Every week there’s a new “100 AI Tools You NEED” thread on Twitter. Most of them are demos pretending to be products.

I lead an engineering team doing real-time data processing. Here are the tools that actually survived more than a week in my workflow.

1. Claude Code — Coding Assistant

What it does: Coding assistant that runs in your terminal.

How I actually use it:

  • Pre-screen code reviews before I look at PRs
  • Generate test cases from specs
  • Debug production issues by pasting stack traces
  • Write documentation that people actually read

What I like: It handles large context windows well. I can paste an entire file and ask specific questions about it. The terminal integration means I don’t have to context-switch into a browser. GPT-4 and Gemini are solid alternatives depending on what you’re doing.

What’s annoying: It can be slow on longer prompts, and the output sometimes needs a fair bit of editing before it’s actually usable. You still have to read everything it gives you carefully.

Cost: Comes with Claude Pro ($20/mo) or use the API.

2. Cursor — When I Need to Write Code Myself

What it does: VS Code fork with AI built in.

How I actually use it:

  • Tab-complete that actually understands my codebase
  • Inline edits: highlight code → describe what to change
  • Chat with codebase context (it indexes your project)

Why I use both Claude Code and Cursor: Claude Code for planning, reviewing, and generating. Cursor for writing and editing. Different modes of work.

The downside: Cursor’s AI features can be hit-or-miss on larger files, and the indexing occasionally lags behind recent changes. It also burns through API credits fast if you’re not paying attention.

3. Notion AI — Meeting Notes and Docs

What it does: AI features inside Notion.

How I actually use it:

  • Summarize meeting transcripts into action items
  • Draft technical RFCs from bullet points
  • “Improve writing” on docs before sharing with stakeholders

Honest take: It’s not revolutionary, but it’s convenient because everything’s already in Notion.

4. GitHub Copilot — The Background Assistant

What it does: Code autocomplete in your editor.

How I actually use it:

  • Boilerplate code (API endpoints, data models)
  • Writing similar patterns across files
  • Unit test generation

Note: I’ve reduced my Copilot usage since getting better with Claude Code. But for quick inline completions, it’s still fast.

5. Perplexity — Research Without 20 Tabs

What it does: AI-powered search with citations.

How I actually use it:

  • “What’s the current best practice for X in Python?”
  • Researching vendor tools before team decisions
  • Quick competitive analysis

Why not just Google? Because I want an answer with sources, not 10 SEO-optimized blog posts saying the same thing.

Worth noting: The citations look authoritative but aren’t always reliable — I’ve caught it confidently linking to sources that don’t quite say what it claims. You still need to click through and verify, especially for anything your team will act on.

6. Granola — Meeting Intelligence

What it does: Records meetings and generates structured notes.

How I actually use it:

  • Auto-generates action items from 1:1s
  • Searchable history of what was discussed and decided
  • Catches things I missed during the meeting

Game changer for: Sprint retros and cross-team syncs where there’s a lot of context flying around.

The catch: Some teammates are uncomfortable being recorded, even with consent. And the auto-generated action items need cleanup more often than not — they tend to be either too vague or oddly specific about the wrong things.

7. Linear + AI — Project Tracking

What it does: Project management with AI features.

How I actually use it:

  • Auto-triage incoming bugs
  • Generate sub-tasks from epics
  • Smart duplicate detection

What I Tried and Dropped

  • Jasper — Too marketing-focused. Not useful for technical content.
  • Otter.ai — Good transcription but Granola’s structure is better for my workflow.
  • Bard/Gemini — Fine for general questions, but Claude and GPT-4 are better for code.
  • Devin — Not ready for production codebases. Maybe next year.

The Real Productivity Hack

Tools don’t matter if you don’t have the habit. On a good day, this is roughly how it goes:

  1. Morning: Review PRs with Claude Code pre-screening
  2. Meetings: Granola running in background
  3. Coding: Cursor + Copilot
  4. Research: Perplexity
  5. Docs: Notion AI for polish

Some days I barely touch half of these. It depends on whether I’m in meetings all day or actually have time to write code.

I estimate these save me several hours a week combined. Hard to put an exact number on it.

This post will be updated quarterly as tools evolve. Last updated: February 2026.

You might also like

📦 Free: AI Code Review Prompt Pack — 10 prompts I use on 15+ PRs/week.

Newsletter: One practical AI workflow per week, plus templates I don’t publish here. Subscribe →