Start here: crawl, walk, run, fly
The number one mistake people make is jumping straight to the flying stage. They see someone on LinkedIn who has built an automated pipeline that generates 50 Instagram ads a day across six platforms and they think that's where they need to be. It isn't. Not yet.
The crawl, walk, run, fly framework is about building real understanding before you build anything fancy. Here's how it breaks down:
This is where you start using products that have AI baked in, without having to touch ChatGPT or write a single prompt. A good example is call transcription tools like Granola. You install it, it runs in the background on your client calls, and it captures everything. You get useful transcripts without having to think about prompts or models. It's AI adoption with training wheels, and that's exactly the point.
This is where you start actually using Claude, ChatGPT, or Gemini directly. You're prompting, getting results back, learning what these tools are good at and where they fall flat. It takes time. That's fine. This is how you build what Keanan calls AI instinct, the ability to feel your way around a model rather than following a script.
Run: making it personal
This is where things get genuinely useful. Both Claude and ChatGPT let you set up persistent project spaces where you can load in context about a specific client, their history, their goals, how they communicate. So when you open Claude to work on a client project, it already knows who you're talking about. No catching up. No re-explaining everything from scratch.
If you've got client calls being transcribed from your crawl stage, you can feed those transcripts into a client project and ask things like: what upsell opportunities am I missing? Are there problems they flagged that we never followed up on?
This is the CMO agent stage. Keanan built himself an AI agent with access to his email subscribers, Google Search Console, and his content plan. Every day it checks in, tells him where he's at against his goals, and prompts him on what to publish next. He spent a couple of years getting to the point where building something like that made sense.
The takeaway isn't that you need to build an agent right now. It's that the progression exists, and you'll get there faster if you start at the beginning.
The internal case for AI: running a tighter agency
Before you start pitching AI consulting services to your clients, use it to make your own operation sharper. Here's where it delivers immediately.
Recording client calls and turning them into action
Most agencies record calls. Very few do anything useful with the recordings. AI changes that. When you've got six months of call transcripts from a client loaded into a project, you've got a searchable record of everything they've ever told you. Problems, goals, half-formed ideas, things they were excited about and then never mentioned again.
That's not just useful for client management. It's a pipeline for work you could be doing for them.
Context switching less, delivering more
One of the real costs of running an agency is the time it takes to re-orient yourself when you switch between clients. If a client emails you about a bug on a site you haven't touched in a week, the ramp-up alone takes time. When you've got an AI project set up for that client with all the relevant context, you're not starting from zero every time.
The external case: becoming your clients' AI guide
Here's a positioning opportunity that not many agencies have fully clocked yet. Your clients are starting to ask about AI. They've heard about it. They don't know where to start. You do.
Just like you became their trusted partner on WordPress, SEO, or email setup, you can become the person they turn to for AI. The window to establish that position is open right now. It won't stay open forever.
Keanan's suggestion: start with where your clients are right now. Most of them are still at the crawl or walk stage. Help them get ChatGPT or Claude set up properly. Show them what a good prompt looks like versus a bad one. Explain the difference between using a model on its free tier versus a paid plan and why it matters. That's genuinely useful advisory work, and it deepens the relationship.
The agencies who are figuring this out for themselves first will be the ones able to walk clients through it later. The experience translates directly.
WordPress and AI: where it adds value
This is the part most AI content skips over, because most AI content isn't written by people who've used these tools on WordPress sites. Here's what Keanan uses them for in practice.
Debugging plugin conflicts and errors
When something breaks on a WordPress site, the old common troubleshooting approach was: search the forums, read six threads, try things, and hope for the best. The new approach: screenshot your plugins list, paste in your error logs, describe what's happening, and let a coding-focused AI model work through it.
Keanan walked through a recent example: a plugin was causing a JavaScript conflict with a theme. What could have been hours of back-and-forth on a Zoom call was resolved in about 30 minutes. The model identified a jQuery versioning conflict and gave a specific fix. The site went live that night.
It won't always be that clean. But it's consistently faster than starting from scratch.
Making sense of PageSpeed reports
Google PageSpeed Insights gives you recommendations, but translating those into actual fixes on a WordPress site isn't always obvious, especially if you're not deeply technical. AI models are good at this: paste in the report, ask for step-by-step guidance on each issue, and get back a plain-language explanation of what to do and why.
Whether that's configuring a caching plugin, adjusting how fonts load, or writing a few lines of custom code to drop into the site, it dramatically shortens the gap between diagnosis and fix.
Coding assistance
AI models are getting genuinely good at writing code, and this is only going in one direction. Keanan uses Claude Code for coding work that would previously have taken him much longer, including projects that had been sitting in the backlog for months because the effort-to-payoff ratio never made sense. Now it does.
For agencies: try assigning one well-scoped ticket per sprint to an AI coding tool. Give it clear context, describe the expected behaviour, and work with it iteratively. The first time will be slower than just doing it yourself. By the fifth time, you'll have a much better sense of where it genuinely saves time.
Worth noting: this space moves fast. What follows is accurate as of early 2026, but check back regularly.
Claude (claude.ai) for most AI work, writing, analysis, reasoning, client project context
Claude Code for AI-assisted coding on your local machine
ChatGPT and Gemini as alternatives when you want a second opinion or hit a dead end
Ideogram for image generation, particularly useful for blog hero images and mood boards
Google Veo for generating short-form video from images
Suno for AI-generated audio, useful if you're producing video content
Granola or Fathom for call transcription at the crawl stage
The key thing about tools: don't expect them to do everything or always do it perfectly. The skill is knowing which tool to reach for, when to trust the output, and when to push back.
Where to start
If you haven't started yet, start with the calls. Pick a transcription tool, run it on your next few client meetings, and see what you've got. Then load one client's transcripts into a Claude project and ask it something useful about that client's business.
That's the crawl. The rest follows from there.
The agencies that will be well-positioned to offer AI consulting to their clients in a year's time are the ones building real experience with these tools right now, not the ones who watched the most YouTube videos about it.
Running a WordPress agency? hosting.com gives you the infrastructure to support your clients at scale, from managed WordPress hosting to dedicated resources that grow with your agency. Find out more at hosting.com.
