My (mostly) minimalistic AI setup as a Senior Engineer in Big Tech
You can stop being overwhelmed by the hundreds of new tools every day
🌶️ Hot take: You don’t need more AI tools. You need a core few that you know well.
If you’re overwhelmed (or sick) of keeping up with the hundreds of new tools that come out every day, this post is for you.
I’m here to tell you that you don’t need hundreds of tools. To be productive with AI, you only need to know your core workflows that you spend the most time on and ask if you can use AI there. For most people, this will result in fewer than 10 tools.
In this article, I’ll walk you through the tools I use and explain how they help me with my core workflows. If you think a tool will be useful for you, I encourage you to try it!
1) Raycast: My AI Entrypoint
I use Raycast to access AI via Cmd+Space, save prompts, and much more.
Quick AI from anywhere: I hit Cmd+Space then tab, and I’m instantly chatting with AI. I don’t need to switch tabs, open browsers, or wait for ChatGPT to load.
Prompt storage: Every time I find a high-quality prompt—whether from a tutorial, course, or blog post—I save it as a “Chat Preset.” I can quickly interact with it any time via Cmd+Space.
YouTube video summaries: Raycast pulls the transcript from any YouTube video and summarizes it. When someone sends me a YouTube video to watch, I first read the summary, then check the key takeaways, and decide whether to watch it fully, skip to specific parts, or skip it entirely. This saves me hours each week.
I can also do the same thing for any webpage I’m on. You can ask any question to the webpage you’re on, where my “question” is usually just one word: “Summarize.” I do this for really long articles where I just want the key takeaways.
2) Perplexity: Deep Research
When you want to learn about a new domain area, Perplexity is the place to go.
You might wonder, why not Google? Google searches often result in many sponsored articles at the top, and the results aren’t as specific to what you need. In Perplexity, you can give it highly detailed information to your needs and get tailored results.
Specifically, I get…
Detailed overviews: I can ask “What are the major developer productivity frameworks?” and get a comprehensive overview with sources in one response.
Better source material: The sources Perplexity finds are often more niche and more helpful for developer cases.
Try searching the same prompt in Google, and you’ll get sponsored articles and pages asking for your email to download a PDF 😵💫
3) ChatGPT and Voice Mode: AI on the go
I primarily use ChatGPT away from my laptop when I’m on the go, as Raycast on iOS only recently launched. I also really love the voice mode, where you can talk back and forth to it.
The use cases are endless here as it’s so general purpose, but a few recent ones:
Travel guide: During my recent trip to Stonehenge, I asked it to roleplay as a master storyteller and tell me about what I was seeing.
Voice mode cases: Voice mode is when you speak to GPT, and it responds back to you via voice, just like a real person. Here are a few use cases:
Preparing for important conversations: You can role-play key conversations, such as an interview or a salary negotiation with a recruiter.
Feedback: You can get feedback on a presentation or pitch.
Real-time translation: I recently had workers over who primarily spoke Spanish. While I knew a little, I was rusty. I used ChatGPT voice mode as a translator—we could both speak into it and understand each other as it spoke each person's language back to them.
4) Cursor: IDE/Editor
If you're not using Cursor or another IDE with Agent Mode, you’re missing out on a significant productivity boost in software engineering. I did a presentation at Infoshare 2025 on many use cases, and I covered how to prompt it well in this article, but here are a few use cases to think about:
Quick scripts for one-off tasks: For any minor task or small script, ask the agent to do it. It excels at these because it doesn’t require a lot of context.
Test writing and fixing: I paste broken tests, describe what they should do, and Cursor fixes them. Or I write a function and ask it to write comprehensive tests.
Error log analysis: When I have large error logs, I ask Cursor to summarize the errors and categorize them. Then we solve them one by one.
Documentation: I select a complex function and ask Cursor to write clear documentation. It understands the context better than I could explain to a human.
Writing features: Let’s be real. Many features are copied and pasted from existing ones and just need some tweaking. Agent Mode IDEs are great at this. Provide them with sample code, explain what’s different this time, and ask them to type it out. It’ll save you plenty of time.
I've been using Cursor the most, but I’m also trying Windsurf. Both are better than coding without AI.
5) MacWhisper: Dictation
MacWhisper allows me to hit the right “cmd” button and dictate into any text input.
Is MacWhisper the best dictation tool? I don't know. I’ve heard people talk about Superwhisper and a few others, but dictation is simple enough that this is fine for me.
I use it primarily with Cursor—I can speak my requirements instead of typing them. On rare occasions, I'll use it for presentation scripts. I record myself talking through slides, get the raw transcript, then feed that into a prompt to clean up grammar and refine it.
6) Superhuman: Email That Stays Clean
I think AI in email makes a lot of sense, and Superhuman does a good job giving you productivity gains with low effort on your part.
Smart filtering: It identifies what appears in my “Important” inbox, keeping it clean. For context, my standard Gmail inbox gets 20-30 emails per day, and my Superhuman “Important” inbox gets 1-5 emails. Everything else gets filtered to “Other,” which I check more rarely.
Auto-follow-ups: It writes automatic follow-up emails if I haven’t received a response.
Smart labeling: It auto-labels emails using the labels I’ve created. Email organization happens automatically.
7) AI Libraries for Building
When I'm building agents or integrating AI into applications, I primarily use:
LLM Calls: For LLM calls, the Vercel AI SDK is the simplest. I’ve also started exploring LangGraph and the OpenAI Agents SDK
Asynchronous agents: For asynchronous agents out of the box, I’ve used OpenAI Codex and Claude Code. More on how to use those in this article.
8) DeepWiki and GitDiagram: Learning
Both Deepwiki and Gitdiagram help you understand how any open-source library works under the hood. For example, here is the page on how OpenAI Codex’s agent loop works, fully documented by DeepWiki:
Similarly, change the “git” in any GitHub repo URL to “gitdiagram” and you’ll get a super-detailed system diagram of how that repo works. Understanding how your tools work under the hood helps you grow as an engineer and maximize the value you get from them.
What I Haven't Explored Yet
Workflow automation tools: Zapier, n8n, and similar tools could probably optimize some of my repetitive tasks. I do a lot of manual email filtering that could be automated with summaries. I just haven’t gotten to it yet!
Creator tools: I've been reluctant to explore AI for content creation. Most require just as much effort as doing it yourself since the real value comes from personal stories and experiences—which you need to tell the AI about anyway. Still, I’ve experimented with giving ChatGPT my outline, bullet points, and unedited stories of what I want to say, then having it generate a version with structured sentences. This has helped reduce my writing time by 10-20%, but I still do most of it manually.
📖 TL;DR
You don’t need a million tools to see huge productivity gains with AI. These were the tools I use because they help me with my core workflows. The tools you choose may be different—you might have more or less, and that’s ok!
Here’s a recap of the tools:
Raycast: Quick AI access via Cmd+Space, prompt storage, and YouTube video summaries
Perplexity: Deep research and comprehensive overviews with quality sources
ChatGPT (mainly voice mode): AI on-the-go, travel guidance, conversation prep, and real-time translation
Cursor: AI-powered coding with agent mode for scripts, tests, debugging, and feature development
MacWhisper: Voice dictation for any text input, especially coding requirements in Cursor
Superhuman: Smart email filtering, auto-follow-ups, and automatic labeling to keep inbox clean
AI libraries: Building LLM integrations and asynchronous agents using the Vercel AI SDK and OpenAI Codex
DeepWiki and GitDiagram: Learning how open-source libraries work under the hood
👏 Shout-outs of the week
MCP auth just got a spec upgrade and Anthropic just rewrote the rules—MCP servers must now offload authentication and token issuance to a dedicated authorization server. No more static API keys or embedded OAuth hacks.
Scalekit gives you a drop-in, OAuth compliant authorization server out of the box — complete with dynamic client registration, PKCE, token introspection, and org-level scoping, so your MCP server stays focused, secure, and spec-aligned.
Read why this shift matters. Thank you to Scalekit for sponsoring the newsletter!
Here were my top reads for the week:
Real-world engineering challenges: building Cursor on
by Gergely Orosz — Similar to point (8) on understanding the tools you use, this article does a great job explaining how Cursor works under the hood.- by Rafa Paez — Loved the detailed walkthrough of how agents work, along with the code examples!
Thank you for reading and being a supporter in growing the newsletter 🙏
You can also hit the like ❤️ button at the bottom of this email to support me or share it with a friend to earn referral rewards. It helps me a ton!
Great post, Jordan! I learned some tools I didn't know and seem worth to try them.
Thanks for the shout out!
For me, Cursor is a game changer. I wrote about it in my newsletter.
I've just installed the Raycast, I think it's just what I was looking for.
Thanks Jordan!