For several months now, we have been coding all our WordPress sites with Claude Code, Anthropic's development agent. Here, without embellishment, is what it has changed for our clients: more accurate quotes, delivery times cut by three on certain projects, and more time devoted to strategic decisions. But also a few limitations we want to share honestly.
Before: the classic pace of a WordPress agency
To fully understand what has changed, you first need to understand how things were done before.
A new WordPress showcase site, at a typical agency pace, takes about six weeks from signing to launch. Two weeks for the mockup, ten days for development, ten days for content, five days for client adjustments, five days for testing and launch. If all goes well, and if the client meets their feedback deadlines.
A custom module for a CRM took about three to four weeks. A complete SEO overhaul on an existing site, two weeks. A WhatsApp Business integration with automation: a month for a basic workflow.
These timelines were industry standards, not outliers. All serious freelancers and agencies operated at roughly this pace.
The bet
After several months of testing on internal projects, we have made the following decision: to move all our projects, without exception, through Claude Code. Not one in two projects. Not just small sites. Everything.
It was a bet, because at the time, agencies talking about "coding with AI" were mostly in marketing, not production. Most used ChatGPT to generate a few snippets of code at a time, which they manually copied into their editor. It was slow, unreliable, and above all unverifiable.
Claude Code, on the other hand, works differently: it's a true agent that reads your project, understands the structure, writes code directly into the right files, runs tests, fixes its errors, and shows you exactly what it did. It's the difference between asking a consultant for advice and having a developer open your editor and work with you.
Six months later: what has concretely changed for our clients
Here are the concrete changes, measured on real construction sites carried out in recent months.
Complete WordPress showcase site: six weeks → ten days
The project that benefited most from the change is the medium-complexity WordPress showcase site. What took six weeks now takes ten days on average. The difference mainly comes from three stages:
- THE initial development drops from ten to three days thanks to assisted generation of the entire Elementor structure, dynamic templates, and plugin configuration.
- THE client adjustments drop from five to two days, because the modifications you request (changing a button color, moving a section, adding a field) are applied in minutes instead of hours.
- THE launch tests drop from five to two days, because the validation chain (backup, deployment, visual check, optimization) is now partially automated.
What doesn't change, and never will: the scoping phase with you at the start of the project, the creation of graphic mockups, and content delivery. On these aspects, AI speeds up nothing because they are moments where your input is essential.
Custom CRM module: three weeks → six days
Developing a custom module for Perfex (the CRM we use and install for our clients) follows the same logic. Where you used to have to manually write hundreds of lines of PHP to respect the CRM's architecture, Claude Code reads the existing structure, understands the conventions, and produces a compliant module. An expense management module with categorization, accounting exports, and bank synchronization that would have taken three weeks in November is now delivered in six days.
SEO overhaul: two weeks → five days
This is probably the most commercially useful gain. A complete SEO overhaul of an existing site — rewriting tags, optimizing load times, restructuring internal architecture, implementing Schema.org, checking compliance with the latest Google guidelines — goes from two weeks to five days. And the result is more robust, because the AI doesn't forget mechanical checks while focusing on strategic choices.
Three concrete questions that clients ask
How long does a complete WordPress showcase website now take?
Ten days on average, compared to six weeks before Claude Code. The gain comes mainly from initial development (from ten to three days), client adjustments (from five to two days), and deployment testing (from five to two days). Initial scoping and creation of graphic mockups do not change, because they depend essentially on the exchange with you.
What about a custom Perfex CRM module?
Six days now, compared to three weeks before. The difference comes from the AI's ability to read the existing CRM architecture, understand internal conventions, and produce a compliant module without rewriting the same plumbing every time. An expense management module with categorization, accounting exports, and bank synchronization that would have taken three weeks just a few months ago is now delivered in six days.
How much difference is there between your quote and the actual time spent?
Less than 10% variance in 90% of cases, on projects completed over the past six months. Previously, variance could reach 30% on complex projects. The reason: AI makes it possible to assess a project's actual workload from the quote stage, rather than discovering pitfalls during execution.
What improves in quality, not just time
The time savings are what you see first. But quality has also improved on three measurable aspects.
Tests are systematic. Before, writing automated tests for each feature took time we didn't bill for and often cut under pressure. Now, every module ships with its test suite, verified and executed. The code that goes to production is tested on a hundred to a thousand automated scenarios, whereas common practice was limited to checking "it works in the normal case."
Regressions are rare. When modifying an existing site to add a feature, the classic risk is breaking something that worked. The AI checks the entire chain after each modification and flags potential breakages before launch. Over the past six months, no client has contacted us for a post-modification regression — compared to an average of one to two incidents per month before.
Documentation is up to date. Every project ships with its technical documentation automatically updated. If you revisit your site in two years and want to know how a particular feature was configured, you have the answer in a readable file. Before, let's be honest, this documentation often got lost in emails and pair-programming sessions.
What this means for you concretely
On the client side, these changes translate into four direct benefits.
Your quotes are more accurate. On the projects we've completed over the past six months, the gap between the initial estimate and actual time spent is under 10% in 90% of cases. Before, this gap could reach 30% on complex projects. You know what you're paying before you sign.
Your deliveries arrive sooner. This seems obvious, but it's the most measurable advantage. If you have an event, a product launch, or critical seasonality, you don't have to wait months.
You get more of our time on the real issues. The time no longer spent writing repetitive code is devoted to more strategic questions: your positioning, your content strategy, your conversion funnel. These are the conversations that make the real difference in your site's performance.
Your post-delivery changes are fast. A minor change request (changing text, moving a section, adding a form field) is handled in hours, not days.
The limits we're willing to acknowledge
We're not selling a miracle product. Here are the limits we've encountered over these six months.
AI doesn't replace judgment. All structural decisions — which positioning, which editorial tone, which information architecture — remain human. AI is an excellent executor; it doesn't decide on your behalf what's good for your business.
Very large projects fall somewhat outside the scope. For a complete overhaul of an information system with ten thousand pages and three interconnected databases, the gain is less obvious because these projects are mainly about strategy, not production. This is rare in the SME world, but it exists.
Systematic human review is required. Everything AI produces goes through a human review before publication. Not because AI often makes mistakes (it rarely does), but because it can't anticipate certain side effects specific to your business. This review takes time and is a significant part of our work.
The learning curve is real. Adopting Claude Code internally, in an agency, didn't happen overnight. It took two months for everyone to become comfortable and for our internal processes to be adjusted. If you're a company wondering, "Can I do this myself?" the answer is yes, but expect several months of learning before reaching industrial efficiency.
Some frequent objections (and our answers)
Will my site be "made by AI"? And what about quality?
Your site is designed, structured, and finalized by the MogaCode team, with AI as a production tool. It's a bit like asking, "Will my dish be made by a robot" when the restaurant uses a modern oven: the cooking remains human, the tool serves the team. On quality, the numbers speak: fewer regressions, more tests, systematic documentation. Quality has improved, not declined.
If you can do this with AI, why should I pay you rather than a freelancer using the same thing?
Because the value of an agency is not found in the act of typing code. It lies in strategy, the experience accumulated across 120+ sites, operational responsibility (security, maintenance, backups), and the ability to intervene 24/7 on a site in difficulty — to understand how our infrastructure makes these commitments achievableAI accelerates work. It neither replaces experience nor responsibility.
Do my personal data go through AI?
No. Your site's code is processed, but none of your personal data or that of your clients is sent to an AI service. We work with strict isolation between source code and production data.
If you save time, will your rates go down?
Honestly, we charge more for value than for time. What changes with AI is that you get more value for an equivalent budget — better tests, better documentation, faster deliveries, more strategic time. The rate reflects expertise and responsibility, not the number of hours typed at the keyboard.
How can I verify what you're saying?
All of our internal tools are published as open source on GitHubYou can inspect them, download them, and have them validated by your own technical team. This is the guarantee of transparency: if we were making anything up, any third-party developer could see it and say so publicly.
Want to see how this could apply to your project?
Whether you're preparing a new site, a redesign, or a custom module, we can together do a quick estimate of what it would look like with our current methods. The estimate takes thirty minutes, and it's without obligation.
Start the conversation on WhatsApp →
For the curious: the tools we use internally are published as open source on GitHub and used by other agencies around the world.