In Conversation with Mark Szymanski

The Client Who Replaced Him with Claude

The Origin of the Story The conversation's central anecdote began with a video Mark published on his YouTube channel with a title along the lines of "My client just replaced me with Claude." He was transparent that there was some deliberate packaging involved in the thumbnail and framing, as is standard for YouTube, but he insisted that the underlying story was genuine and that he made a conscious effort not to be either a hype merchant or a doom prophet.

The video received approximately 30,000 views, which Mark noted is significant for his channel. It also attracted around 200 comments covering a wide range of reactions.

The Story Behind the Video Mark had a client who approached him roughly a year before the video was made needing a website for a new business. When he quoted a typical professional rate in the five to ten thousand dollar range, the client found that too expensive and turned to Fiverr instead, getting a cheap site that, by the client's own admission, was not good. The client came back to Mark, who lowered the scope and price to make it work. The client was happy and the business grew.

A year later, that same client stumbled into Claude and, on a whim, asked whether it could build them a website. Claude produced something functional in front of them. Mark acknowledges that professionals understand how much more goes into a real production website, but the client had something working on their screen and the dynamic had changed irreversibly.

Mark found himself in an ethical bind. He could not in good conscience be a public advocate for AI productivity and then privately discourage his own client from using the same tools.

That felt dishonest. So the video became his way of processing that tension openly and communicating to his audience that this scenario was going to become common.

The Comments and the Meta-Experiment Rather than summarising the comments from memory, Mark fed all 200 of them back into Claude and asked it to analyse them. He shared the results on a live stream. The broad buckets that emerged were: Developer cope: A substantial portion of responses came from people who felt their skills could never be replicated by AI and were resistant to accepting that framing. Mark is empathetic to this position even while recognising it as a form of denial.

Doomsayers: A segment of viewers responded with genuine alarm, treating the situation as catastrophic. Mark does not align with this view either.

Clients: Because the video reached beyond the usual developer and agency audience, actual clients watched it. Some commented to say they had done exactly the same thing, finding AI so useful that they had moved on from their previous service provider, without any particular hostility. They were simply solving a problem.

The Unexpected Lead Perhaps the most striking outcome of the video was that it generated an inbound lead from a reasonably sized home-services business. The prospect had watched the video, recognised in it a description of their own pain, and concluded that Mark's depth of expertise made him someone worth hiring. They wanted a go-to person who could handle their marketing and operations at a higher level. Mark and the client were still in early conversations at the time of the session, but the point stood: radical transparency about the state of the industry can attract better clients, not repel them.

The Ethical Position: Neither Hype nor Doom

Mark returned several times to his guiding philosophy: he refuses to inhabit either extreme of the AI discourse. He is not going to tell clients they cannot or should not use AI tools when he is simultaneously advocating for those same tools in public. He is not going to catastrophise. His message to agency owners who are resistant to engaging with AI is straightforward: you do not have to burn everything down or sell out your values. You can make incremental transitions, become more effective, and continue serving clients well. But other people in your market will not stay still, so the risk of inaction compounds over time.

He explicitly pushes back on the idea that this change is unique to the AI era. He frames it as the latest in a series of transitions: from GeoCities to Flash websites to WordPress to page builders to AI-assisted builds. Each wave brought new entrants, lowered certain barriers, and forced established practitioners to either find new layers of value or be displaced. The current wave is simply faster and more far-reaching than previous ones.

You can make incremental transitions, become more effective, and continue serving clients well.

Workflow: How Mark Builds Client Sites Now

The Shift in Skill Set Mark's background was WordPress with a progression through Elementor, then Bricks, then Etch. He had never built web applications, never worked with frameworks like Next.js or Astro, never touched Supabase, and had no experience with headless CMS setups. Over the past several months, using AI as a collaborator and learning resource, he has acquired working knowledge across all of these. He is careful to note that "working knowledge" does not mean he reads and writes every line of syntax manually, but he understands how the pieces fit together and can direct AI tools to build things he could never have built before.

His key framing: AI has made it possible to be more things at once. You can have design sensibility, business understanding, and development capability without being an expert in all three. The tool handles execution; the practitioner provides context, direction, and judgement.

The Landscaping Client: A Proof of Concept The first real experiment in this new workflow involved a landscaping client whose site Mark had originally built in Elementor, long enough ago that he dreaded going back into it for updates. He decided to use this client as a guinea pig for a headless approach.

Working in WP Studio (or a comparable local development environment), he brought the site down, opened VS Code with Claude Code, and gave Claude the instruction to examine the existing HTML and CSS output from the Elementor site and convert it into an Astro project.

The goal was to create a headless WordPress setup where WordPress handled the data layer but the front end was rebuilt in Astro, making future updates far easier to manage by describing changes in plain language rather than navigating an old page builder.

Claude completed the conversion in roughly ten minutes. Mark describes this as a significant unlock because it means legacy client sites, even those built years ago with whatever tools were current at the time, can be migrated to a more maintainable architecture relatively quickly.

He acknowledges that the harder problem is the sales pitch: how do you tell a client whose website is not broken that you want to change it? He is still working through the framing, though he notes that "we can move faster and service you better" is at least the beginning of an answer.

The Insurance Company: A Full End-to-End Build The most detailed build Mark discussed involved an insurance company. He published a 10- hour raw build video on his YouTube channel documenting the entire process from start to finish.

The workflow began before any code was written. Mark recorded all client conversations (with appropriate notice to the client) and fed those transcripts into a Claude project alongside other context. From this base of information, he and Claude built the site together.

The stack was Astro on the front end and Sanity as the CMS on the back end. It was his first time using Sanity and he reserved final judgement on it, but the build went well and the client was happy.

The more revealing insight came from what happened after launch. The insurance client wanted to integrate their quoting system and an intake SaaS product. In a traditional WordPress environment, this would have meant hunting for compatible plugins, hoping the plugin developer had exposed the right integration points, and accepting the limitations of whatever the plugin ecosystem offered. In the new environment, Mark was able to reach directly for API keys and authentication tokens and wire up the integrations himself with Claude's help. He describes this as a form of liberation: you are no longer constrained by what someone else chose to expose through a plugin.

He uses forms as an illustrative example. A custom form is ultimately HTML and perhaps some JavaScript. The hard part, sending submissions somewhere useful like a Slack channel or a Google Chat, is now a matter of grabbing an API token and wiring it in. Plugin ecosystems exist for good reasons, but they also impose limits that custom builds can now bypass.

The Orchestrator Mindset Mark's preferred metaphor for how to use AI in development is the orchestrator. Rather than typing direct commands and accepting whatever the AI produces, he advocates treating the interaction as a collaborative dialogue. Don't say "add a database." Instead say "I'd like to add a database structure. Let's explore what that would look like." Ask questions. Read the responses carefully. Understand what is being proposed before accepting it.

He combines this with voice dictation tools, particularly mentioning Whisperflow, as a way of making the dialogue more natural. He finds that thinking out loud and having the AI respond conversationally produces better results than typed commands, and that the experience approximates having a junior developer alongside you who has strong technical knowledge but needs your contextual direction to do good work.

The Recurring Revenue Problem

One of the most honest and unresolved sections of the conversation concerns how agencies sustain recurring revenue as the economics of running client websites change.

The Cost Structure Has Changed Static sites and headless builds are cheap to host. Plugin licence fees disappear. The heavy overhead of a typical managed WordPress stack, hosting, page builder licences, plugin costs, and the occasional emergency call when something breaks, is largely gone. This is good for the client and for operational simplicity, but it removes the basis on which many agencies have built their recurring revenue through care plans.

A Live Example At the time of the session, Mark was literally in the middle of transitioning a client from a QuickBooks billing arrangement to Stripe. He reached out to a client who had been paying roughly $150 a month for a care plan on a WordPress site that was more of a passion project than a core business asset. He presented the client with three options: continue as-is, shut it down, or migrate to a lighter static setup that would cost significantly less to maintain. The client chose the lighter route.

Mark is candid that he would rather not lose that recurring revenue, but he also recognises he cannot justify charging for overhead that no longer exists.

Where the Value Shifts To His working answer to the recurring revenue question is to go deeper into client operations rather than sideways into more website maintenance. He spends much more time now asking clients what problems exist in their business beyond the website. Because his skill set has expanded, he can offer things that would previously have been outside his scope: lightweight internal tools, simple web apps, operational integrations.

He gives the example of building his father a simple HTML and JavaScript inventory management app to replace a paper-based system his father's employer had been using. The father uses it every day and it has meaningfully changed how he works. That kind of outcome, solving a real operational problem for a non-technical person, is now within reach for any practitioner willing to learn.

The financial model for these services, build fee plus a maintenance retainer, is broadly similar to the website model, but the value delivered is higher and the dependency on cheap commodity services is lower.

A Caution About Complexity Mark also notes that as you move toward custom builds with AI integrations, SMTP services, rate-limited API calls, and other moving parts, you have to think more carefully about costs.

Free tiers run out. If you are building the same kind of solution for fifty clients, the cumulative cost of email sending services, API calls, and other infrastructure adds up. The days of "plugin costs this, hosting costs this, page builder costs this" are replaced by a more dynamic cost model that requires more active management. You are, in effect, thinking more like a developer and a business owner simultaneously.

You are, in effect, thinking more like a developer and a business owner simultaneously.

Content, Community, and Business Development

YouTube as a Trust-Building Channel Mark's YouTube channel serves multiple functions in his business model. It documents his learning in real time, which he sees as a form of integrity: he is not presenting himself as someone who has figured everything out, but as someone working through the same challenges as his audience. The 10-hour insurance company build video is an example of this approach, offering raw, unedited footage of a real client project from start to finish.

The channel also generates inbound leads, as the viral video about losing a client to Claude demonstrated. His view is that sharing genuine expertise and authentic experience, including uncomfortable experiences, builds the kind of trust that referral networks alone cannot create.

He explicitly recommends that freelancers and agency owners who have relied solely on referrals consider creating some form of public-facing content, not as a marketing exercise but as a demonstration of real relationship and human value. In a world where clients can increasingly self-serve on commodity tasks, the practitioner who is visibly knowledgeable and genuinely helpful has a significant advantage.

The Modern Builder Society Mark's community for AI-augmented web and app builders is called the Modern Builder Society. He describes it as a group of people navigating the same transition together. Details and sign-up information are available through his website at mjs.bio.

On the Broader Industry Transition

A New Wave of Entrants The conversation's host, Vito, draws a parallel to previous moments of industry expansion.

When Elementor arrived, it brought several million new participants into the web-building ecosystem. AI tools are likely to do the same, and possibly at a larger scale. Mark agrees that this is the likely trajectory.

Vito frames it as a shift in interface rather than a shift in fundamental purpose: from hand-coding to drag-and-drop to natural language to voice. The output, websites that generate leads and traffic for clients, remains broadly the same. The interface through which it is assembled changes.

Mark arrived at WordPress in 2018 at a point of relative stability, after the earlier chaos of the GeoCities-to-Flash-to-WordPress transitions. For him, the current moment is the first time he has experienced this kind of rupture personally. Vito notes, with some warmth, that this is what transitions feel like from the inside, and that they tend to happen again.

The Consultative vs. Insurance Model Vito introduces the framing that agencies can no longer position themselves as "the insurance layer," collecting care-plan fees and doing reactive maintenance. The model needs to become proactive and consultative. Mark agrees fully. He notes that the old advice of "sell them SEO or PPC on top of the care plan" was always available but now the pressure to actually do it is more acute, because the commodity maintenance work is eroding faster than anyone expected.

His answer is to become more interested in the full operational picture of each client's business, to ask better questions, and to use the expanded skill set that AI provides to offer genuinely new value rather than repackaged versions of what was already there.

About the speaker Mark Szymanski Mark Szymanski

Mark Szymanski brings a long career in the WordPress and agency space, speaking candidly about the industry's direction in conversation format.