Two years ago, customer support was consuming 20+ hours of my week. I knew I needed a change, but I didn't want full automation. I wanted something different: a minimal automation approach where I'd still review every single reply before it went out.
I started with a simple Raycast AI command and a 26-word prompt. The result? Today, I handle the same volume of customer support in just 5 hours per week while maintaining the human touch my customers expect.
Here's what happens each time a customer message comes in:
- I read the inquiry and form a rough answer in my head
- I select the customer's message and open my Raycast AI command
- I type a quick overview of what I want to say
- AI generates a polished reply that I can copy-paste to HubSpot or email
- I make minor adjustments when needed (rarely more than 5-10 seconds)
The whole process takes under 2 minutes per inquiry.
I'm still fully in control. My name appears as the reply author, and I quality-check every response. But AI handles the writing while I handle the thinking.
This isn't about replacing humans. It's about augmenting them.
The problem with full automation
There's a dangerous myth spreading among founders: that the future of customer support is full automation. "Set it and forget it," they promise. "Let AI handle everything."
I've seen this fail repeatedly.
When you fully automate customer support, you lose the human touch. Customers notice when they're talking to a bot. The responses feel generic, lack context, and miss the subtle cues that human support agents pick up.
Worse, founders often overcomplicate their approach. They spend hours crafting complex prompts with system instructions, tone specifications, personality descriptions, and elaborate guidelines for every scenario. They believe more complexity equals better results.
It doesn't.
The most effective AI customer support setup doesn't remove the human. It augments them. And the best part? My prompt is just one sentence long.
The minimal automation approach
My philosophy is simple: human-in-the-loop is a feature, not a bug.
Quality control matters because my name appears on every reply. If an AI hallucinates or says something I wouldn't say, that's on me. But I've discovered that keeping myself in the loop doesn't mean doing all the work. It means doing the right work.
I handle about 20 customer inquiries per week. Before AI, that consumed 20+ hours. Now? Five hours. Same volume, better quality, faster responses.
The sweet spot is this: speed plus authenticity.
Here's why this works for solo founders and contractors:
- You maintain complete control over what gets sent
- AI handles the writing while you handle the thinking
- You can address complex situations with nuance that fully automated bots miss
- You build genuine relationships with your customers
Setting up your workflow
The technical setup takes five minutes.
Creating the Raycast AI command
- Open the Raycast command palette
- Find the option to create AI command
- Name it "Support reply" and add the prompt
- Select Claude Haiku (for speed), leave creativity on medium
The 26-word prompt
Here's the entire prompt:
Write me a quick customer support reply for {selection} so I can copy and paste it later. Ask me questions if you need clarification on what to answer.
That's it.
No tone specifications. No personality guidelines. No complex system instructions. Just a simple instruction that trusts the AI to understand context.
My daily workflow
Here's exactly what I do for each customer inquiry:
- Read the inquiry: I read through the customer's message carefully
- Form my answer: I think through what I want to say in my head
- Select the message: I highlight the customer's original message
- Run the command: I trigger my Raycast AI command
- Add context: I type a quick overview of what I want to answer (2-3 sentences max)
- Review: AI generates a reply. I read through it and make minor adjustments
- Send: I copy-paste to HubSpot or email and hit send
The entire process takes under 2 minutes. Most inquiries, even complex ones, are handled this way.
Real-world example
Here's an anonymized example of a customer inquiry and my AI-assisted reply:
Customer:
Hi Oliver,
I fixed the problems to login, changed the email address and password. i think i finished the domain setups. i think only need to wait to finish the process in cloudflare.
Pls dont delet my site.
Best regards David
My AI-Assisted Reply:
Hi David,
Thanks for the update! Glad to hear you've fixed the login issues and updated your email and password. No worries, I won't delete your site. Just let me know once the Cloudflare process is complete, and I'll be here if you need any further help.
As a heads up, I've extended your trial by 14 days so you have more time to continue testing our platform.
— Oliver
What's interesting here? The AI captured warmth and professionalism that I, as a developer, rarely express naturally. It proactively offered help. It added the trial extension because the customer had login problems and their trial was about to expire without them actually testing the platform.
Would I have remembered to extend that trial manually? Probably not. That's the beauty of this approach: AI handles the details while I handle the strategy.
Why this works better
The warmth factor
Here's what surprised me: customers don't notice the difference.
In fact, my AI-assisted replies are often warmer than my natural responses. As a developer, my personality isn't naturally warm. I tend to be direct and technical. The AI brings a warmth and empathy that customers appreciate.
I'm noticing customers may actually prefer these responses over my originals. They're consistent, they're helpful, and they feel personal.
Speed and quality
20 hours per week down to 5 hours. That's a 75% reduction.
But here's what matters more: the quality hasn't suffered. In fact, it's improved. I'm catching more details, responding faster, and maintaining consistency across all my responses.
The power of simplicity
Why does my 26-word prompt work so well?
Because it trusts the AI to handle tone, style, and context. Most founders overcomplicate their prompts because they don't trust AI to get it right. But when you give AI clear, simple instructions, it often does better than when you try to micromanage every detail.
Complex prompts create confusion. Simple prompts create clarity.
Common Adjustments I Make
I review every AI-generated reply before sending. Most of the time, no changes are needed. When I do make adjustments, they fall into three categories:
1. Technical Accuracy
I know the platform inside and out. Sometimes the AI gets technical details wrong or misses nuances about how the system works. A quick fix takes 5 seconds.
2. Salutations
AI tends to include lengthy greetings and sign-offs. For a single message, that's fine. But when you have back-and-forth threads with a customer, those "Hi David" and "Best regards" at every message start to feel excessive.
I often trim or remove these in ongoing conversations.
3. Missing context
The AI doesn't know everything I know about a customer's situation. Sometimes I need to add details that aren't obvious from the message itself. A previous interaction, a special circumstance, or something I've already promised.
These adjustments take seconds. The bulk of the work, the actual writing, is handled by AI.
Tips and best practices
Here's what I've learned from using this approach:
- Keep your prompt simple. Complexity often backfires. My 26-word prompt works because it's clear and trusts the AI.
- Always review. Your name is on the reply. A 5-second quality check prevents mistakes.
- Use overviews to guide AI. When you type your quick overview, you're steering the AI in the right direction. Be specific about what you want to address.
- Claude Haiku is fast and sufficient. I chose it because it's fast. For customer support, you don't need the most advanced model. You need speed.
- Make minor adjustments when needed. If the voice doesn't feel right, change it. Don't be afraid to override the AI.
Common pitfalls to avoid
I've seen people make the same mistakes when trying to adopt this approach:
Don't fully automate. The temptation to remove yourself entirely is strong. Resist it. Your human oversight is what makes this work.
Don't over-specify tone and style. Let the AI handle the conversational aspects. Your job is to provide context and quality control.
Don't skip technical accuracy review. You know your product better than any AI. Trust your expertise.
Don't forget your brand voice. Review the AI's output through the lens of your brand. Make adjustments if something feels off.
The bigger picture
This approach applies beyond customer support.
You can use it for emails, documentation, marketing copy, any writing where you want speed without sacrificing quality or authenticity.
The future of work isn't AI replacing humans. It's AI augmenting humans.
Most people are stuck in binary thinking: either fully automate or fully manual. The sweet spot is somewhere in between. A partnership where humans provide the direction, strategy, and quality control, and AI handles the execution.
This is the future of productive work: human + AI, not human OR AI.
Conclusion
Start simple: one command, one prompt.
You're not replacing yourself. You're augmenting yourself. The goal isn't automation. It's freedom to focus on what matters.
For me, that freedom means working 5 hours per week while maintaining excellent customer support. For you, it might mean something different. But the approach is the same: keep the human in the loop, trust the AI to handle the writing, and maintain quality control at every step.
The best AI customer support isn't fully automated. It's intelligently assisted.
That's the future. And you can start today with a 26-word prompt.