There is a tool announcement every Tuesday. Most of them are noise. Some of them are genuinely good. A small number have changed how we actually work. This list is the small number.
We do not get paid by anyone on this list. We just use them every day and pay for them out of our own pocket or our clients’ pockets. If a tool is here, it earned its place by being indispensable, not by being new.
The 10 tools we actually pay for
1. Claude (Pro and API)
Our default model for almost everything that matters. Better than the alternatives at long-context reasoning, code review, and following nuanced instructions without going off-script. We use Claude Pro for chat and the API for everything in production.
- Best at: Long-form reasoning, careful code review, structured output that does not hallucinate.
- What we use it for: Code generation, document analysis, and as the brain of every agent we build for clients.
- Cost: $20/month for Pro. API costs vary.
2. Claude Code
The terminal coding agent that turned into our daily driver. We replaced most of our IDE-based AI workflow with Claude Code over the past three months. It handles multi-file edits, runs tests, fixes its own mistakes, and gets out of the way when you want to type.
- Best at: Complex multi-file refactors, building features from a spec, debugging problems that span the whole repo.
- Why it beat Cursor for us: Better at planning, better at autonomous multi-step work, better at knowing when to stop.
3. Cursor
Still the best AI-native code editor. If you want something that feels like writing code with a smart pair partner, Cursor is the one. We use it for fast iterative work, prototyping, and any project where the keyboard feels better than the agent.
- Best at: Tight feedback loops, inline suggestions, and the moments when you want to be in control.
- Cost: Free Hobby tier, $20/month for Pro.
4. Linear
Not technically an AI tool, but the AI features are now the best part. Linear’s auto-triage, summarization, and PR linking are how we actually run projects. The 2026 update added an agent that can pick up tickets and draft pull requests. It is rough, but it is real.
5. Perplexity Pro
We use Perplexity for any research task where citations matter. It is faster than Google, better than search-augmented chat models, and the Pro tier gives you Claude Opus and GPT-5.4 in the same interface.
- Best at: Competitive research, fact-checking, and building briefing docs you can hand to a client.
- Cost: $20/month.
6. Granola
The meeting notes tool that quietly replaced everything else. Granola listens, transcribes, and lets you write your own notes alongside the transcript. The summary is good, the search is good, and it never joins a Zoom call as a creepy bot.
7. Resend
The email API we ship every client project on. Clean SDK, real deliverability, and the inbound webhook just made our cold-outreach loop work. If you are still on SendGrid, you should evaluate.
8. Supabase
The database, auth, and storage layer for every AI app we ship. Postgres under the hood, real-time subscriptions, edge functions, and a vector store that actually scales. The pgvector integration is the reason we stopped reaching for Pinecone.
9. n8n (self-hosted)
The workflow engine we use when a client needs automation that connects 8 different SaaS tools and we do not want to write 8 different adapters. Self-hosted because we do not want our clients’ data sitting on someone else’s server. Cheaper than Zapier and infinitely more flexible.
10. ElevenLabs
Voice cloning and text-to-speech that does not sound like a robot. We use it for client demos, voice agents, and the voice mode in our own products. The 2026 v3 model added emotion control that actually works.
4 tools we have stopped paying for
ChatGPT Plus
Not because GPT is bad. GPT-5.4 is excellent. We just found ourselves opening Claude first, every time, and never going back. The interface and the integrations are great. The model just is not our default anymore. Your team may feel differently. Try both.
Zapier
We replaced Zapier with n8n eighteen months ago and never looked back. The pricing for any non-trivial workflow gets brutal fast, and the AI-native automations we needed just were not there. Zapier has improved, but we are not paying to find out.
Notion AI
Notion the workspace is great. Notion AI as an add-on never earned its keep for us. We pay for raw model access through Claude and use Notion as a clean canvas. Your mileage may vary, especially if your team lives entirely inside Notion.
Jasper
The category that copywriting AI tools used to own has been eaten alive by general-purpose models. There is no marketing copy task that Claude or GPT cannot do better, faster, and cheaper. If you bought into the category in 2023, it is probably time to re-evaluate.
The tools we are watching but not paying for yet
- Devin (Cognition). Got better, still not better than Claude Code for our work. Worth keeping an eye on.
- v0 by Vercel. Great for component prototyping. Not a daily driver for serious work.
- Replit Agents. Promising for non-developers. We have not found the production use case yet.
- Mistral Large. European, fast, cheap. We use it for specific cost-sensitive jobs but not as a default.
The pattern across the keepers
Every tool on the keeper list does one thing extremely well, has a real API, and respects your data. The tools we cut were either trying to do too much, were getting eaten by the underlying model improvements, or were just not worth the friction.
Buy fewer tools. Use the ones you keep harder. The teams shipping the most are not the teams with the most subscriptions.