Senior WordPress developer with AI: what an agency that calls itself AI-first looks for

In late September 2026 Blacksmith Agency, a remote US WordPress agency that calls itself AI-first in its job posts, was looking for a senior developer to lead projects. It wanted two things from the candidate at once: use AI to work faster, and still write and review code themselves. The same company was separately hiring an AI Engineer to build internal tools. All three postings are closed now, but they show clearly how an agency that bet on AI splits the work between people and models.

This is one company, not a market study. In mid-October the official WordPress.org Jobs board listed four postings dated 5-13 October: freelancers for client sites, plugin integrations, a WooCommerce store and a part-time technical lead. None had AI in the title.

#What does a senior WordPress developer job with AI require?

The Senior WordPress Developer posting from 27 September and its WordPress Development Lead variant from 30 September describe the same role. The senior is expected to:

• agree the scope of work with the client and tie it to a business goal,
• lead their own WordPress projects: themes, templates, plugins and blocks,
• speed up the team’s work with AI,
• be the person the client talks to about technical matters, and look after the client’s sites after launch.

The requirements are strong WordPress and PHP, modern HTML, CSS and JavaScript, excellent English and experience leading projects, “not just closing tickets”. The job is remote from the US, Canada, Latin America or South Africa, with clients across the US.

The role description ends with this line:

“Drive efficiencies with AI, and lead builds that are fast, clean and custom. You still write and review real code.”

Source: Blacksmith Agency posting on remoteitjobs.app, 27 September 2026

The lead version adds that everyone on the team gets access to the best AI tools, “so you spend less time on busywork and more on the decisions that matter”.

#How much does an AI-first agency pay a senior WordPress developer compared with an AI Engineer?

The Senior posting page gives no salary. The copy of the lead role on the jobfound.org aggregator lists 80-120K USD a year. The AI Engineer at the same company was to earn 3-4.5K USD a month, which is 36-54K USD a year.

The AI Engineer was to build agent workflows on LangChain or LangGraph and internal tools that cut manual work in design and development. The role required two years of programming in Python or Node.js and integrations with the OpenAI or Anthropic APIs. WordPress and PHP were only a plus. So the agency paid the most to the person who knows WordPress, talks to the client and is responsible for the code, not to the agent specialist.

The salary ranges come from an aggregator, not from the agency’s own page, and the locations cover countries with very different costs of living. Do not apply these figures directly to European projects. My rates are on the pricing page.

#Do WordPress developer jobs require AI skills now?

Three postings from one company cannot answer that. The WordPress.org Jobs board from the same period had no posting with AI in the title. What the postings do show is how a company that sells itself as AI-first writes its roles. It did not replace the senior with an agent. It split the work into two roles: a person responsible for the project and the code, and a specialist who builds automation for the team.

A similar split shows up in small teams in Poland. One person is responsible for architecture, code review, WooCommerce and blocks. A second uses AI for first drafts of code, tests and documentation, supervised by the first. The trouble starts when both roles are merged into one junior with Cursor and the client is promised senior-level work.

#What can AI do in a WordPress project, and what can it not do?

On 15 October WordPress.com updated its Agent. Generating a site now takes less than half the time it used to, and most sites are ready in under four minutes. The Agent recommends plugins, installs and activates them from chat, and on request updates the ones already installed. Since March 2026 WordPress.com has also offered MCP tools that let an AI assistant save posts and pages, as drafts by default. I cover the details in my post on the Agent and MCP write.

The safeguards in MCP write are sensible. According to the WordPress.com announcement of 20 March 2026, the assistant confirms every step before making a change, and categories and tags are removed permanently only after an extra confirmation. The announcement also mentioned 30 days in the trash for deleted posts, pages, comments and media, but the current MCP tools reference keeps the trash only for posts, pages and comments: a file deleted from the media library is gone for good, and content that used it loses the image. Every change shows up in the activity log. The write features work on every paid WordPress.com plan. These are good safeguards for content. They do not cover theme code, plugins or integrations, because MCP write does not change those at all, and they do not cover sites on your own hosting.

A client who has seen that demo asks: why pay a senior developer? WebDevStudios answered that on 6 October:

“AI is very good at the 20% of a website you can see. It doesn’t ship the 80% you can’t, meaning the parts that get a site found, keep it secure, protect visitor data, and let a marketing team run it without filing tickets.”

Source: WebDevStudios, 6 October 2026

In practice that 80% means connecting the shop to the warehouse and accounting systems, backups, user permissions, language versions and testing after every update. The WordPress.com Agent will update a plugin when asked in chat, but it will not then walk through the checkout in the shop, check that the invoice reached the accounting system, or notice that the contact form stopped sending messages. Someone has to put these checks on a checklist and run them after every update.

In the same week Gutenberg 24.2 turned on the React 19 experiment in the plugin, unless a theme or plugin with known compatibility problems is active. Custom blocks written for React 18 can stop working in the editor after a change like that. An AI assistant will not notice, because it does not know anything changed. A developer who checks the blocks on a test copy of the site before updating production will. More in my notes on Gutenberg 24.2.

#What does React 19 break in custom Gutenberg blocks?

React 19 removes several APIs that have been marked as deprecated for years but still sit in older blocks and plugins. The React 19 upgrade guide lists, among others:

• ReactDOM.render and ReactDOM.hydrate, deprecated since React 18. They are replaced by createRoot and hydrateRoot.
• ReactDOM.unmountComponentAtNode, replaced by root.unmount().
• ReactDOM.findDOMNode, deprecated since 2018. Refs are used instead.
• propTypes checks and defaultProps in function components. Default values move to the function parameters.
• String refs in class components.

In WordPress code the first case is the most common. Older blocks and admin screens mount components with the render function from the @wordpress/element package, which calls ReactDOM.render under the hood. Since WordPress 6.2 the package has createRoot, and that is the one to use. The second common case is defaultProps in function components generated by an older block template or by a model trained on old code.

The check usually takes an hour per project:

1. Search the theme code and your own plugins for: render(, findDOMNode, unmountComponentAtNode, defaultProps, propTypes.
2. On a test copy of the site, install Gutenberg 24.2 and open every custom block and every settings page in the editor.
3. Watch the browser console. Mounting errors and warnings about removed APIs show up right away.
4. Make the fixes before updating production, not after the client reports that the editor does not work.

An AI assistant will do step one and suggest fixes. Steps two to four need someone who knows the project and knows which blocks matter to the client. That is the same work the Blacksmith posting describes as reviewing real code.

#How do you check whether a WordPress developer really reviews AI-generated code?

Asking “do you use AI?” tells you nothing, because everyone says yes. It is better to test what the posting asked for: whether the candidate can read code and find a bug in it. The interview can follow four steps:

1. Give a short description of a site and a list of three plugins they are allowed to use.
2. Ask for a plan: what AI does, what a person does and at what point a change reaches the live site.
3. Show a piece of AI-generated code with a planted bug, for example a form without a nonce check or user data printed without escaping.
4. Ask what they would do when the shop has to be connected to a warehouse system: stay on WordPress.com or move to their own hosting.

Here is an example of code for step three, in the form models often return it:

add_action( 'admin_post_save_note', function () {
    update_option( 'client_note', $_POST['note'] );
    wp_redirect( admin_url( 'options-general.php?page=client-note' ) );
} );

add_action( 'admin_notices', function () {
    echo '<div class="notice">' . get_option( 'client_note' ) . '</div>';
} );

An experienced developer will point out four problems here. There is no nonce check, so the form can be submitted from another site on behalf of a logged-in administrator. There is no capability check with current_user_can(), so any logged-in user can save the note. Data from $_POST goes into the database without sanitize_textarea_field(). The note is printed without esc_html(), so an injected script runs in every administrator’s dashboard. On top of that, exit is missing after wp_redirect().

A candidate who reviews code every day finds the bug in a few minutes and explains how to fix it. A candidate who only pastes model answers usually starts talking about models. Then add questions about how they actually work:

• Which changes may AI make without a person’s approval, and which may it not?
• Who reviews the code before it goes live?
• Who picks up the phone when a change breaks the site on a Friday afternoon?
• How do you know AI saves time? Do you measure it, or does it just feel that way?

A good answer to the last question includes numbers: time from brief to first version, time from test version to release, and the number of rolled-back changes.

#What should the path of a change from AI to production look like in WordPress?

The line from the posting, “You still write and review real code”, only makes sense if the team has a path on which that review happens. For a WordPress project, five elements are enough:

1. A separate branch in the repository for every task. Code from AI goes there, never straight to the server.
2. Automated checks before review. WordPress Coding Standards for PHP_CodeSniffer have rules that catch a missing nonce, missing escaping and unsanitized input, which are exactly the bugs in the example above. Add static analysis on top, for example PHPStan.
3. A person reviews the changes. Someone reads the diff line by line and is accountable for what they let through.
4. A test copy of the site with data that does not break UK GDPR or the Irish GDPR (ICO / DPC). Changes are checked on the copy, and shop customer data is anonymised before copying.
5. A backup and a rollback plan before release. Everyone knows who restores the previous version and in how many minutes.

This path does not depend on whether a person or a model wrote the code. The difference is that with AI the code grows faster, so points 2 and 3 carry more load than before. A team that sped up writing without strengthening review simply ships bugs faster. If your company does not have such a path, a WordPress security audit or ongoing maintenance is a way to get one from an outside team.

#What does Anthropic’s study say about hiring junior engineers?

Update, 18 October 2026. On 16 October Anthropic published a study of employment inside the company. Two findings matter for this post. By August, the working hours of interactive AI agents outnumbered employee hours six to one. Over the same period junior technical positions made up a shrinking share of hires, and by April the engineering and research teams had cut back hiring for lower technical levels. The study measures the share of hires and gives no absolute drop in the number of juniors.

The results have to be read with the limits the authors point out themselves. It is one company studying itself, at the front of AI adoption, and the authors present it at most as a possible leading indicator for the rest of the economy. The productivity comparisons rely on employees who chose for themselves when to use agents, and the authors note that this choice was not random, so the numbers do not prove that agents caused the change. For a WordPress agency the useful lesson is narrower: when routine coding moves to agents, the junior role is the first to change, which is why the onboarding covered in the next section matters more now than it did a year ago.

#How do you onboard a junior WordPress developer who works with an AI assistant?

A junior with Cursor can write a block in an hour. Before they get access to a client site, even a test one, they need to master four things:

1. Escaping data, nonces and capability checks, with no exceptions like “that is what the model wrote”.
2. Reading changes line by line before merging them into the main branch.
3. Recognising tasks that AI should not get: passwords and keys, changes on production, database migrations.
4. Running every database query through $wpdb->prepare(). Models readily glue an SQL query together with a value from $_GET, because that is how half of the old tutorials they learned from looked. One such line in a WooCommerce orders query is enough for an outsider to read the shop’s customer data.

Without this, a junior just produces bugs faster. With it, after a year and a half or two years they can lead their own projects. That difference is why the AI-first agency in this example required code review from its senior: someone has to teach juniors what is wrong in generated code.

#Are companies that use AI hiring fewer junior developers?

Update, 24 October 2026. The Anthropic study describes one company. The working paper How Does AI Change Labor Demand? Evidence from 41 Countries from the Stanford Digital Economy Lab, by Bharat Chandar and Bouke Klein Teeselink (PDF, 21 September 2026), covers 1.25 billion job postings and 154 million employment records from 41 countries between January 2021 and March 2026. The authors infer AI adoption from job ads that involve generative AI and compare foreign affiliates of adopting multinationals with similar affiliates that did not adopt. By March 2026 the junior share at adopters was 1.9 percentage points lower. That came mostly from hiring above junior level (up 6.7 percent), not from cutting juniors: the 2.5 percent drop in junior employment is not statistically significant, and total employment rose about 3.3 percent, a result the authors call marginally significant. Computer and mathematical occupations, the group most exposed to AI, grew their share of employment by 0.8 percentage points while the junior share inside that group fell 3.3 points. The junior share fell in 23 of the 31 countries estimated separately.

The authors state the limits themselves: the data come mainly from LinkedIn profiles, which over-represent managers and professionals; the firms studied are large multinationals, not WordPress agencies; and the window covers just over three years after ChatGPT. Next to the Anthropic numbers the direction is the same: development work grows, but fewer people enter it at the bottom. When you choose an agency, ask who reviews the code and for how many years they have done it. An agency that cannot find a senior to review AI output can bring one in without a hire through white-label WordPress support for agencies.

#How do you choose a WordPress agency that works with AI?

The same questions an agency asks a candidate, a client can ask an agency. Before signing a contract, ask for four things:

• A description of the path a change takes: from AI through a test copy and code review to production. Who makes the decision at each step?
• An example of rolling back a change that broke a site. How long did it take?
• A clause in the contract stating that the code and access belong to you, not just a licence to the finished site.
• An explanation of how the agency tests blocks and the theme after Gutenberg and React updates.

Spell out the clause about code and access as specific points in the contract, because “we will hand everything over when the project ends” guarantees nothing:

• the repository with the theme code and your own plugins sits in the client’s account, and the agency has access to it, not the other way round,
• the client has their own WordPress administrator account and access to the hosting, the domain and DNS,
• a list of paid plugins, with whose name each licence is in and when it renews,
• where backups are stored and how to restore them, tested at least once,
• a list of outside services connected to the site (payments, mailing, analytics, API keys for AI models), with who pays for each one.

The last point is new. Sites where the agency connected its own API key for an AI model, for example to generate product descriptions, stop working on the day the agency switches that key off. The client should know about every such key and be able to replace it with their own.

Also watch out for a discount offered “because AI does half the work”. An honest agency can lower the price for the stages AI has actually shortened, for example the first version of a template. It should not lower the price for code review, security and on-call support, because AI does not take responsibility there. An agency that discounts the whole project usually saves on testing.

Agencies based in the EU have clear advantages over teams spread from California to South Africa: the same time zone as clients in Germany and Scandinavia, UK GDPR and the Irish DPC regime as the starting point, and the option to meet in person. You cannot beat teams from cheaper countries on rate, so an offer should describe these advantages rather than promise a lower price.

#When should you outsource a project instead of hiring a full-time senior developer?

Not every company has full-time work for a senior developer. In that case you look for an agency that in practice does what the Blacksmith posting describes: it uses AI for first versions and routine work, while a person is responsible for architecture, integrations and releasing changes.

Ask for a quote when:

• you need a WordPress developer who takes over the code and reviews, not just content updates,
• you are rolling out an MCP or AI integration on your own hosting,
• competitors promise that “the Agent will do everything” and you want to know what that really costs to maintain.

In your request, describe the goal, the stack, whether WooCommerce is involved, how many languages and whether you have a test copy of the site. Pricing is individual, based on scope. The form is on the contact page.