The 2024 wave of AI productivity tools was mostly underwhelming. The 2025 wave got better. The 2026 wave, after a year of model refinement and workflow integration, has produced a small set of tools that genuinely save time for working professionals.
The distinction between useful and useless AI tools is the same as it has always been: does it replace a real cognitive task, or does it generate noise that has to be cleaned up later? Most tools do the second. A handful do the first. Here are the ones that do the first.
The five that earn their subscription
1. Claude (or ChatGPT, depending on taste) for first drafts
The single best use of a frontier LLM is producing a first draft of something you would otherwise have stared at a blank page to write. The draft is rarely good. It does not need to be. Its job is to give you something to edit, which is psychologically different from writing from zero.
Real use cases that work:
- Outlines and section structure for long documents
- First-pass email replies where the emotional content is the bottleneck, not the words
- Summaries of long documents you do not have time to read carefully
- Boilerplate-heavy work: contracts, NDAs, status updates, project briefs
Real use cases that do not work:
- Anything that requires domain expertise the model does not have
- Anything that requires citation or current information (use a tool with web access)
- Anything where the cost of being wrong is high
The mistake most people make is treating the output as finished. Treat it as a competent colleague's first draft that you would have to rewrite anyway.
2. Granola for meeting notes
Granola is a meeting transcription tool that solves the actual problem, which is not transcription but the post-meeting cleanup. It joins the call, takes notes, and produces a structured summary with action items and decisions.
The output is editable, syncs to your calendar, and lets you copy action items directly into your task manager. It is not cheap (around 14 euros per month), but for anyone in five or more meetings a week, it pays for itself in saved time and fewer forgotten commitments.
3. Raycast for keyboard-driven everything
Raycast is a launcher that has quietly become the most productivity-dense app on macOS. It is Spotlight but with superpowers. You can:
- Run any app or shortcut from the keyboard
- Calculate, convert, look up definitions, paste snippets
- Manage windows, snippets, calendar, clipboard history, system controls
- Extend with community plugins for almost every common workflow
It is free for personal use. The paid tier (around 10 euros per month) unlocks AI features and pro extensions. For anyone who spends the day at a computer, the muscle memory of Raycast cuts more time than any single AI tool.
4. Superhuman for email triage
Superhuman remains the best email client for anyone with a demanding inbox. It uses AI for triage, scheduling, and short reply suggestions. The keyboard-driven interface is faster than anything else available.
The downside is the price (about 30 euros per month) and the closed ecosystem. It works with Gmail and Outlook but not with IMAP. For users of those two providers, it is genuinely better than the alternatives.
5. Notion AI for connected knowledge
If you already use Notion, the AI features are worth turning on. They are not transformative in the way that LLM chat tools are, but they make small-but-frequent tasks faster: summarizing meeting notes, drafting from templates, restructuring documents, generating action items from pages.
The real value of Notion AI is integration. The AI understands your existing pages, so it can pull context from a project database, summarize a long-running thread, or generate a status update from a series of meeting notes. Chat tools cannot do this.
What to skip
Most "AI notetakers" that are not Granola
There are now dozens of meeting transcription tools. Most of them produce a transcript you do not read, in a format that does not integrate with your task manager, with a price that makes them hard to justify. Granola's structural summary is the reason it stands out.
AI writing tools for content
The marketing copy for AI writing tools has gotten louder as the actual differentiation has shrunk. Most paid writing tools are now GPT wrappers with a custom prompt and a markup on the API cost. The result is the same as ChatGPT with a stronger system prompt, but at five to ten times the price.
AI image generators for stock photography
AI image tools are useful for illustrations, mockups, and concept images. They are not useful for editorial photography, which still requires real photographs of real subjects. Most sites using AI-generated "stock photography" look uncanny and dated quickly.
Most "AI assistant" browser extensions
Most of these are wrappers around the same models, often with weak privacy practices. If you want an AI assistant in the browser, use the sidebar in ChatGPT or Claude directly. You will get a better model with cleaner data handling.
A practical adoption order
If you are starting from zero, the order that gives the highest return per hour saved:
- Set up a keyboard launcher (Raycast on Mac, PowerToys on Windows). One weekend of learning. Pays back forever.
- Use a frontier LLM for first drafts. Free tier is enough to start. Build the habit before paying.
- Try a meeting transcription tool if you are in five or more meetings a week. Granola is the current leader.
- Pay for one specialty tool that targets your biggest specific bottleneck. Email, calendar, design, coding — whichever is your actual pain point.
- Do not adopt more than three paid tools. Tool fatigue is real and the marginal time saved per added tool is negative.
The goal of all this is to make a working day feel less compressed, not to add another layer of software to manage. A small set of well-integrated tools beats a large set of barely-used subscriptions.
What I actually use day to day
The list above is what I would recommend to someone else. The list of what I actually use day to day is shorter, and it is worth describing because it differs from the recommendation in instructive ways.
I use Claude for first drafts of the writing I do. The use is narrow. Outlines for articles. First-pass emails where the emotional content is the actual bottleneck. Summaries of long documents I have been sent and need to respond to before a meeting. Boilerplate-heavy work — NDAs, contract amendments, project briefs. The work I do with Claude is the work I would have spent an hour staring at a blank screen for. It is not the work that defines my job. It is the work that gets in the way of the work that defines my job.
I use Granola for every meeting I am in. This is roughly fifteen hours a week. The post-meeting cleanup that used to take fifteen minutes per meeting now takes three. The compounding effect across a week is roughly two hours saved. Granola costs €14 a month. The cost-per-hour-saved is €0.60, which is the best return on any subscription I pay for.
I use Raycast probably two hundred times a day. The actual muscle memory of the application — command-bar to switch apps, snippets for things I type repeatedly, clipboard history, window management — has become invisible to me. I would notice the absence of Raycast within an hour if it disappeared.
I use Notion AI for the connected-knowledge tasks it does well. Pulling context from a project database. Generating status updates from a series of meeting notes. Summarizing long-running threads. The use is small in volume but high in value, because the work would otherwise require me to manually piece together context from multiple places.
I do not use Superhuman. I tried it for a month. The keyboard interface is excellent. The closed ecosystem is a real limitation. The €30 a month is not justifiable on a personal account, and I am not on a corporate account that would cover it. The recommendation stands for people whose inbox demands more than mine does.
What I do not use
A few tools that get a lot of marketing attention and that I have either stopped using or never adopted.
Otter.ai. I used Otter for two years before switching to Granola. The transcription quality is similar. The post-meeting summary was the difference. Otter produces a transcript. Granola produces a summary. The summary is what I needed.
Motion. The AI calendar tool that promises to schedule your day for you. I tried it. It scheduled my day badly. The algorithm does not understand which meetings require preparation and which are drop-ins, and it scheduled my focused work time during the slots when my colleagues were most likely to interrupt. I cancelled after two months.
Reclaim. Similar to Motion. Similar disappointment. The AI did not understand my actual work patterns, and the suggestions felt like they were generated by someone who had read about my job but not done it.
Notion AI for actual writing. I tried generating articles, blog posts, and reports with Notion AI. The output is competent and generic. For writing that needs to sound like me, I write it myself. Notion AI is for the connected-knowledge tasks described above, not for content generation.
The adoption mistake
The single biggest mistake I see people make with AI tools is adopting too many at once. They pay for Claude, ChatGPT, Gemini, and Copilot simultaneously. They pay for three meeting transcription tools to compare. They pay for every AI browser extension that promises to save time. The result is ten subscriptions, none of which they use deeply, and a working week that is more fragmented than before.
The fix is the opposite direction. Use one tool deeply. Understand what it does well and what it does badly. Add the second tool only when the first has been fully integrated into your workflow. Add the third only when the second has. The cost of an unused subscription is small. The cognitive cost of switching between similar tools is large.
I have been refining this stack for about eighteen months. The five tools I use today are the five tools I would have recommended to my past self. They will probably change. The principle will not.
The honest summary
AI productivity tools in 2026 are at the point where the marketing has caught up with reality. The tools that work are tools that solve a real problem. The tools that do not work are tools that wrap a model in a custom prompt and charge a markup.
The five tools above are what I would recommend to anyone who is starting from zero. The adoption order matters. The first draft use case for LLMs is the easiest entry point because it pays back immediately and requires no integration with existing workflows.
The mistake is not using AI tools. The mistake is using too many, or using the wrong ones, or treating the outputs as finished when they are drafts. Done well, these tools save hours. Done badly, they add a layer of work. The difference is intentional.