You don't have to be a software developer to build software anymore. That's Jack Arturo's position, stated flatly and more than once – and he builds software for a living. Jack is the creator of WP Fusion, the plugin that connects WordPress membership sites to CRMs like ActiveCampaign. He broke his leg badly enough to spend almost two months in bed, and he used the forced downtime to automate the work he could no longer do. Then he went much further.
His advice for operators is much simpler than his own setup. You start with a desktop app, read-only access to your inbox, and one question: how can you help me? Agents are what that conversation grows into.
This is a breakdown of the three decisions his advice comes down to: where to start, what to build versus buy, and how much access to hand over.
Where to start with AI agents: read-only access and one question
Asked what he'd advise a $5 million-a-year membership business running on WordPress, Jack doesn't reach for anything exotic. Download the Claude desktop app, which is free to get started. Connect it to your email with read access only, no permission to send or delete anything. Then tell it to go through your messages, spend some time, and tell you how it can help. His suggested first routine: have it run every day, tell you what you don't need to read, and flag what you can unsubscribe from.
From there it makes concrete suggestions. “Add me as a bot to your Slack workspace and I'll summarize things for you. I can answer member questions while you're asleep. I can do billing and refunds.” Each suggestion is a switch you flip deliberately. If you don't like how something behaves, you say so and it puts a rule in place. If you're not sure it's accurate, you can tell it to prove it.
This works as a first step because it's reversible. Read-only means nothing can go wrong while you learn what the tool is good at, and an hour or two of conversation is the whole onboarding. If the first step earns your trust, the next is agents: the same tool, given a standing job.
You don't have to be a developer to build
The standard objection is that coding is for developers. Jack's counter is Katie Keith, co-founder and CEO of Barn2, a fellow WordPress product owner from his mastermind. She's not a developer, but she got tired of her reporting plugin and built her own from scratch. It does exactly what she needs.
What Katie brought wasn't coding ability. It was the habit of identifying problems, asking the right questions, and giving feedback until the output does what you wanted.
Jack's read is that this is the skill most successful entrepreneurs already have, whether or not they've ever written a line of code.
His own version of the point: “Ninety percent of the code I write is in programming languages I've never learned.”
AI accelerates people who know the fundamentals
Vic raised the obvious worry: if AI produces something in a domain where you can't judge quality, you won't know it's wrong until it fails in front of your members. Jack pushed back, at least for software. The non-developer founders he talks to can't read the code, but they know when something doesn't work, and when it breaks they ask what caused it. The tool tells them, and suggests the fix. You never have to look at the code.
Where Jack agreed without hesitation is copy and design. A trained copywriter spots unusable AI copy in seconds. A non-copywriter ships it. So the split the conversation lands on: software forgives, because errors surface and the tool helps you repair them. Everywhere else, AI gives massive leverage to people who know the fundamentals of what they're asking for, and far less to people who can't evaluate the output. Point agents at work you already know how to judge.
Build or buy: a two-question test
When something is missing from your stack, how do you decide whether to have AI build it or buy an existing product? Jack's test is two questions:
“Can you describe what it should do and do you know when it's done?”
His worked example is a login form. He used to build them with Gravity Forms: lifetime license, login add-on, Ajax endpoint configuration, custom styler, half a day gone. When he needed a simpler one (a couple of inputs, spam validation, somewhere to post the data) he just asked for it. It works, it's fast, and there are no plugin updates waiting to break it.
That last point matters more than it looks. A commercial product has to keep shipping features to justify itself, and every new feature is a new way to break your site. In Jack's framing, a WordPress site runs on updates from some 80 different companies every week, and on any given week one of them breaks something. A small tool you own just sits there.
From there the decision is plain. If the need is tightly scoped and you can describe what ‘done' looks like, try building it: ask the AI whether it could do it in an afternoon and whether the result would run without maintenance. If it confidently says yes, try it. If it says it can't, throw it out and buy the product.
The far end of the curve, and why supervision matters
One caveat before the impressive part: Jack's own setup is the far end of the curve, not the starting point. His agent, AutoJack, runs on a memory system that Jack says holds about 95,000 of his past decisions. Most of the time, he says, it would make the same call he would, which is how it runs his company day to day. It can also buy things on Amazon and call his bank.
Jack can operate at that end because he invented much of the underlying technology himself and knows exactly what his agent is doing at any moment. Most operators don't have that visibility, and shouldn't copy the setup. What they should copy is the caution that comes with it: an AI musician Jack built got a voicemail from a producer asking for a call back, and obliged, four times, at four in the morning. His phrase for the general problem is what happens if you let your agents loose.
A think tank that Jack's neighbor was part of came up with a seven-stage model for integrating agents into the workplace, and Jack walked through it. Stage one is a connection to something like your Slack. By stage four the agent is a regular member of the team: daily check-ins, briefings, real collaboration. By stage seven it's woven into every process, and your creative people get to just be creative. The model describes depth of integration, but it also works as a trust ladder, and that's the rule worth taking from it: grant each new level of access only after the agent has earned it. Read-only inbox first; bank details maybe never.
When the agent acts as you
Everything in the last section cuts both ways. The access that makes an agent useful is the access that makes it risky. Jack's agent can buy on Amazon, call his bank, and sign up for and cancel credit cards. Give something your email, your phone, and your payment details, and it no longer assists you – it acts as you, in every channel you've connected, at a speed no employee would match. A human assistant who misunderstands you sends one bad email. An agent can send a hundred, spend money, and book calls while you sleep.
The 4 a.m. phone calls are that risk in miniature. Nobody told the musician agent to call the producer four times in the middle of the night. It had a goal, a phone line, and no supervision, and it improvised. Scale that from one awkward night to an agent holding your banking details, and the question stops being how much time agents can save you. It becomes who is watching them.
Vic's analogy: giving an agent access is like dating. You keep your distance at first, then you relax, and one day you realize it has your debit card. The escalation is gradual enough that no single step feels reckless. Which is why the staged access from the previous section isn't caution for its own sake – it is the entire safety mechanism. Vic also mentioned that Jack is now building a security layer to stop agents from getting hijacked by their own instructions, after it happened to his own team. The person with the most capable setup in the room is the one building guardrails. That is not a coincidence.
And even at Jack's end of the curve, there is a human in the loop. When his coding tools hit a question they can't answer, it escalates to AutoJack; when AutoJack can't decide, it goes to Jack's watch. Supervision isn't the training wheels you take off once the agent proves itself. It's the architecture.
Where the durable value sits
Jack is blunt about the shelf life of anything you build: “Whatever you come up with, it might only last for a month.” The platforms will ship your clever tool as a free feature soon enough.
His conclusion is that what compounds is trust – being known as the person people count on to solve their problems. His advice to product companies is to rein in costs, automate what's automatable, and be good to the user base you have. For operators it comes down to this: let agents buy back time and remove friction, then spend what you get back on the service and authority that outlast any tool.
What membership and eLearning operators can take from this
Jack Arturo runs a WordPress product company, not a membership site. But WP Fusion sits between membership sites and their CRMs, and his advice maps directly onto membership operations. Three things transfer:
Start read-only, with one question. Connect your inbox and ask how it can help. What comes back is a concrete list of things it could take off your plate, and you choose which switches to flip.
Point AI at your team's strengths, not around them. AI gives massive leverage to people who know the fundamentals of what they're asking for. Your copywriter with AI produces more and better copy. Your support lead with AI covers more members, around the clock. The same tool in the hands of someone who can't judge the output just makes mistakes faster. Enhance the expertise you already have on payroll before chasing the kind you don't.
Treat saved time as a budget. The first wins come back as hours: inbox triage, Slack summaries, member questions answered while you sleep, billing and refunds off someone's desk. Decide where those hours go before they evaporate. Jack's answer is to spend them on the service and authority that outlast any tool – being the team your members count on when something breaks.
Test every gap against the two questions. If you can describe what a tool should do and you'd know when it's done, try building it before you buy another subscription. If you can't, buy the product and keep your energy for problems that are actually yours.
Grant access in stages, and keep a human in the loop. The agent that summarizes your Slack this month can answer member questions next quarter and process refunds after that – if it earns each step. And even when it does, the questions it can't resolve should escalate to a person, the way Jack's own setup escalates all the way to his watch.
None of this needs a strategy deck. It takes one afternoon and one question.
Jack Arturo is the founder of Very Good Plugins and the creator of WP Fusion, the integration layer between WordPress and your CRM.
Blog: https://drunk.support/
WP Fusion: https://wpfusion.com/
Very good plugins: https://verygoodplugins.com/
Running a Membership or eLearning Platform on WordPress?
MemberFix provides managed WordPress operations for membership, eLearning, and community platforms — so your team never has to worry about the tech breaking during a launch or losing members to site issues. We handle development, support, updates, and monitoring with SLA-backed guarantees. Trusted by industry leaders like Echelon Front, TechServe Alliance, and hundreds of others since 2014.
About MemberFix Radio:
We interview experts and share insights for membership site owners and subscription business leaders building sustainable, profitable recurring revenue models.
