The AI tools I actually use every day, and the ones I deleted after a week.

Here is my real stack, split three ways: what I use daily, what I keep only for clients, and what I deleted because something cheaper or free already did the job.

Which AI tools do I actually use every day?

A small core, each doing one job I could not do as fast without it.

Claude is the one I open first. It is where I think, brainstorm, research, organize, and write. The reasoning holds up on complex work and the writing does not read like a machine wrote it, which matters when the whole point is to sound like a person. ChatGPT is still in my stack, but I demoted it. I stopped using it for writing because the output feels too obviously AI-generated, and I now use it for what it is genuinely best at for me: prompting and image generation. Perplexity handles deep research, the citation-backed kind where I need sources I can trust rather than a confident guess.

Around that AI core sits the working stack. For design, Photoshop and Illustrator for the real work, Canva for speed, and Ideogram, Pinterest, Behance and Dribbble for finding direction. For video, Premiere Pro and After Effects for the heavy edits, CapCut and VN for fast turnaround. GoHighLevel runs the CRM. Notion runs tasks. For automation I use a trio, Zapier, n8n, and Make, plus GoHighLevel's own, because no single one covers every job cleanly. ElevenLabs handles audio when I need a voice. Lovable builds websites fast.

That looks like a lot, but notice the shape. One tool for thinking, one for research, one for images, and then specialist tools for the crafts I actually practise. Nothing overlaps. The moment two tools started doing the same job, one of them left, and that is the whole discipline.

Which AI tools do I keep, but only for clients?

The expensive, high-end ones that do not earn their price for my own work but absolutely earn it on a client budget.

Midjourney, Higgsfield, and HeyGen sit here. They make genuinely excellent images, video, and avatars. But for my own content the cost does not justify the use, so I do not keep them running for myself. When a client needs that specific quality and their budget supports it, I use them on the project. That distinction matters more than most keep-or-cut lists admit. The question is never just “is this tool good.” It is “is this tool worth its price for whose work.” A tool can be a clear delete for you and a clear keep for a client in the same week.

Which AI tools did I delete, and why?

The ones that turned out to be duplicating something I already had, or solving a problem I could solve for free.

Descript went first. It is a capable video and audio editor, but it was overkill for how I work. I rebuilt its core workflow from free pieces: extracting audio with a free tool, transcribing and pulling subtitles with Maestra, then importing the subtitle file into whatever editor I am already in. Same result, no subscription. Copy.ai went because it duplicated what Claude already does for me, at a higher price for a narrower job. Grammarly went for the same reason, duplication, I use Quillbot for that need instead. And HeyGen, for my own content, lost to the simplest alternative there is: a real person actually filming. AI avatars still need heavy editing and they read as slightly off, and for a personal brand, authentic beats fast.

Notice the pattern in every deletion. None of them were bad tools. They were redundant, or overpriced for the job, or beaten by something free. That is almost always why a tool should go, and it is a much more useful test than chasing whichever one is newest.

The tool that actually burned me

Perplexity's top enterprise tier is the one that cost me a real lesson.

It is the heavy tier, listed at three hundred and twenty five dollars per seat per month, and it is genuinely powerful if you run research at volume with a whole team leaning on it. That is exactly the point. It is built for mid-to-large enterprises running Perplexity as core infrastructure, not for a small team or a solo operator. I found that out the way you always find it out, by paying for it and then watching how little of that capacity I actually used. The tool was not the problem. The mismatch between its scale and mine was.

That is the quiet scar nobody warns you about. You do not learn a tool is dead weight from a review. You learn it after you have paid for the premium months to test it properly, which is the only way to really know. The cost of building a good stack is the money you spend on the tools that do not make it. There is no shortcut around that, only a discipline for making the cuts fast once you know.

So how do I decide what stays and what goes?

One process, and it is the whole spine of how I keep the stack lean.

I test everything on the free version first. If it looks like it could add real value, I pay for a few months of premium and use it on actual work, not a demo. If it delivers consistent value over that stretch, it stays. If something else does the same job cheaper, I switch without loyalty. And there is one more test that is specific to running an agency: if a tool is priced so high that a client would be better off having us do the work than buying and learning it themselves, that tells me something about where the real value sits.

The industry has landed in the same place, for what it is worth. The current consensus is that a professional AI stack runs somewhere between forty and two hundred dollars a month, that the winner is not the person with the most tools, and that if two tools do the same job you cut the more expensive one. My three hundred dollar research seat was a direct violation of that, and the market agreed with the lesson after I had already paid for it.

Building an AI stack is not about collecting tools. It is about removing friction, one bottleneck at a time. The tools change every few months. The discipline does not.

If you are carrying a stack of AI subscriptions right now, do one thing this week. Open your billing and list every AI tool you pay for, then mark the ones you have actually used in the last seven days. Whatever you have not touched is the tool to cancel first. You will almost certainly find at least one seat you are paying for and not using, and cutting it is the fastest money you will make all month.