There's a new AI tool dropping every other day. Most of them I try once, get mildly impressed, and never open again. Some I actively deleted — not because they were bad, but because they added friction instead of removing it.
Here's what actually survived my workflow. No affiliate links. No sponsored mentions. Just what I'm using right now, in April 2026, and why.
The 5 That Stayed
1. Claude (Anthropic)
I use Claude for almost everything that requires extended reasoning — drafting, editing, figuring out why a piece of code doesn't work, thinking through a decision. The thing that makes it different from most tools isn't raw intelligence. It's conversation. Claude actually remembers what you said three messages ago and builds on it. It doesn't drift.
For me it's usually open in a tab from 9am until I close my laptop. I use Projects to keep context persistent across sessions so I'm not re-explaining myself every time.
When I use it: Drafts, research, debugging, working through decisions I'm stuck on.
2. Perplexity
I stopped Googling things. Not dramatically — I didn't announce it — I just noticed I was getting better answers faster from Perplexity. It cites its sources, so I can actually verify things. And it doesn't serve me a list of SEO-optimized articles I have to triage before finding the thing I wanted.
The thing I like most: it gives me a direct answer, then shows me where it came from. It's how Google search should have evolved.
When I use it: Factual lookups, research that needs citations, anything where I'd have previously opened 6 browser tabs.
3. Cursor
I write code. Not every day, but regularly. Cursor is VS Code with Claude and GPT-4 built in — you can highlight a block of code and ask it to explain, fix, or refactor it without leaving your editor. Tab autocomplete that actually understands what you're building, not just what you've typed.
The real value isn't the autocomplete. It's that it reduces the context-switching tax. I used to code → get stuck → open browser → ask ChatGPT → copy back into VS Code. Now that loop is compressed into a single conversation inside the editor.
When I use it: Any coding session.
4. Notion AI
I live in Notion already. All my notes, projects, writing drafts — everything's there. Notion AI is just... already in the place where the content is. I use it to summarize meeting notes, clean up rough drafts, and generate first outlines when I know what I want to say but haven't said it yet.
It's not the most powerful AI. But placement wins. The best tool is the one you actually use, and I'm already in Notion.
When I use it: Summarizing, cleaning up notes, first-draft outlines.
5. ElevenLabs
I generate voiceovers. For a small project, for demos, occasionally just to hear something back. ElevenLabs is still the best I've found for this — the voices sound like actual humans, not the text-to-speech robots we grew up with.
I don't use it every week every week, but it shows up when I need it and delivers every time.
When I use it: Narration, demos, anything that needs a voice but doesn't need me on camera.
The 3 I Deleted
Jasper
I wanted Jasper to work. It's purpose-built for marketing copy, it's polished, and the feature set is impressive on paper. But it always produced content that sounded like content. You know the kind — technically correct, covers all the points, completely forgettable. I could never get it to match my voice, no matter how many settings I tuned.
If you're running a team and you need consistent-enough output at scale, Jasper might be right for you. For solo work where voice matters, I found it added revision time instead of saving it.
Midjourney
I subscribed to Midjourney for about three months. The images it makes are genuinely beautiful. But I kept running into the same problem: I couldn't get it to make the specific thing I was picturing. I'd spend 40 minutes iterating on a prompt and end up with something impressionistic that wasn't quite right. For professional output, I still needed a designer.
I moved to DALL-E 3 inside ChatGPT for quick ad-hoc images where I don't need perfection — it's faster to iterate with because I can describe it conversationally. And for anything real, I hire someone.
Otter.ai
I tried Otter for meeting transcription. It works fine for that exact use case. But I discovered I didn't actually go back and read those transcripts. I was capturing notes and then ignoring them, which meant I was just adding a step for no return. I switched to building a habit of taking manual notes in Notion during calls instead — I retain more and I'm not creating a pile of unread transcripts.
The Pattern I've Noticed
The tools I kept all have something in common: they remove steps instead of adding them. Claude is already where I'm thinking. Perplexity is faster than search. Cursor lives in my editor. Notion AI is in my notes.
The tools I deleted were powerful but required a separate workflow. They were additions, not integrations.
If you're evaluating AI tools right now — and there are a lot of them — I'd think about that question first. Does this fit where I already am, or am I learning a new place to live?
I share what I'm learning about AI every few days at denismoroz.ai. If this was useful, the newsletter goes deeper.
