New AI products appear every week, but only a small number earn a permanent place in a working routine. The ones that do tend to share a trait: they remove a slow, repetitive step rather than adding a novelty. This roundup covers the tools that were drawing the most attention in late November 2025 across dictation, meeting notes, browsing, automation, creative production and research, with a short description of what each one does and who it is for.
Dictation and meeting notes
Wispr Flow is a voice-to-text application for Mac, Windows and iPhone that goes beyond transcription. According to the company, it cleans up what you say as you speak, removing filler words, false starts and mid-sentence changes of direction, and produces a finished sentence ready to paste into a message, document or AI prompt. The pitch is speed: dictating prompts and messages faster than typing them, without having to edit afterward.
Granola addresses a different pain point. It is a meeting notepad that records and transcribes calls without sending a bot into the meeting. Users jot short notes during the call, and the app combines those notes with the full transcript to produce a structured summary afterward. The two products have begun to overlap: Granola has added voice typing to its Mac app, and Wispr has launched a meeting notetaker, so the choice between them increasingly comes down to which workflow a user starts from.
Browsing and automation
Comet, the browser from Perplexity, folds research, video summaries, chat about the open page and task automation into a single window. Perplexity describes use cases such as pulling a list of products mentioned in a YouTube video, triaging an inbox or analyzing a website’s content, all handled by an assistant that can see and act within the browser.
For heavier automation, n8n is an open-source workflow platform that connects applications and AI models through visual flows and agents. Common deployments include processing incoming newsletters, routing customer service requests and automating sales follow-ups. It can be run on n8n’s cloud service or self-hosted on a server, which gives organizations more control over data and typically lower costs at scale, at the expense of managing the installation themselves.
Image, video and sound
Several platforms now bundle many generative models behind one interface. Freepik, Krea, Leonardo and Artlist each offer access to multiple image and video models, letting users switch between them without separate subscriptions. Replicate takes a developer-oriented approach, hosting thousands of open models, from large language models to video generators, behind an API.
Sound has become the next frontier. Mirelo generates synchronized sound effects, Foley and ambience directly from uploaded video footage, which the company says removes the work of finding, trimming and syncing individual effects. MMAudio, an open research model, performs a similar video-to-audio task and is available through free web front ends.
For fixing footage and images rather than generating them, Magnific is a widely used creative upscaler that adds detail to low-resolution pictures, and Topaz Labs’ Astra applies the same idea to video, addressing softness, warping and missing detail. Google’s Nano Banana image model, meanwhile, has become a go-to for consistent characters, thumbnails and graphics, and its Pro version, released in November 2025, promises better text rendering and layout control.
Research and workflow building
Google’s NotebookLM remains a standout for research. Users load documents, links and recordings into a notebook, and the tool answers questions grounded in those sources and turns them into other formats, including quizzes, mind maps, reports, audio overviews and, as of November 2025, infographics and slide decks.
Finally, the major assistants are increasingly places to build small tools rather than just chat. Gemini’s Gems, ChatGPT’s custom GPTs and apps, and Anthropic’s Skills for Claude each let users package instructions and files into reusable workflows that handle a recurring professional task in a few steps.
Taken together, the list suggests where the practical value of AI currently sits: not in the newest model announcement, but in tools that shave minutes off tasks people already do many times a day.
