For years, one of the biggest obstacles to building an online business was access. Website software cost more. Hosting was harder to understand. Research took longer. Publishing could mean learning code, hiring somebody, or living with a website that looked as though it had been assembled during a power outage.
That problem hasn’t disappeared completely, but it has changed. Now we have website builders, research tools, design platforms, analytics, training, AI, automation and enough advice to remain very busy without necessarily building much of anything. Tools aren’t the scarce part anymore. What’s scarce is the judgment about which ones are worth touching, what deserves attention, what can wait, and — now that AI is doing more of the work — what exactly we’re handing over when we let it.
Another dashboard doesn’t answer much of that.
Tools Were Never the Bottleneck
Open almost any online-business platform and you’ll find plenty to do. Build a website, research a market, generate an article, start an email list, make a video, check the analytics. Install something. Connect it to something else. Then automate the thing you just connected five minutes ago.
None of that is inherently foolish. The problem is that capability usually arrives before context. Someone who already runs a local business isn’t solving the same problem as someone who hasn’t picked one yet, and having an established website calls for a different first move than a domain name sitting on top of an empty WordPress install.
Even two beginners aren’t starting from the same place. One might have twenty hours a week while the other has four. One can write; the other would rather have dental surgery than draft a paragraph. Add somebody who knows the software cold and still hasn’t worked out who they’re trying to help, and you’ve got three different starting lines.
Put all of them in front of the same collection of tools and technically you’ve given everyone freedom. Whether you’ve given anyone progress is a separate question, because a system can be perfectly capable of doing something that isn’t worth doing.
The Feed Doesn’t Know Your Business
Online advice makes this worse because most of it arrives without enough context. You need a website. Websites are finished. Start an email list now — or don’t waste time on email until you have traffic. Publish more. Publish less. Use AI everywhere. Never let AI near your writing.
Some of those statements can be sensible in the right circumstances. The circumstances are usually the part that goes missing. Recommendation systems can be very good at surfacing content people are likely to find interesting or useful, but they aren’t sitting down with your business, your resources, your customers, your skills and your stage of development before deciding whether the advice actually fits.
So I care less about whether something “worked” in the abstract than about who it worked for, under what conditions, and at what stage. That usually tells you more.
AI Has Made the Judgment Problem More Obvious
AI has made many kinds of production much easier, and that’s genuinely useful. I use it because it can research, organize information, challenge ideas, compare approaches, help examine a problem and turn rough material into something much more workable.
But easier production creates its own trap. A weak premise used to be slowed down by the labour required to do anything with it. Now one weak premise can become an article, five social posts, a newsletter, a video script, several graphics and next week’s content calendar before anyone stops to ask whether the original idea was any good.
Making the thing is getting easier. Deciding whether the thing should exist hasn’t been solved at all. Somewhere in all of this, we’ve also started using automation as though it describes one simple thing. It doesn’t.
Delegation Is Not the Same as Automation
If I ask AI to summarize several documents so I can examine the important differences, that’s delegation. Asking it to draft several alternatives and then choosing among them myself is more delegation. But if a system decides what gets published, publishes it, judges the result and changes strategy without anyone looking, we’ve moved into something else. We’ve given it authority.
Those arrangements shouldn’t be treated as interchangeable. The question isn’t only whether AI can do the task. More and more often, it can. What matters more is what happens when it gets the task wrong.
Letting AI resize fifty images doesn’t keep me awake at night. Organizing research notes or suggesting internal links isn’t the same as deciding an unsupported claim should be published under my name because it sounds plausible. Nor is that the same as changing a customer promise, publishing advice with real stakes, deleting something important or altering the direction of a business without review.
There is a strong counterargument here, though. Human review isn’t automatically valuable just because a human is involved. People rubber-stamp things. They miss errors, get tired, bring their own biases to the review, and sometimes know less about the problem than the system they’re supposedly supervising. A mandatory click on an approval button can easily become theatre.
So I’m not arguing that humans should sit in the middle of everything forever. That would be silly. Sometimes a person in the loop really is just slowing things down. In other situations that person is the reason a bad decision doesn’t make it any further. The difficult part is knowing which situation you’re in.
Capability Does Not Create Authority
This is the part that gets blurred very quickly. AI capability keeps expanding, but authority doesn’t come with it. We hand that over ourselves.
A model may be able to research a question, write a recommendation and explain its reasoning in wonderfully tidy prose. None of that tells us whether the evidence was good, whether the sources were current, or whether one bad assumption sent the whole thing off in the wrong direction. Fluency certainly doesn’t establish truth.
That’s where the Second Question comes in:
How do we know this?
Not as a slogan. As an actual working question. Where did the information actually come from, and is the source saying what the summary claims it’s saying? Is it current, or is it a fact wearing the clothes of company marketing, community opinion, or somebody’s prediction dressed up a little too nicely — and what would actually change our minds about it?
That isn’t hostility toward AI. It’s what stops useful assistance from quietly acquiring more authority than we’ve actually earned the right to give it.
What Counts as Verification
One of the laziest forms of “verification” is producing an entire piece of work with AI and then asking the same AI whether it’s accurate. It can catch things — I do use it that way — but it’s not much of an independent check.
The stronger check usually happens outside the answer itself. That means opening the source, finding the underlying number, checking the date, comparing competing accounts, and looking for the original research instead of the article describing the article describing the research.
Sometimes the answer gets weaker when you do that, and that’s fine. The evidence might support “possibly” rather than “yes.” It might only hold under particular conditions, or there may simply not be enough evidence yet to say much at all. Knowing that is useful too.
The point was never to defend the first draft because you’ve already spent time producing it. The point is ending up with something the evidence can actually hold up.
Better AI Systems Should Help Us Disagree With Them
A lot of business tools still seem to begin with some version of what would you like to create? That’s useful when production is genuinely the problem. Often it isn’t.
I’d rather a system knew what I was trying to accomplish, what I’d already tried, what evidence mattered, what assumptions were still shaky and which decisions I hadn’t delegated. I’d also like it to remember why something was done, not just the final choice.
That starts to matter once AI is working with you over longer periods instead of appearing for one prompt and disappearing again. If a system recommends changing a page next month, I want to know what happened the last time we touched it — what improved, what failed, and what we deliberately left alone because changing it would have been stupid.
Otherwise we end up with very sophisticated software rediscovering last Tuesday.
The Dashboard Still Has a Job
None of this means dashboards are useless. We still need tools, analytics, publishing systems, research, training, communication and somewhere to see what’s happening. I just don’t think the dashboard should be the beginning of the thinking.
Analytics are useful once I know what I’m trying to learn from them. Automation earns its place once we’ve thought about how much authority it deserves. And I’ll take an AI recommendation gladly, provided I’ve got enough context to actually disagree with it.
The toolbox can stay. Just don’t empty it onto the floor and call that a strategy.
Working This Way Now
You don’t need some future platform to work this way. Start with the actual problem rather than reaching for the tool first, and let AI do as much of the digging, sorting, comparing and drafting as makes sense. I see no prize for doing manually what a machine can do perfectly well.
What I don’t want to do is confuse how much work I’ve handed over with how much authority I’ve handed over. If an answer matters, I want to check it somewhere outside the answer itself. And the more it costs to be wrong, the less interested I am in treating a plausible-looking response as sufficient.
I’d also keep some record of what happened. That part is dull enough that people skip it, which is probably why it matters. Write down what worked and what didn’t. If AI gets something wrong and you fix it, keep the correction instead of quietly cleaning up the evidence that the mistake happened.
After a while, patterns start showing up. Certain sources hold up better than others, certain checks keep catching real problems, and some of the human review you thought was essential turns out to contribute almost nothing. At that point, giving the system more room isn’t an article of faith about autonomous AI. It’s a decision based on what you’ve actually seen work.
The Next Version Starts With a Decision
I don’t know exactly what the next generation of online-business software will look like. Anyone who says they do has more confidence than evidence.
But I do think the problem has changed. Access to tools once limited what ordinary people could build online. Now the harder part is increasingly deciding what deserves to be built, what can safely be delegated, what needs checking and where human judgment still earns its place.
A dashboard can help manage the work, and AI can do an astonishing amount of it. Neither one gets to decide, simply because it’s capable, whether the work was worth doing in the first place.
And for now, I’m quite comfortable leaving that problem with us.
