AI can sort hundreds of messages out of your inbox, but you still have to catch the email that could become your next major project. For agency owners, time saved doesn’t automatically mean money saved.
On The WP Minute’s Agency Action Show, Kurt von Ahnen and Toby Cryns discuss where automation has helped their businesses and where it has left expensive work unfinished. Their experiences keep returning to human judgment, even as the tools become more capable. That tension starts with the daily problem of sorting business email.
AI and spam filtering still need human judgment
Toby already uses Gmail’s built-in spam filtering through Google Workspace. He thinks it does a decent job compared with how chaotic his inbox could be. Yet he still missed an email the previous week.
That experience complicates the idea that AI can take over the work of a virtual assistant. Filtering messages is part of the job, but recognizing an unanswered message that deserves attention requires another kind of review.
Toby also sees AI on both sides of the spam problem. In his view, AI has helped increase the volume of unwanted messages, while businesses now use AI to sort through that volume. Human problems still need human attention.
Kurt’s Gmail experiment
Kurt tested a more customized approach using Magi and Claude. He asked the tools to examine a Gmail account and propose a way to handle its spam problem.
The resulting plan separated spam, newsletters, and promotional messages while keeping important person-to-person email in the inbox. It also connected to Slack so Kurt could receive daily summaries of what the system moved.
This was a sample project on an account he didn’t use for day-to-day business. That distinction mattered because he wasn’t ready to give the same system control over his main inbox.
Filtering messages doesn’t remove the work of responding
Kurt’s experiment moved large batches of messages, sometimes 100, sometimes 90, sometimes 72. However, he gradually stopped reviewing the Slack summaries. He saw the notifications, assumed the system was doing a good job, and largely ignored the email account.
An important message from his wife about their vacation stayed in the inbox. A few other messages also appeared where he expected them. Those successes were reassuring, but they didn’t resolve his concern about what might disappear from view.
He described a hypothetical inquiry from a school with 3,000 students that wanted to migrate its e-learning website. Missing that message could mean losing a project worth $70,000 or $80,000.
Someone still has to check and reply
Toby brought the discussion back to the remaining work: someone must read the important messages and respond. Even an AI-generated draft may need changes before the owner is comfortable sending it.
He compared that with hiring an employee. Getting 80% of the work done can still leave a consequential 20% for someone else to catch or correct.
Kurt raised the human alternative: depending on what an owner’s time is worth, paying someone for 90 minutes or two hours of daily email review might make sense. The conversation left that choice open. AI filtering, human assistance, and handling the inbox personally each leave different responsibilities with the owner.
Cheaper project management didn’t get the work done
For Toby, the cost of relying on cheaper approaches had already become tangible. He hired a local project-management company that charges substantially more than he wanted to spend.
The added expense was hurting project budgets, but his earlier combination of AI and inexpensive project management hadn’t delivered the result he needed. He described the new spending as the penalty for getting that decision wrong.
His experience put a limit on the promise of low-cost automation. A cheaper system can still become expensive when projects need additional people to get them back on track.
Kurt’s new contract focused on the customer-facing work
Meanwhile, Kurt was about to start a project-management contract with a different division of responsibilities.
His team at Manana No Mas would handle client onboarding, customer-facing updates, project oversight, and keeping the work on schedule. Another company already had the technical team to build the project.
That company lacked someone who could manage the customer-facing side. So it hired Kurt’s team for that role rather than for development.
The arrangement showed a specific kind of agency work that remained valuable: coordinating delivery and communicating with the customer, even when another team handled the technical implementation.
AI savings depend on how the team works
Toby was also trying to understand whether AI could reduce the time spent on design.
He had brought in a local firm that included a designer, and he described the people involved as excellent professionals. Still, he wanted to know how they used AI in their process. If a task was expected to take 10 hours, could a different workflow bring that down to five?
He was encountering similar friction with his full-time team. Their use of AI wasn’t matching the depth of use he thought the business needed.
For Toby, this wasn’t a statement of complete trust in AI. He felt pressure to use it because he needed better margins.
That created another responsibility for the owner: understanding how the tools fit into the actual work, rather than assuming that having AI available would automatically reduce billable hours or delivery costs.
Removing Elementor lifted a desktop score from 56 to 92
Kurt shared a completed experiment with a clearer result. Using AI-assisted coding, often called vibe coding, he removed the Elementor website builder and other bloat from an abandoned website.
His goal was to make the site more efficient. He wanted better page speed, less unnecessary weight, and no ongoing payment for a page builder he no longer wanted to use.
The site’s reported desktop page-speed score improved from 56 to 92.
Kurt also thought the site looked better after the changes. That was his subjective assessment, while the score gave him a concrete performance result.
He was careful not to frame the project as a general accusation against Elementor. He described it as a heavy tool that he had chosen to remove from his own sites.
For this experiment, AI helped him finish a specific change with benefits he could identify: a higher desktop score, an appearance he preferred, and one less software expense.
Rebuilding WooCommerce features raises a maintenance question
Toby was considering a broader change on a complex WooCommerce project with about 30 extensions.
His direction to the developer was to keep WooCommerce and the important payment, checkout, and shipping integrations. That included Stripe and the site’s UPS shipping API connection. He didn’t want to rebuild those integrations.
However, he wanted to reconsider roughly 25 other plugins and reproduce only the features the project needed.
Build the feature, without the rest of the product
A fly-out cart was his example. If the requirement was a cart panel that appeared when someone clicked the cart button, he saw that as a focused feature AI could help build.
The code could live in one custom plugin or several smaller plugins. The structure mattered less to him than having efficient components without features the client didn’t need.
The developer still had to consider updates
Toby’s developer pushed back on maintenance. Custom plugins would still need updates, and the team would have to decide how much to trust AI with that work.
Toby’s response was that he believed commercial plugin developers were also using AI to update their products. From his perspective, custom development could use similar tools while carrying less feature bloat.
This was a proposed direction for the project, rather than a completed result like Kurt’s Elementor experiment.
A custom wish list led to more customization requests
Kurt understood the appeal of focused custom code, but his FluentCart experience made him more cautious.
A customer wanted a wish-list add-on with several particular requirements. Kurt used AI-assisted coding to build it, met the requested functionality, and didn’t charge the client. He treated the work as a learning exercise to see whether he could pull it off.
Then the client wanted more changes, including a different heart shape and changes to the button.
At that point, Kurt decided he would rather return to an established add-on costing $79 per year. It included options and support, and it gave him a clearer boundary to present to the customer.
WordPress flexibility can expand the work
Kurt’s frustration extended beyond that one feature. He felt WordPress customers often expected anything they disliked to be adjustable because the platform allowed customization.
He contrasted that with a SaaS platform such as Kajabi. In his comparison, a SaaS provider could insist that the interface remain as supplied while continuing to charge its monthly fee.
WordPress agencies, meanwhile, often respond by offering another tweak.
For Kurt, the paid add-on was attractive because it reduced the ongoing negotiation over details. Its cost bought an existing product, support, and a defined set of options rather than another open-ended custom assignment.
Contractor introductions can create project-management friction
Kurt’s contract model often puts his team in charge of only part of a job. A client may then hire someone else for another portion and introduce that contractor by email.
One of Kurt’s personal trigger points is seeing a freelancer use a Yahoo, Gmail, or AOL address. He reads that as a concern about professional presentation.
He also acknowledged that his assumptions about why the client hired someone else, such as choosing a cheaper contractor, were assumptions. A personal email address alone didn’t establish the contractor’s ability.
Still, the introduction affected his willingness to bring that person into the working relationship.
Responsibility stays with the person who hired the contractor
Kurt was clear about the boundary he sets when a client hires a separate freelancer.
His team will report if that contractor falls behind. However, it won’t accept responsibility for the timeliness of work from someone it didn’t employ or hire.
If the outside contractor misses a deadline, the client must address that issue with the contractor.
His email-address frustration was a personal reaction. The responsibility boundary was a specific part of how he handled shared projects.
Client-facing roles change expectations about dress
Toby’s comparable trigger point came from interviewing project-manager candidates.
He had completed two late-stage interviews that day, and both candidates appeared in T-shirts. Because he was hiring people who might represent the agency in front of clients, he expected a more deliberate presentation.
He also acknowledged the irony: he was wearing a T-shirt during the conversation.
His concern was that clothing could become a distraction. He wanted clients to remember the work and the communication, rather than what someone wore.
Kurt dresses differently for different audiences
Kurt gave two contrasting examples from his own work.
When creating content for LifterLMS, Chris wanted him in a T-shirt because that supported a relaxed, relatable presentation. For training with the Marine Retailers Association or Powersport Academy, Kurt wore a button-down dress shirt.
He tied that second choice to the customer’s expectations. When someone is paying $45,000 for training, he feels a professionally dressed instructor should show up.
Both hosts recognized that a T-shirt doesn’t tell anyone how capable a person is. Their concern was perception, especially when an interview or a paid client engagement called for a different presentation.
Toby summed up the distraction: the conversation had become about T-shirts instead of the quality of someone’s work.
Software support is part of the buying experience
Kurt raised another frustration with professional presentation: buying software and then discovering that support happens in a Facebook group or a course community.
One product directed him to a community containing instructional courses and a social feed where customers could ask questions.
He could see the appeal. The setup was creative and affordable. Still, as a paying customer, he wanted ticketed support and confidence that a real team was handling his request.
Toby connected this with the earlier examples. An email address, a shirt, or a support destination can become the thing customers discuss instead of the service itself.
The support arrangement therefore mattered to them alongside the software’s features. A community might provide useful information, but it wasn’t the same experience Kurt expected from a support ticket.
AI exposed missing task updates, but Toby still had to act
Toby’s biggest operational frustration involved accountability.
Some projects were months behind. The local project-management firm was helping with only a handful, while the business had a much larger project load.
He wanted team members to stop depending on project managers to discover delays and chase explanations.
“You need to tell the project manager why it’s late three days before it’s late.”
The expectation was a short daily comment
Toby sent an escalation email requiring employees to leave a comment on every task they worked on that day.
The update could be brief. “Made 50% progress” was enough to show what had happened. He also pointed out that ClickUp AI could help write the comment.
After sending that message, he used AI to build an app that pulled reports from ClickUp. It showed which employees had logged time, which tasks they had worked on, and whether they had left a comment.
The report made follow-up easier
The app identified missing updates from two employees he had just emailed. It also supplied text he could copy into direct messages to follow up.
That reduced the work of finding the problem and preparing the message. Toby still had to confront the employees and manage the situation.
His preferred direction was capable staff supported by AI. The frustration came from employees failing to meet a simple expectation, even when tools were available to make it easier.
The AI subscription stack keeps growing
Kurt initially thought Magi would give him access to the tools he needed through one account. In practice, some connections weren’t available, and he ran out of credits.
He then bought lifetime deals on additional tools as backups.
That prompted a wider look at what he was already paying for. Canva and Descript carried monthly costs. X Premium gave him access to Grok. Other tools and accounts added more spending.
Kurt admitted that fear of missing out played a role, especially when someone offered a lifetime deal.
New features don’t necessarily replace existing tools
He was also interested in new Grok bots, but using them required a separate upgraded Grok account.
When he researched whether those bots could replace tools such as Claude, Gemini, or Perplexity, the answer he received was that they would add features rather than replace the existing accounts.
That was the source of his concern: each appealing capability could become another expense.
Toby recognized the same pattern in his own subscriptions. During the discussion, he remembered that he also paid for Perplexity. Some days he used his subscriptions heavily enough to hit limits, but the growing number of accounts was still difficult to track.
Replacing one tool can leave another bill in place
Toby was reconsidering established tools as well as new AI products.
His team was heavily invested in Figma, and designers liked using it. Yet he wondered whether he still needed to spend about $100 per month on it when he could produce something useful in Lovable.
He hadn’t settled that question. Figma was part of the existing workflow, so the comparison involved how people worked as well as what each tool could produce.
ClickUp raised a similar issue. Toby paid for AI add-ons for meeting recording and its Brain features. That morning, he connected to ClickUp through an MCP connection and wondered whether that could replace some of the paid functionality.
However, he was also paying for Claude to support the new approach.
A possible replacement therefore didn’t establish a saving by itself. He still had to work out which subscription could go away and which bill would remain.
Agency pricing hasn’t caught up with every new cost
Kurt’s larger question was whether agencies were recovering these expenses through client billing.
He recognized that AI could save time. But he also wanted to know whether tool costs were making it into the business’s pricing, or whether the agency was absorbing them.
He wasn’t certain Manana No Mas had accounted for that consistently. Some agency rates might already be high enough, while other work could be priced at half the value he believed it delivered.
A solo owner may capture the savings differently
Toby sometimes imagined operating as a solopreneur with the current AI tools.
He thought that model could produce stronger margins because he could apply the tools directly to his own work. His existing agency, however, still had older pricing and client expectations attached to that pricing.
Kurt recognized a related choice in his own schedule. He could delegate a task or keep it himself and retain an 80% margin.
At the same time, he questioned whether keeping the work was the smartest decision. Delegating it could free him to sell another project.
Neither host resolved that tension. Staffing, delivery time, software costs, and opportunities to sell more work pulled the decision in different directions.
Where Kurt and Toby continue the conversation
Toby’s agency, The Mighty Mo, handles WordPress security, SEO, and lead generation. Kurt runs Manana No Mas, and both hosts pointed to LinkedIn as a way to connect with them.
They also highlighted The WP Minute’s open Slack community for people who want stronger connections within WordPress.
A discussion about cameras, audio, and other production equipment would wait for another episode. This conversation stayed focused on the systems behind agency work and what it costs to keep those systems running.
Automation still leaves the owner accountable
Kurt’s inbox experiment and Toby’s ClickUp app show two different limits of automation. A tool can move messages or reveal missing updates, while the owner still has to review what happened and respond.
The clearest savings came from a specific completed task, such as Kurt’s website improvement. Broader promises about replacing staff, plugins, or subscriptions remained tied to maintenance, client expectations, and actual project margins.
A cleaner inbox is useful. Its business value still depends on whether the important email gets answered.