My AI stack keeps getting smaller

I counted my subscriptions and nearly fell out of my chair

August 18, 2026

In partnership with

I have a confession. I sat down one afternoon and added up every piece of software I was paying for, and the number was embarrassing. Half of those tools I had not opened in months. Some I could not remember signing up for at all.

So when I came across a post from an AI creator who had done the opposite of what everyone else does, who had ruthlessly cut his stack down to the essentials instead of adding to it, I stopped scrolling and read every word.

What I love about his approach is the honesty. He is not chasing the shiny new thing. He is naming the five tools he actually opens every single day, and explaining why each one quietly replaced something bigger and more expensive. His list got smaller over the years, not bigger. That is rare enough to be worth studying.

Blu Dot surpasses 2,000% ROAS with self-serve CTV ads

Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:

After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.

The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.

“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”

Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.

*Ad

He uses ChatGPT for pictures, and for nothing else

Here is the twist that made me sit up. The creator does not use ChatGPT for writing or for work at all. He uses it purely to make images. Infographics, quick visuals, the kind of edits he would normally open a heavy design app to handle. Photoshop is simply gone from his life.

His favorite move is the built-in edit mode. Once you generate an image, you do not start over to tweak it. You click on the image, ChatGPT drops into an edit mode that is genuinely easy to use, and you say what you want changed while staying on the high quality setting.

One tip he stresses is the part most people skip. If you are making an infographic, grab a reference image first, from Pinterest or anywhere else with a look you like. He says the prompt does almost nothing without a visual reference to anchor the style. I think that single detail explains why so many AI image attempts come out looking flat and generic. People are describing a style to a model that has never seen the one in their head.

The tool that fixed his laziness, not his calendar

This one surprised me more than any of the others. He uses Wispr Flow, a dictation tool, and he does not frame it as a Siri replacement. He frames it as a fix for a very human problem, which is that we under-prompt our AI because typing long, detailed instructions is tiring.

So he talks instead of typing. When he is speaking, he gets into what he calls a yapping flow state, dumping a big, rich chunk of context by voice in seconds. The answers come back faster and better, and as a bonus he burns fewer tokens, because he front-loads everything at once instead of dragging the model through a long back and forth.

The reasoning underneath is the part worth keeping. The quality of what you get out of an AI depends almost entirely on the quality of what you put in. Voice removes the friction that keeps most people typing lazy one-liners and then blaming the model for a lazy answer.

Winning Startups Aren't Bigger. They're Leaner.

Founders deploying AI agents across sales, marketing, and customer service are closing more with smaller teams. 65% more sales leads. Less headcount.

Download the free Practical Guide to Agentic GTM for Startups and start building the stack your competitors dream of.

*Ad

One model quietly eating a whole shelf of software

Once images are handled elsewhere, he reaches for Claude for basically everything else, and he argues it replaces far more than other chatbots.

The list he gives is genuinely wide. Finance tasks and legal cleanups. Design work. A coding tool, connected out to services like Netlify and Supabase. Email drafting, once it is connected to Gmail. A personal assistant, once it can see his call notes. Building spreadsheets from scratch, starting at an empty file.

The pattern there is worth noticing, and it is not really about which model won. It is that one capable AI, connected to your actual accounts and tools, absorbs the jobs you used to spread across five separate apps. The connections are what turn a chat window into a coworker. Without them you have a very smart thing in a box.

The boring choice he refuses to apologize for

He is short and blunt about this one. He is fully on Google Docs and Sheets, and he admits he feels a little bad for people still wrestling with Word, Excel, and Teams.

The logic is simplicity and speed. Cloud first, always synced, easy to share, and it plays nicely with everything else in his stack. No heavy installs, no folder full of files named final, final2, and final_actually_final.

I do not think the specific vendor is the lesson here. The lesson is that the document layer should be the least interesting decision you make, and if yours is generating friction, it is quietly taxing everything above it.

Stop typing what you could say in 10 seconds.

Wispr Flow turns your voice into clean, professional text inside any app. Emails, Slack, client updates — speak once, send without editing. 4x faster than typing.

*Ad

Twenty slides, because eight is never enough

For presentations he skips Google Slides and PowerPoint entirely, and reaches for Gamma. His reason is the Templates feature, which gets you to a polished, professional design almost instantly.

The workflow he shared is small enough to steal today. Open Gamma, go to the left menu, search for Templates, and scroll through them. Pick one, hit Create from template, then prompt it. The prompt he uses is simply to make twenty slides on the topic.

That slide count is the clever bit. He notes that if you do not specify a number, Gamma tends to fall back to eight slides, which is almost never enough for anything real. As he puts it, the world still runs on slides and emails in 2026, so getting a deck done fast is a genuine edge, not a vanity metric.

The whole list, and the honest footnote

Laid out plainly, his daily stack is ChatGPT for images, Wispr Flow for dictation, Claude for everything and then some, Google Workspace for documents and spreadsheets, and Gamma for slides, plus a meeting notes tool feeding transcripts into the pile.

Then comes the footnote I respect him for including. He still pays for twenty-plus other tools in the background. This is not a story about a man who cancelled everything. It is a shortlist of what he opens every day, and everything else is infrastructure he barely thinks about.

That gap between what we pay for and what we open is the real number. Most people have never measured it.

The line that stuck with me

His closing thought is the one I keep turning over. His stack keeps shrinking, not growing. He wonders out loud whether in three years he will be paying for exactly one tool, and whether that tool ends up being Google or Claude.

I do not know the answer, and I am not sure it matters. What matters is the direction. The goal was never to collect more AI toys. It is to find the few that quietly do the work of many, and then cut the rest loose without ceremony.

So here is the thing you could do tonight, and it takes about ten minutes. Open your billing history, list every subscription you are paying for, and mark the ones you have actually opened this month. Whatever is left unmarked is your answer.