great at digesting and rephrasing text—fast.

• Coding and spreadsheets: Copilot, Claude, and ChatGPT can troubleshoot code, write formulas, and assist with low-level data tasks.

• Search with context: Perplexity and Gemini are moving search closer to “answers” instead of just “links,” especially when users need synthesis or opinions.

• Scheduling and productivity: Copilot, Apple Intelligence, and Google’s integrations are improving calendar help, task reminders, and doc drafting—but they’re still clunky across apps.

That’s real utility. But it’s narrow and task-based. These tools are assistants—not replacements—for human judgment, creativity, or emotional nuance. Think of them as well-informed, digital Swiss Army knives: compact, handy, and versatile. For the right job, they’re a terrific tool.

Claude’s Response: 

ChatGPT’s list captures the basics well. What’s worth adding is how far the tooling has evolved beyond chat interfaces. Autonomous AI agents — tools like Claude Code and CoWork — can now execute multi-step tasks, not just respond to prompts. An insurance agency could, for example, instruct an agent to pull renewal dates from an AMS, draft outreach emails for each client, and flag any accounts with coverage gaps — all without step-by-step human direction. That’s a different category of capability than drafting a single email.

The general capabilities ChatGPT describes — writing, summarizing, coding assistance — are the foundation. Agents build on that foundation to handle workflows.

That said, ChatGPT is right about the limitations. AI excels at pattern recognition and text manipulation, but it can’t replace judgment about whether a client actually needs additional coverage, or read the emotional subtext when someone says they’re “fine” during a benefits review.

The real transformation isn’t that AI does everything — it’s that it handles routine cognitive tasks well enough that professionals can focus on relationships, strategy, and complex decision-making. 

Alan to Claude:

You jumped ahead from high-level consumer use to how AI can be used by insurance agencies. That’s fine. Please provide some examples of what an agency with a (in your case) Claude subscription can do.

Claude’s Response:

With a Claude subscription, an insurance agency can handle several routine tasks more efficiently:

• Document analysis: Upload carrier communications, policy changes, or compliance updates and ask Claude to summarize key points or flag important deadlines. Instead of reading through dense carrier bulletins, agents get the highlights in plain English.

• Client communication: Draft clear, empathetic emails explaining complex topics like why a claim was denied or how a new benefit works. Claude can maintain a professional tone while avoiding jargon that confuses clients.

• Proposal preparation: Feed Claude client information and coverage requirements, then have it draft initial proposal outlines or comparison summaries. It won’t replace the agent’s expertise in selecting appropriate coverage, but it speeds up the documentation process.

• Training material creation: Convert technical product information into digestible training content for new agents or create quick reference guides for complex products.

• Meeting preparation: Upload client files and have Claude generate discussion points for renewal meetings or suggest cross-selling opportunities based on current coverage gaps.

The key limitation: Claude can process and reorganize information brilliantly, but it can’t make coverage recommendations, assess client risk, or handle nuanced compliance questions. It’s excellent at the “paperwork” side of agency work – the reading, writing, and organizing that takes time away from actually advising clients.

Alan to ChatGPT:

Here’s what Claude had to say about what insurance agencies can do with a Claude subscription. I assume they can do the same with ChatGPT, right? Do you agree with Claude’s list? Anything to add?

ChatGPT’s Response:

Yes, with a ChatGPT subscription—especially GPT-4o or enterprise-level access—agencies can do almost everything Claude describes. The capabilities are broadly similar across platforms when it comes to document summarization, drafting emails, and creating internal content. I’d say Claude’s list is accurate and practical.

Here are two additions worth considering from the ChatGPT side:

• Workflow automation (via API or integrations): ChatGPT can power backend automations—like sending templated follow-up emails after client meetings or flagging policies nearing renewal—if integrated through platforms like Zapier, Make, or custom API calls. This goes beyond just content generation into actual process streamlining.

• Voice and multimodal support: ChatGPT now supports voice interactions and can interpret images, PDFs, and spreadsheets more fluidly in consumer tools. That’s handy for things like reviewing census forms or summarizing plan grids without needing to retype them.

That said, Claude’s point about limitations stands: no consumer AI can safely replace professional judgment. These tools don’t understand insurance the way agents do—they just handle the parts agents shouldn’t have to waste hours on.

Alan to ChatGPT:

You and Claude have both cautioned against assuming AI can safely replace professional judgement. That’s now. What about in the next year or three?

ChatGPT’s Response: 

Over the next 1–3 years, AI will absolutely get better at pattern recognition and generating insurance-specific language—but it’s unlikely to reach the level of safe, autonomous decision-making for agents.

Why?

• Lack of domain understanding: AI models still don’t understand insurance. They generate responses by predicting word patterns—not by grasping concepts like insurable interest, state-specific compliance, or client suitability. No amount of training makes that true understanding click.

• High stakes and low tolerance for error: Insurance involves financial risk, legal exposure, and client trust. A hallucinated plan feature or overlooked coverage detail isn’t a typo—it’s a lawsuit waiting to happen. AI isn’t ready to own that liability.

• Opaque reasoning: Even as models improve, their internal logic remains largely inscrutable. You won’t know why the AI recommended a certain plan—or what assumptions it made—unless you retrace every input. That’s not practical for advisors with a license on the line.

That said, we’ll likely see safer AI assistants emerge: tools that flag potential gaps, suggest next steps, or pre-fill documents for agents to approve. Think “advisor co-pilot,” not “autopilot.” If agents stay in the loop, AI will be a force multiplier. But alone? Still risky.

Alan to Claude:

I noted that you and ChatGPT had both cautioned against assuming AI can safely replace professional judgement. But that’s now. What about in the next year or three? Here’s ChatGPT’s response. Do you agree?

Claude’s Response:

I’d agree with ChatGPT that professional judgment will remain essential, but I think they’re being too conservative about the timeline for practical improvements.

Within the next year, I expect we’ll see AI that’s much better at handling multi-step workflows. Instead of just drafting emails, AI might pull client data from your AMS, check recent interactions, and suggest personalized outreach based on policy anniversaries or life events. That’s not replacing judgment – it’s giving agents much better information to make decisions with.

The bigger change will be in compliance assistance. AI could potentially flag when a recommendation conflicts with state regulations or carrier guidelines before the agent submits it. Not making the decision, but catching obvious mistakes early.

Where I disagree with ChatGPT is the assumption that “professional judgment” is this monolithic thing. Some judgment calls are actually pattern recognition at scale – like identifying which small business prospects are most likely to need voluntary benefits based on industry and employee demographics. AI could get quite good at those analytical tasks.

But the core relationship work – reading between the lines when a client says they’re “thinking about it,” knowing when to push and when to back off, understanding family dynamics that affect life insurance decisions – that’s genuinely human territory.

The question isn’t whether AI will replace agents, but whether agents using AI will dramatically outperform those who don’t. I think that gap will widen quickly.

Alan to ChatGPT:

Here’s what Claude said about your answer. Does it have a point?

ChatGPT’s Response:

Claude absolutely has a point—especially about nuance in “professional judgment.” I was drawing a harder line to highlight the risks, but Claude is right that some tasks agents consider judgment-based are actually pattern recognition problems—and those are squarely in AI’s wheelhouse.

Their example of identifying good voluntary benefit prospects based on industry and demographics is spot on. That’s a strong use case for AI: lots of data, clear outcomes, and repeatable logic. I also agree that AI flagging compliance risks before submission is realistic if paired with good data hygiene. AI can be provided with current regulations and used to flag potential conflicts against an agency’s correspondence and submissions.

Where I’d still hold the line is in AI acting independently on that information. Even if the AI can say, “This looks like a mismatch for California small group rules,” it still takes a licensed human to interpret that warning in context. The danger is when tools feel more accurate than they are—and agents overtrust them.

So yes, I may have overstated the limits slightly, but only because the consequences of AI error in this space can be so serious. The real opportunity is exactly what Claude ends on: AI-augmented agents will pull ahead fast. The challenge is making sure they’re using it wisely.

 Conclusions:

Alan to AIs:

What you both seem to be saying is that AI is going to change the way insurance agencies do business in the next few years. Some of what you’ve said will be heartening to agents. Some of it will be terrifying. We’ll talk about the limits of AI’s abilities in the next chapter, but to wrap up this one, in 100 words or less, what should agents and their staff take away from this discussion so far?

Claude’s Summation:

AI is already useful for routine cognitive work – document analysis, drafting communications, and organizing information. That usefulness will expand rapidly. Agents who learn to use AI as a productivity tool will have a significant advantage over those who don’t.

The key insight: AI won’t replace agents, but agents using AI will outperform agents who aren’t. Start experimenting now with simple tasks like email drafting or document summarization. Focus on what AI does well—processing and organizing—while you concentrate on relationships, strategy and judgment. The window to gain this competitive edge is open, but it won’t stay open forever.

ChatGPT’s Summation:

AI is here to assist, not replace. It can streamline tasks, surface insights, and reduce busywork—giving agents more time for what matters: building relationships and advising clients. The agencies that benefit most won’t be the ones with the most tech, but the ones who use it thoughtfully—pairing AI’s speed with human judgment and empathy. Swiss Army knives can do a lot—they just can’t do it all.

Alan’s Summation:

AI is already changing the way agencies work. And it’s not going away. NextAgency has started adding AI features. Our first addition was adding AI to the text editor used to create descriptions in notes, tasks, and tickets and to help improve emails. More AI tools are in the pieline. Significantly, we’ll be implementing these tools cautiously, because we’re aware of the dangers AI introduces. We’ll explore these dangers in the next chapter.