Ask HN: What useful AI tools do you use every day? | Hacker News Reader

Ask HN: What useful AI tools do you use every day?

42 pointsrajkumarsekar72 comments

There are thousands of AI tools launching every month, but very few become part of our daily workflow.

I’m curious, what AI tools or features do you genuinely rely on every day? This could be anything from coding copilots and writing assistants to niche productivity tools, automations, or personal hacks using LLMs.

View on HN

Something never thought I would say: google AI previews. They actually helped me a lot during Iran's internet shutdown last week. I wrote a blog post about it: https://ahrm.github.io/jekyll/update/2025/06/20/iran-interne... .

damn; this article triggered my dormant frustration over Google sunsetting Webcache. Unbelievably useful feature that went away kind-of suddenly for weak reasons.

Deepl Write. It helps me improve my lousy German by showing me a better way to write what I came up with.

Google Lens to identify plants.

BirdNet is Shazam for birds.

ChatGPT for vague questions about everything. I feel like a child asking mom about the world again. It reduced the friction of curiosity.

It’s pretty wild that I have a real world Pokedex.

NotebookLLM. Load in technical manuals and datasheets, then ask it questions and check the parts it references instead of searching tens of thousands of pages across dozens of diferrent documents. It's the most useful AI tool I've tried so far for embedded work.

Isn't Claude (i.e. general LLM) enough? I seem to have luck having it interpret datasheet PDFs for me.

1. ChatGPT I use it mostly to dump quick ideas and then expand them later by using it as my thinking partner. Also I use chatgpt for brainstorming ideas for marketing campaigns, writing copies etc. I have few custom GPTs created for my every use case

2. Claude I use Claude projects for writing articles and for SEO optimisation

3. Cursor I use Cursor for coding daily. Now quite used to the flow it creates and makes my coding super fast

4. Sora and Gemini For image generation. Mostly I need that to share social media posts

All other AI tools come and go but these are 4 constants for me for last few months.

everything thats wrong with The Internet in one comment

why do you think so

If it works it ain't stupid (or wrong). I can imagine dozens of setups where such tools can bring massive velocity in small/single man team, ie indie/mobile game development

Your first point resonates with me.

I am often using a mobile device and away from my computer, but want to put something in my Obsidian notes or a Trello card.

ChatGPT speech rec is excellent (so much better than iOS), so often I will start a temporary chat, press the Mic button, and say something like :

“Take the following dictation and repeat it back to me without adding anything. Don’t alter the content, but clean it up if needed… “. Then I ramble on for a while, with pauses when I need to gather my thoughts for the next bit. Then, when I get the paragraph back, I long press for the copy option and paste it back into my other app.

How’s your experience been with Cursor compared to GitHub Copilot?

I have an actual rubber duck on my desk, which means I only use AÍ for stuff like “refactor and write tests for this stuff Ducky and I designed”.

But I think your question is fundamentally flawed—-we all use AI in phone and editor autocomplete, searching, summarization and whatnot on a daily basis now in office tools. It is using it for actual useful output that counts, and for that it is still below what I deem acceptable.

Fair point. AI is everywhere now, even when we don’t notice. I was aiming more at tools people actively choose for real output. Love the rubber duck method, by the way!

Kagi Translate.

Kagi Summarizer (via bookmarklet) to help weed out articles that won't be worth reading (even here on HN) or to get the one sentence that matters.

Kagi Assistant. This I use for random things, like pasting a webpage's source and having it figure out what to fill in on this form to make an RSS feed where none exists.[1] (This, in my current push to move most of my needed information updates both out of email and into RSS, as well as to reverse what I call "the arrow of information" [I want to batch-check information updates on a schedule instead of getting interrupted by incoming data piecemeal].)

Kagi Assistant (and the various LLM models) are also great when I realize that I don't understand a particular grammar or usage point in a foreign language that I'm learning: the models usually create clear explanations and then a few practice exercises for me to do.

[1]: https://feedmaker.fly.dev/

Current workflow that might stick:

I’m using SuperWhisper for voice dictation and casually chatting with Claude Code, which I just start in the Obsidian vault directory.

I’m currently onboarding into a new project and need to gather and structure information quickly. So I just jot down whatever comes to mind — in natural language, no structure.

Claude takes care of organizing it: generating #tags, creating [[links]], and making things retrievable later.

After a couple of weeks, I can ask things like:

• What did I do yesterday?

• Who is “John Doe”?

• Which project uses Java version 21?

• …

Happy to share more if anyone’s curious.

(English reviewed with a bit of help from ChatGPT — non-native speaker here.)

please do!

I'm experimenting with a somewhat similar workflow.

I have a Boox tablet I use to take handwritten notes on any relevant topics during the day. And with that I create the following workflow: 1. transform handwritten notes into a markdown document using chatgpt 2. with this document I ask chatgpt to create a summary of yesterdays activities and a list of unfinished things I still need to work on today.

This approach isn't perfect but is surprisingly useful.

Perplexity - for search (Google replacement), summarization and rewriting, basic research and making presentations (using Perplexity apps)

Granola - transcription and meeting notes, searching across notes, recalling action items

I've played around extensively with ChatGPT, but Perplexity now covers my use cases. I'm looking to test Claude, primarily because Perplexity does not currently support MCP servers, and I need an assistant who can answer questions across all my work files (Google Drives, Calendar, Slack messages, GitHub, etc.).

Have you tried google deep research? i'm curious how well perplexity compares to it.

I've been using Gemini Deep Research to replace my Google search since it does a web search and provides links you can check yourself to any citation that the resulting report uses.

Granola sounds super handy for recall and meeting notes.

Claude Code. It's so amazing, and not just for code. You can use it for pretty much all kinds of projects. See, for example, this post of a user sharing his automated workflow https://www.reddit.com/r/ClaudeAI/comments/1liylon/were_unde...

None that has come with the flood of LLM-based tools. I only use one language translator [1], but it was already of such good quality before the LLM wave that I didn't even notice the change under the bonnet (if there was one at all).

[1] https://www.deepl.com/en/translator

Travelling with ChatGPT is incredible. Translate this Japanese menu into romaji (latin character Japanese) and English. Explain the difference between Turkish yoghurt dishes haydari and cacik. Walking around Manila with voice mode talking me through some basic Tagalog phrases I might need.

>Explain the difference between Turkish yoghurt dishes haydari and cacik

wow something that you can find literally in the second sentence of the wikipedia article for haydari.

And the other, vastly more complex, stuff?

Btw, it's the third sentence.

Haydari is a type of yogurt dish similar to a thick cacık, made from certain herbs and spices, combined with garlic and yogurt.[1] It differs from cacık in that the recipe contains no cucumber and calls for strained yogurt or labne.[2]

it is in the second sentence.

https://en.wikipedia.org/wiki/Haydari

The third sentence is where it tells me how they taste different and what they are ordered for (meze vs accompaniment) which is what I wanted.

My answer from ChatGPT was much better, and I didn't have to gamble on it being in the Wikipedia article, or which one to search for (cacik has nothing).

So, anyway, about that other complex stuff?

I just think you're impressed by stuff which was already available and portraying it as stuff that is through chatgpt. Like a Tagalog phrasebook? That's been around a long long time.

Everything in an LLM was already available, that's how they work.

Short of hiring a human, there has never been a phrasebook I can talk to out loud and it talk back. That's what I did that day in Manila, earbuds in, walking and talking, acting out scenarios like ordering food in Tagalog with the assistant.

> literally in the second sentence of the wikipedia article for haydarI and cacim

It's impressive that you know where things are and how long it take to find the infomation even BEFORE hitting search, trully.

That’s an awesome use case, feels like having a multilingual travel buddy in your pocket.

- postgress/pgvector and ollama in a jupyter notebooks[1] to search my documents/projects and other data

- Mistral as search engine and for natural language processing (mostly summaries of something)

- Co-Pilot, mostly as autocomplete and sometimes to ask questions or let it write tests or do refactoring

- AWS AI Services to create tasks, tickets and stories from templates in JIRA or YouTrack.

- Github/Co-Pilot and Gitlab/snyk for code review and security analysis in PRs

[1] basically https://github.com/roskakori/wolkenlose-ki-fuer-zu-hause/tre... and now i try also https://github.com/ggozad/haiku.rag

Wow it’s amazing how hacker news which is primarily filled with technology type people has no few using ai or has no idea what they are doing with ai.

The saying "Can't see the forest for the trees" really makes a lot of sense now.

NotebookLM to explore sources in bulk, prepare vacations etc

Claude Code for coding and as basis for my AI assistant/ghostwriter (productivity, planning, prioritization, writing articles, newsletters, social media posts, etc

Gemini & Sora for images, also replicate and Flux (much less now)

Veo 3 for video (tests)

code2prompt to create AI Mega prompts out of my notes

LM Studio and ollama for local LLMs

Dockmaster for installing/running MCP servers. Hopefully soon Desktop Extensions

Various Obsidian extensions for AI

Oh and tons more stuff

None.

Very rarely I need a bash script or systemd service written from a command line, or just something where I know what to search for and what to replace it with.

Then I use Co-Pilot.

The Jetbrains code helper AI is 99% useless, also inconsistent.

Claude.ai + Claude Code + Overlay(https://overlay.one) The last one is built by me, a tool to interact with websites.

Augment Code (https://www.augmentcode.com/) extension in VSCode. The remote agent feature is fantastic and the local agent is worth buying credits for. I've tried almost every AI editor (Cursor, Windsurf, Roo/Cline etc.), tried CLI-based coders (Claude Code, Aider, Codex etc.), and have used them all with ton of useful MCPs and in the end, I've had the best results with Augment.

Tabs open right now: Chatgpt, claude, and grok.

Local LLM: deepseek r1 and qwen2.5coder; phi4 for my project.

Pycharm pro as my main IDE. I <3 pycharm.

I have been giving Void IDE a try this week but it's just frustrating me more than anything.

Has to be a bug where when i approve it to edit the file and it reads maybe 5 lines, stops, still says its running but isnt doing anything. then the chat keeps going like an error happened; but then it really errors out. I'm not ruling out that im just screwing it up somewhere.

For features: Transcript of a Zoom call, ask the LLM what you could do better, how you might be perceived, what the LLM thinks you wanted to achieve (vs. what you really wanted to achieve) etc. Gave me great insights and helps me every day.

Self-marketing: Started Marvai this week https://github.com/StephanSchmidt/marvai/ as an AI tool for installing useful Claude Code prompts

None day to day.

I may use a tool time to time, but I do not use anything daily (on purpose at least, as all software nowadays has AI running in the background that I do not care for).

Warp.dev and VSCode Copilot (I use Cursor et al too it could be any of them).

Warp because I can't remember many commands (copilot in the terminal works too but you lose agent mode). They've just release v2 today, looking forwards to try it as a free Claude Code.

Looking to switch to Dia or some agentic browser as main driver but at the moment content with Firefox and Grok as the default search engine.

I have my own personal digital assistant in the form of a telegram bot. Why telegram? Because it is in a group with me, my wife, and the bot. It manages our calendars, reminders, shopping list, and a few other things. When we think of something that might be useful, I add it. It's been great, and because it's on telegram, my wife uses it without issues.

How does it manage your calendars and shopping lists? Does he just print the events and shopping lists into the chat or does he have access to like a google calendar?

It fetches our calendars from different providers and injects upcoming events in the context. We haven't felt the need to do 2-way sync'ing yet. Reminders and shopping list are fully managed by the AI, they get stored in json files

Other than the obvious heavyweights, here are a few others:

1. notebooklm for deep-dive into any document

2. Notion AI for QA on my own documents (works really well)

3. cartesia.ai for very good and cheap audio generation

4. veed.io for automated shorts generation with voiceovers and background imagery

5. zenquery.app for data analysis on my large csv and json files

6. regrowth.so for building my own brand on twitter by copying others

7. syft.ai for news summaries (actually works)

Robot vacuum is the most useful for sure.

Nice! I haven’t tried one yet, but you’re making me curious, might be time to test if AI can finally tackle real-world dust :)

I use Google Gemini. Can't say it's super useful but it's only $15/month so worth trying. It's better than search for coding things, the image and video generation is cool for memes I guess. The code it produces is mostly useless but it's ok with concepts.

I like that it seems to read way more sources than other AI when deep research, but hate that it always assume "research" to mean a verbose report with much filler.

I don't use any generative AI tools at the moment (I don't even have autocomplete enabled on my phones; I'm a caveman, I know), but I use Tesla FSD, Cornell Merlin Bird ID and have robot vacuums and stuff at home.

i find grok far superior to chatgpt and copilot, but none of them worth more than $10/m. I am about to switch to agents and pro subscription level because I've heard good feedback about those. AI is perfect at small and easy tasks

are you paying for any of them or only using the free versions? and which models specifically are you using?

my contractors provide me paid versions and it's my choice which one to use. Most of the time I use them as a sanity check and autocomplete. Anytime I get to anything complex like writing rust - every AI is useless to different levels

PS. Maybe I am a bad example, since I don't even track with model was where. I tick different boxes and see which one works better. In general, not a good experience

there are a lot of mixed reviews on this tool but chatgpt still is one of my most used ai tools - from work to personal life.

i find it helpful to sort of help me make sense of things when i have lots of ideas and unable to structure it

I've even used it for emergencies - trying to locate a place while visiting another city and both maps and the place's insta guide failed .. it was pouring down there. (maybe that's why im partial lol)

i have also started enjoying notebooklm for contextual content generation

gemini - ive tried but it keeps disappointing. currently exploring perplexity

I’m learning German, and my listening comprehension particularly needs work. So I create mini stories on different topics using Claude, then ElevenLabs to do narration.

I do a very similar thing. I have Gemini create me a story based on a historical activity that happened on that day with an instruction to use A2 and below constructs.

Then I have it give me:

- A transliteration (I'm still getting used to the non-Latin alphabet)

- A list of vocabulary from the story

- Grammar tips

Then it is emailed every day around lunch.

I need to go back and tweak it, though, because Gemini really likes starting stories off saying that the sky is clear and the sun is shining.

Cool! Does Gemini support emailing, or is this handled separately? I've been interested in creating something similar, but haven't devoted the time.

It's done in Lambda; with the email template, script for sending via Resend, and script for packaging for Lambda all done with V0.

I'm not the most versant AI user so far, I was wondering, what is currently the best tool for asking math questions (lessons, proofs, examples) ?

claude code for writing code, chatGPT for everything else (brainstorm ideas or writing down documentation) and Cringe Guard (https://github.com/pankajtanwarbanna/cringe-guard - i've built this one to avoid cringe content on my feed)

VSCode with Copilot using Claude Sonnet 4 as the model, running in Agent mode.

Is there a way to have that run automatically where it does not require clicking continue each turn?

See https://code.visualstudio.com/docs/copilot/chat/chat-agent-m...

These replies would be more useful if they included the subscription(s) cost.

Surprisingly, many people in this thread use no AI tools at all...

My company has a specific format for weekly updates, and, frankly, it's a pain to put together.

So what I do is I take notes in a doc throughout the week (Obsidian periodic weekly notes) and just send all of it to a custom GPT that creates the update for me. I usually then spend ~10 minutes cleaning it up.

I'm generally allergic to having an LLM write for me, but seeing as it _must_ be in the requested template and ChatGPT saves me a couple of hours, I let this case slide.

duck.ai

Duckduckgo's llm offering with no sign in required and more privacy.

No idea how I missed this one. Thanks, its quite cool!

cursor ofcourse, the tool I’ve paid most in my career

Goblin.tools

Amp aka https://AmpCode.com. I’m one of the core engineers building it - happy to answer any questions. We built Amp with Amp. So I guess I rely on Amp ;)

My own AI agents platform, AI Construx (https://aiconstrux.com). The MVP should be released in a few weeks time.

Reply on news.ycombinator.com