The shortlist
- Cursor — IDE-resident coding agent.
- Cloak — overlay copilot for meetings and interviews.
- Claude (web / desktop) — long-form thinking and writing.
- Granola — meeting notes that don't drown you.
- Vercel AI Gateway — multi-provider model router for shipping apps.
1. Cursor — the IDE
Cursor is where most of our code happens. The agent loop is mature, the multi-file edit story is the best in class, and the model selector lets us route per-task instead of being locked to one provider. If you only adopt one AI tool in 2026, it's a code editor that knows how to actually edit code — and that's Cursor.
Runner-up: GitHub Copilot for teams already standardized on GitHub.
2. Cloak — the overlay
The category we built. Cloak runs invisibly in the background of every meeting, interview, and screen-share. It's the answer to "what if the AI was just there instead of being a tab you have to alt-tab into mid-conversation?"
The pragmatic reason it's in the top five: meetings and calls eat the parts of the day where you can't open another app, and that's exactly the time the overlay shape wins.
Runner-up in the category: Cluely.
3. Claude — long-form thinking
Claude (4.5 / 4.6 Sonnet and Opus) is where we drop large documents, long traces, and any task that needs the model to actually think instead of pattern-match. The 200k+ context and the "thinking" preface output makes it the default for editorial work, deep research, and code-review-style tasks.
Runner-up: GPT-5.x for tasks where you need tighter structure and faster latency.
4. Granola — meeting notes
Granola hits the sweet spot: it transcribes, summarizes, and respects the time you spent actually paying attention. Combined with Cloak, the workflow is "Cloak helps me during the meeting; Granola helps me remember it after."
Runner-up: Fireflies, Otter.
5. Vercel AI Gateway — the router
When you ship a product that calls LLMs, you don't want to be tied to one provider's API. Vercel's AI Gateway sits between your app and 100+ models, gives you per-call cost and latency, and lets you switch providers with a config change. It's how we keep Cloak's managed tier model-agnostic.
Runner-up: OpenRouter, LiteLLM.
What didn't make the top five
- ChatGPT (consumer): excellent, but we mostly use Claude desktop or provider APIs directly through Cursor / Cloak / the gateway.
- Perplexity / Phind: great for one-off research; we use them, just not daily.
- Replit Agent / Bolt: impressive demos, not part of our daily flow.
- v0: we use it for UI scaffolding, but the output usually moves into Cursor for finishing.
The pattern across all five
None of these tools tries to do everything. The winning AI products in 2026 are the ones that pick one job in the workflow and own it. Cursor owns the IDE. Cloak owns the overlay. Claude owns long-form. Granola owns meeting notes. The gateway owns routing.
The losing pattern is "AI workspace that does everything." The user has to learn one new app instead of slotting AI into the apps they already use.
Add Cloak to your stack
Download from the home page. Plays well with everything else on this list.