AI-First Operations: A Practical Guide

This post offers a pragmatic approach to integrating AI and Large Language Models (LLMs) into operations at software companies.

I offer a structured thought process for identifying opportunities to leverage AI, while cautioning against overly complicated solutions.

Read on for the full guide 👇

The AI-First Operator's Question

AI-first operators ask one question for every operational bottleneck: How can we use AI/LLMs to solve this?

The boring truth is that AI is just one tool to reach for, among many. Sometimes it's the right tool, sometimes it's not.

Do not blindly try to apply AI to every problem, it's just "operations" after all.

Today, I think AI is most useful to operators in these 6 situations:

1. Thought partnership - stress test and sharpen your thinking and work
2. Pattern recognition at scale - extract insights from thousands of customer calls
3. Content transformation - translating technical docs into sales materials
4. Decision augmentation - prioritize bug fixes by impact
5. Knowledge synthesis - answer questions across multiple documents
6. Repetitive cognitive work - first-draft responses or initial analysis

These are each use-cases where AI can augment human capability. 

Though, it still doesn't replace judgment or empathy, and I don't expect that to change anytime soon.

The 2x2: Speed and Scope

When it's clear that AI is the right tool, you need to decide which type of AI solution to implement.

This breaks down along two dimensions:

• Speed of impact: How quickly do you need to see results? (Fast vs Slow)
• Scope of impact: How big of an impact is needed? (Individual vs Organizational)

This creates four distinct approaches.

Personal AI (Fast + Individual)

Start experimenting with

• ChatGPT/Claude/Gemini chat
• MCPs for interconnectivity
• Projects for repeat use cases
• Skills, GPTs, agents, artifacts, whatever the labs cook up next.

No budget approval, no meetings, just immediate feedback and leverage for you.

Example: A customer support agent is struggling to quickly respond to a complex customer inquiry. Instead of spending hours researching the issue, they could use ChatGPT to summarize relevant documentation and generate a draft response.

Native AI Features (Fast + Organizational)

Your existing tools probably already have AI features. Hubspot Breeze, Jam.dev, Zapier agent, Intercom Fin, etc...

Flip the switch and the whole team benefits instantly.

No, don't even think about vibe coding. Your needs are probably not that unique and you can bet that the market has, or will shortly create, a solution that works for you.

Example: You are a customer success manager and just got passed a new account. Instead of asking your sales colleague for a summary of the deal, you use the Hubspot Breeze AI agent to create the summary for you.

New Software (Slow + Organizational)

There are two cases where a new piece of software makes sense

• The core system is outdated or missing
• The functionality is truly novel

This can be a painful quadrant due to change management, budget approvals, and hoping the software provider raises their next round.

• No CRM? Consider Attio or Lightfield before Hubspot.
• No support platform? Maybe Pylon or Plain is a better decision than Zendesk
• Need to clip video for socials? Opus clip is a great place to start

Example: You have so much inbound demand that you can't provide demo's to everyone. You want an AI agent to give a good-enough demo to people you can't get to so they can get their questions answered and see the software.

The Dead Zone (Slow + Individual)

Slow results limited to just you?

Uh, just avoid this.

Conclusion and Key Principles 

The AI-first approach means systematically answering "How can we use AI/LLMs to solve this?" for operational challenges. 

Use the matrix to guide your approach.

Most solutions live in the top half, expensive transformations are rarely the answer, and keep in mind these principles:

• Don't fall for the hype: These tools are awesome but they are not panaceas, and many of them are not quite ready for prime time.
• Start Simple: Begin by exploring personal AI tools and native AI features. Avoid implementing complex AI solutions when simpler alternatives exist.
• Iterate: Continuously monitor the performance of AI-powered solutions and make adjustments as needed.

 Start fast, think small, prove value.