The New Wealthy Affiliate Looks Promising—Part 3 of 5

How Much Control Should Ace Have?

From Advice to Action

In Part 2, I looked at how Ace might help us understand, strengthen, and improve what we have already built before encouraging us to create more.

But while working through that, another question came to me.

What happens when Ace moves beyond studying the business and recommending what we should do?

At some point, an agentic AI may also be able to act.

That could mean preparing or publishing content, sending messages, changing settings, starting workflows, spending money, deleting material, or making other changes that affect the business.

I am not saying that all of those things have been confirmed as finished Ace features.

I am asking what kind of control we may need if Ace becomes capable of doing them.

This series grew out of Kyle’s Friday, July 24, 2026, Beer With Kyle presentation about agentic AI and the rebuilt Wealthy Affiliate:

Beer with Kyle Fridays: Agentic AI and a First Look at the New Wealthy Affiliate

Kyle’s presentation introduced the broader idea of Ace connecting different parts of the business into agentic workflows.

That could be extremely useful.

But the more Ace can do, the more important it becomes to decide what it should be allowed to do by itself.

Which actions can it handle after we have already given permission?

Which ones should it prepare and hold for review?

And which ones should always require fresh approval because our money, reputation, relationships, or business may be affected?

I do not think the answer is to prevent Ace from acting.

The question is how to gain the benefits of an active AI business assistant without quietly giving away control of the business.

Why the Ability to Act Matters

Right now, I can ask an AI to study a problem, explain what it finds, and help me decide what to do.

But I am still usually the one moving the information from one place to another and carrying out every step.

An agentic system may be able to connect those steps.

It could study the business, prepare a plan, carry out the parts we approve, watch what happens, and bring what it learns back into the next decision.

That could save a lot of time.

It could also help us manage work that now has to be handled across several different tools.

So I am not arguing that Ace should only explain things and then wait for us to do everything ourselves.

That would leave out much of what could make an agentic system useful.

My concern begins when Ace acts without enough context, without clearly defined limits, or without the owner understanding what is about to happen.

There is also the question of mistakes.

Some mistakes may be easy to correct.

Others could cost money, affect a customer or business relationship, publish something under our names, or remove material we still needed.

Ace should be able to do real work.

But the ability to act needs to come with permissions, safeguards, and a clear way for the owner to remain in control.

Four Actions That Got My Attention

A useful place to begin is with four words:

Send, spend, publish, and delete.

Those are not the only actions Ace may eventually be able to take.

But they helped me see where the question of control becomes real.

If Ace sends something, it may be speaking in our name.

If it spends something, it may be using our money.

If it publishes something, it may affect our reputation, traffic, customers, or legal responsibilities.

If it deletes something, we may not be able to get it back.

The same concern could apply to changing prices, altering affiliate links, contacting customers, modifying settings, or starting a paid campaign.

But these four give us somewhere to begin.

They also do not all carry the same level of risk.

Fixing a spelling error is not the same as publishing a new article.

Organizing files is not the same as deleting them.

Preparing an email is not the same as sending it to an entire list.

Recommending a campaign is not the same as spending money on it.

So I do not think the question is simply whether Ace should be allowed to act.

We also need to ask what kind of action it is taking, what could happen if it is wrong, and whether we should have to approve it first.

Send

One of the clearest places where Ace would need limits is communication.

If Ace can send emails, reply to messages, follow up with leads, contact customers, post to social media, or send newsletters, then it may be speaking in our name.

That could save a lot of time.

It could also create problems very quickly.

A message may sound wrong for the situation. It may make a promise we did not intend to make. It could include information that should have remained private, misunderstand what the other person was asking, or damage a relationship that took years to build.

This is different from Ace helping us prepare a reply.

I would be comfortable with Ace drafting a message, explaining what it is trying to accomplish, and holding it for review.

That would still save time without giving up control.

Routine messages may eventually be different.

For example, a member might approve a narrow set of low-risk replies in advance, such as confirming that a form was received or sending a standard follow-up after someone requests information.

But even then, the limits would need to be clear.

What counts as routine?

What happens if the message includes a complaint, refund request, legal concern, personal problem, or something else the system does not fully understand?

Ace should be able to recognize when a message has moved outside the approved situation and stop.

It should also show us exactly what it plans to send before anything important goes out.

The basic rule I would want is simple:

Ace may help write and prepare communication, but anything unusual, sensitive, or capable of affecting trust should require human review before it is sent.

Spend

Another place where Ace would need clear limits is money.

If Ace can buy advertising, increase a campaign budget, renew a service, purchase credits, hire an outside service, or start a paid tool, then it is no longer only helping us make a decision.

It is using our money.

That does not mean Ace should never be allowed to handle paid actions.

In some cases, it might be useful for it to manage routine expenses we have already approved.

But I would want the limits to be very clear.

For example, a member might allow Ace to keep an existing advertising campaign within a set budget.

That is different from allowing it to increase the budget, start a new campaign, purchase a new service, or sign up for another subscription.

Ace should be able to recommend a paid action without carrying it out automatically.

Before any important expense, it should explain:

• what it wants to purchase;

• why it believes the expense is needed;

• what it is expected to accomplish;

• how much it may cost;

• whether the cost is one-time or recurring;

• and what could happen if the recommendation is wrong.

Members should also be able to set limits.

Those might include:

• a maximum amount Ace can spend without fresh approval;

• categories where no automatic spending is allowed;

• warnings before a budget reaches a certain point;

• and a requirement that every new subscription, vendor, or recurring charge be approved separately.

There is also a difference between spending money and using credits or other resources already inside the platform.

Even then, I would still want Ace to tell me what it plans to use, explain why, and get my approval before using it.

A balance of tokens, credits, or advertising funds may not feel exactly like money leaving a bank account, but it still represents something limited that could be wasted.

The basic rule I would want is this:

Ace may help us plan, compare, and manage approved expenses, but it should not quietly decide that our money or limited resources are available for whatever it believes the business needs next.

Publish

Publishing is another area where Ace could save a great deal of time.

It may be able to help research an article, organize the material, prepare a draft, add links, format the page, create images, check disclosures, and get everything ready to go live.

That would be useful.

But preparing something for publication is not the same as deciding that it is ready to represent us publicly.

Once an article, email, social media post, video, book, product page, landing page, or other material is published, it can affect how people see us and our businesses.

It may contain a factual error. It may say something more strongly than we intended. It may make an unsupported claim, use the wrong tone, include a bad link, or weaken the voice we have spent years developing.

It could also create problems with affiliate disclosures, copyright, privacy, or other responsibilities that still belong to the business owner.

For those reasons, I would want Ace to show me the final version before anything important is published.

It should also show me:

• what it changed;

• why it made those changes;

• what sources or information it relied on;

• what it is uncertain about;

• and whether anything still needs human verification.

There may eventually be some low-risk publishing actions that a member chooses to approve in advance.

For example, Ace might be allowed to schedule material that has already been reviewed, correct a formatting problem, or update a link that the owner has already approved.

But publishing new material under someone’s name is different.

Ace can help prepare the work. It can help improve it. It can even help us notice problems we missed.

The final public version should still require the owner’s approval.

Delete

Deleting something may be the clearest example of an action Ace should not take casually.

It may find an article that appears outdated, duplicated, weak, or disconnected from the rest of the website.

It may also find old files, abandoned campaigns, unused images, inactive workflows, or other material that no longer seems useful.

Sometimes deletion may be the right answer.

But Ace may not understand everything that material is connected to.

An older article may still have backlinks, bring in a small but important group of visitors, support another page, preserve useful research, or contain material we plan to use again later.

The same problem could apply outside the website.

A story note may look like an abandoned idea when it is actually Canon. An older draft may contain the only surviving version of a scene. A file may seem duplicated even though the differences matter.

That is especially relevant to my own work.

I have story development spread across many ChatGPT threads. Some of it is Canon. Some of it is a possibility, a rejected direction, a contradiction, or a question we have not answered yet.

An agentic system could save me an enormous amount of time by finding, organizing, and comparing all of that material.

But I would not want it deciding on its own which version was official or which material could be deleted.

It could show me the conflicts, explain what appears duplicated or outdated, and recommend what should be kept, combined, archived, or removed.

I would still need to make the final decision.

In most cases, I think Ace should archive before it deletes.

It should also show us:

• what is being removed;

• why it recommends removing it;

• what other material may depend on it;

• whether links, workflows, or projects will be affected;

• and whether the action can be reversed.

Permanent deletion should require clear approval.

The basic rule I would want is this:

Ace may help us identify what no longer belongs, but it should not permanently erase part of the business—or part of its history—without the owner deciding that it is safe to do so.

The Risk Depends on the Action

The more I thought about these examples, the clearer it became that one permission setting probably would not be enough.

Some actions may be easy to reverse.

Others may cost money, affect trust, expose private information, damage a relationship, or create a problem in another part of the business.

Ace may need to look at questions such as:

• Can the action be reversed?

• Does it involve money or limited resources?

• Will it be public?

• Could it affect customers, subscribers, or business partners?

• Does another system or project depend on it?

• Are there legal, ethical, or privacy concerns?

• How difficult would the mistake be to repair?

A low-risk action might be something we have already approved, and Ace can safely carry out within clear limits.

A higher-risk action should require more explanation, more review, and a fresh decision.

To me, the basic principle is:

The greater the possible consequence, the stronger the approval requirement should be.

Different Levels of Permission

That led me to wonder whether Ace may need more than a simple choice between “allowed” and “not allowed.”

Maybe the permission system could work in levels.

Explain Only

Ace studies the situation and explains what it found.

It may point out a problem, identify a risk, or show us several choices, but it does not take any action.

That may make sense when the issue is complicated, the consequences are unclear, or Ace does not have enough information.

Recommend

Ace tells us what it thinks should be done and explains why.

It should also tell us:

• what information it used;

• what it is assuming;

• what other choices may be available;

• and what could happen if the recommendation is wrong.

Then it waits for us to decide.

Prepare for Review

Ace may draft the email, prepare the article, build the workflow, organize the files, or set up the proposed change.

But it does not send, publish, spend, delete, or complete the action until we review and approve it.

This may be one of the most useful levels.

Ace could do much of the work without taking the final decision away from us.

Carry Out Approved Low-Risk Actions

There may be some actions we deliberately allow Ace to complete within clear limits.

For example, it might:

• correct approved formatting problems;

• update links we have already reviewed;

• organize files into an approved structure;

• schedule material that has already been accepted;

• or perform routine maintenance we have clearly defined.

Even then, Ace should keep a record of what it did and make the action easy to review.

Fresh Approval Required

Some actions should always require a new decision.

That may include:

• spending money;

• publishing important new material;

• sending unusual or sensitive messages;

• deleting or permanently removing content;

• changing prices;

• starting a new subscription;

• or making a commitment in the owner’s name.

Approval for one action should not automatically become permission for every similar action afterward.

Different members may want different levels of automation.

Some may want Ace to do very little without review.

Others may become comfortable giving it more responsibility after they have learned how it works.

The system should allow that flexibility.

But I think the safest starting settings should protect members who may not yet understand what they are authorizing.

Ace may be able to earn more freedom through clearly defined permissions and a history of reliable work.

It should not begin with broad authority and leave us to discover the risks afterward.

Ace Should Explain Before Acting

Getting approval is not enough if the member does not understand what Ace is asking permission to do.

Before an important action, Ace should explain the action in plain language.

It should tell us:

• what it plans to do;

• why it recommends doing it;

• what information it used;

• what it is assuming;

• what it does not know;

• what may be affected elsewhere;

• what could go wrong;

• and whether the action can be reversed.

That explanation should come before the approval button, not after the action has already been taken.

Ace should also be honest about how confident it is.

There is a difference between saying that the evidence strongly supports a recommendation and saying that something is only one possible explanation.

For example, Ace may notice that traffic fell after an article was updated.

That does not necessarily prove the update caused the decline.

A search-engine change, seasonal interest, stronger competition, or something else may be involved.

If Ace is uncertain, it should say so.

It should not make every recommendation sound equally definite merely because confident wording is easier to follow.

This matters because many members may approve an action based largely on how Ace explains it.

If the explanation leaves out uncertainty, assumptions, or possible consequences, then the approval is not fully informed.

I would also want Ace to give me reasonable alternatives.

It may recommend deleting an article, but perhaps updating, combining, redirecting, or archiving it would solve the problem with less risk.

It may recommend buying a tool, but perhaps something less expensive—or something already available within WA—would do the job.

Ace should not overwhelm us with every imaginable possibility.

But it should tell us when there is more than one reasonable path.

The goal is not simply for Ace to ask, “Do you approve?”

The goal is for us to understand what we are approving and why.

Undo, History, and Accountability

Even with good permissions and informed approval, mistakes can still happen.

Ace may misunderstand an instruction.

We may approve something before noticing a problem.

A change that looks helpful at first may cause trouble somewhere else.

That is why control cannot end with the approval button.

We also need to be able to see what Ace did afterward and, when necessary, undo it.

Ace should keep a clear record of:

• what it recommended;

• what information it used;

• what the member approved;

• what action it took;

• when the action happened;

• what changed;

• and whether the change can be reversed.

For content, that could include a side-by-side comparison showing the earlier version and the revised version.

For a workflow, it could show which steps were added, removed, or changed.

For a message, it should preserve the exact version that was sent.

For spending, it should show what was purchased, how much was used, and which approval allowed it.

This record should be easy for the member to understand.

It should not require digging through technical logs to figure out what happened.

I would also want safeguards such as:

• previews before important actions;

• version history;

• backups;

• temporary archives;

• rollback;

• and recovery periods before permanent deletion.

These protections become even more important when one change affects several connected parts of the business.

For example, Ace may update an article, change its internal links, adjust a related email, and modify a workflow connected to that page.

If the result creates a problem, can the member reverse the whole group of changes?

Or would each part have to be found and repaired separately?

A system is not fully under our control merely because it asked permission before acting.

We also need to know what it did, understand the consequences, and have a practical way to correct or reverse the action when something goes wrong.

Protecting Human Voice and Judgment

Business control is not only about whether Ace can click a button, send a message, or make a change.

It is also about whether the business still reflects the person who built it.

Ace may be able to study traffic, conversions, engagement, rankings, and other measurable results.

That could help it identify patterns and recommend what appears most efficient.

But efficiency is not the only thing that matters.

A recommendation may look good according to the numbers and still work against:

• the owner’s values;

• the voice of the business;

• long-term trust;

• relationships with readers or customers;

• future plans;

• or the reason the business was created in the first place.

I can see this problem in my fiction work.

The material for several stories has accumulated across a large number of conversations, and later decisions do not always sit beside the earlier ideas they changed.

An agentic system could help trace those developments, gather related material, expose contradictions, and prepare a clearer continuity record.

But organization is not authorship.

Even after the system assembled the evidence, I would still need to decide which ideas remain part of the story, which ones were only explored, and how conflicts should be resolved.

The same principle applies to a business.

Ace may be able to show us what appears to work best, but it may not understand why we chose a slower path, why one small group of readers matters, why we will not promote a certain product, or why a project that earns little money still matters to the larger plan.

It should be able to ask questions and learn those reasons.

It should also remember the rules, values, and corrections we give it.

But it should not quietly reshape the business around whatever is easiest to measure, automate, or monetize.

Ace should help us think more clearly and carry out the decisions we make.

The final judgment should remain with the person who understands the history, purpose, and consequences of the work.

A Business Partner, Not the Business Owner

When I step back and look at all of this, I still come to the same conclusion.

The ability to act may be one of the things that makes Ace valuable.

It may be able to connect different parts of the business, prepare work, carry out approved actions, watch the results, and bring what it learns back into the next decision.

That could save us a great deal of time.

It could also help us manage work that now has to be handled across several tools, websites, files, and communication channels.

But the more Ace can do, the more important the safeguards become.

We may need:

• clear permissions;

• explanations we can understand;

• approval requirements based on risk;

• records of what happened;

• and practical ways to undo mistakes.

The point is not to stop Ace from acting.

An agentic system that could never act would leave much of its potential unused.

The point is to make sure its actions remain connected to our intentions, limits, knowledge, and approval.

Ace could become an analyst, organizer, strategist, workflow assistant, and active business helper.

But it should not become an unsupervised replacement for the person who owns the business and must live with the consequences.

We should still decide what the business is, what it stands for, where it is going, and which risks are worth taking.

Ace may help us see more clearly and work more efficiently.

It should not quietly take ownership of the decisions.

Looking Ahead to Part 4

Even when we understand what Ace plans to do and approve the action, there is another kind of control we will need.

We will need to understand what the action costs.

That may include money, but it may also include tokens, credits, usage limits, processing time, or other resources inside the platform.

A simple task may use very little.

A larger workflow involving research, writing, images, video, automation, or several connected businesses may use much more.

Members should not have to guess how expensive a task may become after Ace has already started.

Before beginning a major workflow, I would want Ace to explain:

• what resources the task may use;

• whether the estimate is certain or only approximate;

• what parts of the job are likely to use the most;

• whether a lower-cost approach is available;

• and when it will stop and ask before continuing.

That is the subject I want to examine in Part 4.

How will tokens, credits, limits, and usage costs work—and how much control will members have over them?

This Five-Part Series

Part 1: The New Wealthy Affiliate Looks Promising—But Will It Understand the Business We Already Built?

Part 2: The New Wealthy Affiliate Looks Promising—Will Ace Improve What We Already Have Before Creating More?

Part 3: The New Wealthy Affiliate Looks Promising—How Much Control Should Ace Have?

You are here.

Next—Part 4: Tokens, usage, limits, and what “anything and everything” may cost.

Sources and Related Discussion

Kyle’s original presentation:

Beer with Kyle Fridays: Agentic AI and a First Look at the New Wealthy Affiliate

Kyle’s later discussion of the rebuilt platform:

If You Build It, They Will Come. So We Are Building It.

Kyle has described the larger goal of Hubs as understanding the business as a connected system.

The questions in this article about sending, spending, publishing, deleting, permission levels, and safeguards are my own exploration of what human control may need to look like as Ace becomes more capable.

They should not be read as confirmation that every action discussed is already a finished or announced Ace feature.

A Note About AI Assistance

Yes, I used ChatGPT to help me organize, draft, and polish this blog. But my hands were in every part of it.

The ideas, questions, concerns, and final decisions are mine. I reviewed every section, challenged wording that did not sound like me, requested changes, and decided what stayed in the blog.

I also added my own plans, experiences, opinions, concerns, and examples wherever they helped explain what I was trying to say.

AI helped me shape and organize the work. It did not replace my judgment, my experiences, or my voice.

JD