AI Applications
New AI Apps Worth Your Time Right Now (And the Ones to Skip)
A practical rundown of the newer AI apps people are actually using in 2025, grouped by what they do: agentic browsers, coding agents, voice notetakers, video and image generators, and on-device assistants. Includes a test for separating durable tools from thin wrappers, and what to do once you have picked two or three.
CSEWhy ·
A practical rundown of the newer AI apps people are actually using in 2025, grouped by what they do: agentic browsers, coding agents, voice notetakers, video and image generators, and on-device assistants. Includes a test for separating durable tools from thin wrappers, and what to do once you have picked two or three.
The new AI apps actually worth installing
If you want the short answer: the genuinely new category in 2025 is agentic apps, tools that do multi-step work on your behalf instead of just answering. The ones people keep using after week three fall into five buckets, listed below. Everything else is a chat box with a new logo.
I have tried most of these. A few I dropped within days, which I will say where it applies. The list is deliberately short because a list of fifty AI apps is useless to you.
One thing to notice as you scan the table: almost every app here is doing something a human assistant would do, not something a search engine would do. That shift is the whole story of the last eighteen months. Chat answered questions. Agents take actions, and they fail in new and interesting ways when they do.
| Category | Apps to know | What it actually does | Honest catch |
|---|---|---|---|
| Agentic browsers / computer use | Comet, Claude in Chrome, OpenAI's Operator-style agents | Reads pages, fills forms, books, compares, clicks through multi-step tasks | Slow and occasionally confidently wrong. Always watch the first few runs |
| Coding agents | Claude Code, Cursor, Windsurf, GitHub Copilot Workspace | Writes, edits and tests code across a whole repo, not one file | Brilliant for prototypes. Still needs someone who can read the diff |
| Meeting and voice capture | Granola, Fathom, Otter, Fireflies | Records, transcribes, pulls out decisions and action items | Everyone ships this. Pick the one your team already has, not the newest |
| Video and image generation | Sora, Veo, Kling, Midjourney, Freepik, Higgsfield | Text or image to short video, product shots, ad variants | Clip length and consistency are still the ceiling. Great for hooks, not films |
| On-device assistants | Gemini on Android, Apple Intelligence, Copilot in Windows | Summarise, draft and act inside apps you already use | Often good enough that a separate app becomes pointless |
The ones I would actually open tomorrow morning
Granola, because it does not join your call as a bot. It listens on your laptop, you type rough notes, and it merges both into something you would actually send to a client. For anyone doing five calls a day, this is the highest ratio of saved time to setup effort on this page.
Claude Code or Cursor, if you write any code at all. Not because they autocomplete, but because you can hand them a vague task and they will open files, make changes across the project, run tests and come back. A student who uses this well ships a working project in a weekend. A student who uses it badly ships something they cannot explain in an interview. Both outcomes are common.
NotebookLM is the quiet one. Upload your PDFs, lecture slides, a messy folder of research, and ask questions that are answered only from those sources. For exam prep and for anyone who has to read a lot and remember it, it beats a general chatbot because it will not invent a citation to please you.
For video, Veo and Kling have reached the point where a founder can produce a decent eight second ad cut without a shoot. Realistic expectation: you will generate twelve versions and use one.
How to tell a real app from a wrapper in five minutes
Most new AI apps are a prompt, a payment page and a waiting list. You can filter them fast. The question is not whether it uses GPT or Gemini underneath, because almost all of them do. The question is whether it owns anything the model does not already give you free.
- Does it hold your data? An app that remembers your files, your repo, your calendar, your past outputs gets better the longer you use it. A wrapper restarts from zero every session.
- Does it take an action, or just produce text? Booking, posting, editing a file, raising a ticket. Action means real engineering behind it.
- Can you reproduce it in ChatGPT in two prompts? If yes, do that and keep the subscription fee.
- Is there a free tier that lets you reach a real result, not a teaser? Confident products let you finish one job before asking for money.
- Who is accountable when it is wrong? Serious tools show sources, diffs or logs. Wrappers show a paragraph and hope.
Installing apps is not the same as having AI skills
Here is the part most people get wrong. Downloading eleven AI apps feels like progress and produces almost none. The person who gets paid more is not the one with the longest tool list, it is the one who can take a real workflow, decide which part should be automated, build it, and show what it saved.
A concrete version of that: a sales team spends four hours a week turning call recordings into CRM notes and follow up emails. The app that transcribes the call is maybe fifteen percent of the solution. The rest is deciding what a good follow up looks like, connecting the transcript to the CRM, handling the calls where the agent gets it wrong, and convincing the team to trust it. That work is the skill. The app is the easy part.
So pick two apps, not ten. Use them on actual work for a month. Then try to rebuild one of your own manual processes end to end, and notice how much of it the app does not cover.
If you want that practice with structure and feedback rather than alone on a weekend, that is exactly what the AI Creator Fellowship is built around: eight weeks of building with these tools, on real projects, with people reviewing what you ship. And if the problem you are staring at is a business process rather than a career move, AI Automations is where teams bring their actual workflow and get it built.
What to do this week
Open your calendar for last week and find the task you repeated most. Then pick one app from the table above that touches it directly. Install it, use it on real work for five days, and keep a note of where it broke.
That note is more valuable than any list of new AI apps, including this one. It tells you precisely where the gap between what AI can do and what your job needs still sits, and that gap is where your next skill lives.
Want help deciding which AI skills are worth your next three months? Tell our team what you do and what you are trying to get to, and we will point you at the right path, program or not.
FAQs
1. Which new AI app is best for students?
NotebookLM for studying, because it answers only from the sources you upload and cites them, which makes it safer for exam prep than a general chatbot. For project work, pair it with a coding agent like Cursor or Claude Code if your course involves any programming.
2. Are paid AI apps worth it, or is the free tier enough?
For most people, the free tier of one major assistant plus one specialised tool is enough. Pay only after an app has saved you measurable time for two or three weeks, and cancel anything you have not opened in a month.
3. What is an AI agent app, and how is it different from a chatbot?
A chatbot answers; an agent takes multi-step actions like browsing, filling forms, editing files or running code, then reports back. Agents are more useful and more risky, so supervise the first several runs before trusting them with anything that costs money or touches live data.
4. How do I keep up with new AI apps without wasting hours?
Do not track launches. Track your own repeated tasks, and go looking for a tool only when a task becomes annoying enough, which cuts your search from thousands of apps to maybe five.
5. Are AI apps safe to use with company or client data?
Check two things: whether the app trains on your inputs and where the data is stored. Enterprise or team plans usually turn training off by default, while free consumer tiers often do not, so keep confidential documents out of free tools unless the policy says otherwise in writing.
6. Will AI apps replace the jobs they automate?
They replace tasks faster than jobs. The realistic pattern in 2025 is that roles heavy in repetitive drafting, transcription and first-pass analysis shrink in headcount while the people who design and supervise those automated workflows become more valuable.