Is Your AI Getting More Expensive to Use? Time to Give It a Checkup

A few days ago, a friend who runs an e-commerce business found me to vent.

He said his AI bill kept climbing the more he used it. Yet he wasn't getting more work done, and the output wasn't any better.

I asked him: how many skills and how many plugins have you installed—do you remember?

He froze. After thinking for a long while, he said: no idea. Quite a lot, anyway.

That's the problem.

Your AI Workspace Is Quietly Gaining Weight

What does it mean for a workspace to gain weight?

When you first got started with AI, the toolbox was clean. Then you added skills, built projects, connected apps, configured agents—each one handier than the last.

Six months on, you can't even say what's piled up in there.

Doesn't it remind you of your phone? Buttery smooth when it was new, sluggish after a year. The phone didn't change—what piled up in the background did.

Your AI workspace is the same. And none of what's accumulated just lies there quietly. It's working. Doing exactly the jobs you never asked it to do.

Think about it:

Six months ago you built 3 skills, each with its own job—or so it seemed at the time. Looking back now, they all do the same thing. Say one sentence, and you might wake up 3 employees doing duplicate work at once.

Then there's the artifact (an interactive document or dashboard you build inside Claude) you set up to keep an eye on your data, refreshing itself every single day. But how often do you actually look at it? Once a week. So whose tokens are those other 6 refreshes burning?

And a pile of project folders full of stale context. Your business moved on long ago, but these "memories" are still stuck six months back. The AI reads them diligently every time, uses them faithfully, and quietly pulls you off course.

Sneakier still: skills fire when they shouldn't, and scheduled tasks burn money around the clock on things you stopped caring about long ago.

Every single one is stealing your money and slowing you down.

What to Do? Let AI Run the Checkup on Itself

The fix, in one sentence:

Set aside one day each month and let AI audit your entire AI system.

That's right—no manually combing through configurations yourself. Let AI inspect AI.

You'll dig up a bunch of stuff. Correction: a pile of junk.

Three duplicate skills? Delete. An artifact idling through daily refreshes? Make it weekly. Project context six months out of date? Clear it out.

It's like servicing a car. Your car won't remind you to change the oil; ignore it for too long, and one morning it leaves you stranded.

By the way, the barrier to building these automations is now absurdly low. You don't need to learn to code—just describe what you want in plain language. Take Claude Cowork: it can plan its own steps, remember context across sessions, and connect to the tools you're already using.

The tools hurdle? Basically cleared.

But Will a Clean System Make Your Content Good?

No.

I recently came across a case I really liked. AI strategist Nicky Saunders built her own "AI creative director"—always on, 24 hours a day. It knows her voice, her style, and the kinds of topics that land best with her audience.

What does "know her" actually mean?

It's not that she said one line to the AI—"please imitate me." She laid the foundation first.

A lot of people use AI for content like this: open the chat window, type "write me a post," and get back something everyone who sees it will scroll right past.

Why? Because you haven't fed it anything. So all it can do is fob you off with the average.

Nicky's approach is to first teach the AI two skills:

First, teach it how you speak.

Second, teach it how you write.

Note that these are two different things. The way you talk out loud and the way you put pen to paper are completely different. And on different platforms, your tone shifts again. Teach both of these properly, and the AI's first draft sounds like you.

With the foundation laid, we get to my favorite part.

One small daily habit feeds an entire content pipeline.

Every day she speaks a few passages into her phone—whatever comes to mind, like keeping a journal, except she talks instead of types. The practice comes from an old classic on creativity.

Those voice notes are her library of raw material.

Then a scheduled workflow takes over: a single morning "brain dump" automatically becomes topic ideas, visual assets, and publish-ready drafts for every platform.

Tweets, newsletters, video scripts—all of it grows out of those few voice recordings.

She almost never "sits down to create content." She just talks every day. The AI does the rest.

The system is called DraftLoop.

A Few More Things Happening in This Space

That was the do-it-yourself part. Now a few platform-side developments, all relevant to those of us doing marketing with AI.

First: ChatGPT ads have arrived in Europe.

OpenAI has rolled ChatGPT Ads out across 31 European markets. The product started testing in the US roughly six months ago; now it's officially going international.

Where do the ads run? In the conversations where you're researching, comparing prices, and making decisions.

The business logic is blunt: free users see ads; paying Plus, Pro, and Enterprise users are buying peace and quiet.

A note for advertisers who want in: for now, you have to go through OpenAI's ad solutions team and designated partners. The self-serve ad console won't open until late summer, according to OpenAI.

Second: ChatGPT can now manage your iMessage.

ChatGPT has connected to Apple Messages. Using natural language, you can have it search chat threads, dig up old messages, draft replies—even send messages or delete them.

The same capability extends to Codex and ChatGPT Work, giving business follow-ups and replies a new playbook.

Rest assured: it only reads messages when you ask, and it processes them locally on your device. AI-drafted messages still need your nod before they're sent.

Third: Google hands the choice back to readers.

Google has launched an embeddable "Preferred Sources" button. Readers use it to designate the outlets they favor, and from then on, that outlet shows up more often across Search, Discover, Google News, AI Mode, and AI Overviews.

For people who make content, this could be a lifeline. AI is reshaping search behavior, and traffic keeps getting harder to earn. Now readers can ask for you by name.

By the way: Discover now supports natural-language customization, and Google News audio briefings understand you better, too.

Fourth: OpenAI turns "admin work" into a conversation.

ChatGPT Work and Codex have gained an Admin plugin. Admins no longer dig through settings menus—they just say it: check usage, manage members, adjust permissions, set spending caps.

It can also run routine workflows automatically—for example, routing approval requests to Slack or Teams, with access granted automatically once conditions are met.

Of course, it works within your existing roles, permissions, and approval frameworks. It doesn't overstep.

Finally, Back to That Friend from the Start

After hearing all this, he went home and spent an afternoon giving his AI a health check, then deleted half of everything.

The next day he told me: it's faster now. And cheaper, too.

See, in the AI era, tools keep getting cheaper and more powerful—everyone can afford them. What actually sets you apart is how you feed yours: run regular health checks so redundancy can't steal from you; lay a solid foundation so the AI can't fob you off with averages; then build the daily habit of talking, and let the content grow itself.

Everyone has the tools. Your voice is the moat.

Here's wishing you, too, an early health check for your AI.