The AI tools I actually open every day

By Kasra Farivari · May 6, 2026 · 6 min read

I've used a lot of AI tools over the last two years. Most of them got tried for a week, used inconsistently for a month, and then quietly abandoned. The ones I'll write about here are the ones I open every working day, often multiple times a day. The list is shorter than you'd expect, given how much I write about this stuff.

This post is descriptive, not prescriptive. I'm not telling you what to use. I'm telling you what I use, why, and where each one earns its keep.

Claude — the everyday writing partner

This is the one I open most. I use Claude for:

Drafting PRDs from a context dump (covered in detail in an earlier post on PRDs).
First-pass user research synthesis (detailed here).
Rewriting emails that are too long, too defensive, or too curt.
Sanity-checking my reasoning before I send something risky.
Asking dumb questions I'd be embarrassed to ask a coworker.

What I don't use it for: anything I need to be confident is factually correct without checking. Anything that requires up-to-the-minute information. Coding (I dabble, but I'm not an engineer).

The reason I default to Claude over the alternatives, for writing: the prose feels less generic. It pushes back. It will tell me my framing is off rather than dutifully producing the deliverable I asked for. That conversational feedback loop is more valuable to me than any individual output.

ChatGPT — the lookup

I use ChatGPT mostly for fast lookups and quick rewrites. The kind of stuff where I want a competent answer and I don't need a thinking partner. "What's the standard structure of a vendor MSA?" "Rewrite this in 30% fewer words." "Explain X in plain English."

Why ChatGPT and not Claude for these? Honestly, mostly muscle memory and the search integration. For pure utility lookups it's a solid pick.

I keep both apps installed and use them differently, the same way I use Notes for personal stuff and a separate tool for work writing. The tools are similar; the contexts are not.

Notion AI — in-doc help

I write inside Notion all day, and the in-doc AI is genuinely useful for short tasks: summarizing a long doc into a TL;DR, expanding a bullet list into prose, or rewriting a paragraph in a different tone. The advantage isn't capability — it's friction. Highlighting a paragraph and hitting a hotkey is much faster than copy-paste-into-Claude-paste-back.

I don't use it for serious drafting. The output is fine, but it's not better than what I'd produce myself, and I lose the "thinking partner" loop I get with Claude. For "make this slightly better in place," it's perfect.

Gemini — image and quick visual stuff

I'll occasionally use Gemini for image-related tasks — generating placeholder visuals for a wireframe, getting a rough icon idea, looking at a screenshot and asking what's in it. I don't use it as a primary writing tool; the prose feels stilted to me compared to Claude.

It's also better than the others, in my experience, at multi-modal tasks. If I have a screenshot of a competitor's app and I want a structured description of the UX flow, Gemini handles that well.

Otter / transcription tools

Not strictly AI in the LLM sense, but the same wave. Transcription has gotten cheap and good enough that I record almost every customer call I'm allowed to record, and I get clean transcripts back within minutes. That alone has changed how I do user research more than any single LLM tool.

The catch: transcripts have errors, and you have to read them with the audio handy if anything important is on the line. The 95% accuracy claim is real but the 5% that's wrong is often the part that matters.

The tools I tried and dropped

For balance, a few that didn't stick:

AI meeting note tools that auto-summarize standups. The summaries were fine. They didn't tell me anything I didn't get from being in the meeting. Net negative because I started skimming the meeting expecting the summary to bail me out.
AI roadmapping tools. A category that seems to assume the bottleneck in roadmapping is producing the document. The actual bottleneck is the political alignment, which AI doesn't help with.
AI competitive intelligence dashboards. Pulled features from competitor products and tried to synthesize them. Output was rarely better than five minutes of manual digging on the competitor's app store page.

If anything, the pattern in what I dropped is: tools that automate a workflow where the output document was never the bottleneck.

What I expect to change

A year from now this list will look different. My guesses, with low confidence:

Voice-first interaction with Claude or ChatGPT will be a bigger part of my flow. I already do this when I'm walking and don't want to type.
Some category of "agentic" tool — running long-running tasks across my docs, calendar, and email — will be useful enough to make this list. Right now most of them are demos that don't survive contact with my real workflow.
Image tools will be standard enough that I won't write about them as a separate category.

We'll see. If you want to nerd out about any of this, my LinkedIn is the best way to reach me. And if this is your first post here, the intro post explains what the rest of this blog is about.