AI Tools That Have Transformed My Daily Workflow

The essential AI systems that have earned a permanent place in my productivity stack and the creative workflows that combine them

Every day, new AI tools emerge. I get my hands on most of them, experimenting with their capabilities and limitations. But only a few have truly earned a permanent place in my daily workflow.

A friend recently mentioned that she had already interacted with several AI systems before her first cup of coffee, which got me thinking about which tools have become essential in my own routine. Since I'm asked about this frequently, here's my current AI toolkit.

Core Thinking & Writing Tools

Claude (by Anthropic, Specifically, Claude 3.7 Sonnet): This has become my primary thinking partner. I use the desktop app extensively for brainstorming, writing, and organizing ideas. The "artifacts" feature is particularly valuable, allowing me to create and iterate on structured content seamlessly. Projects help me keep related work organized.

Example: I rely on Claude as my primary editor for my Pseudorandom Bits newsletter. I've set up a dedicated project with a detailed system prompt that outlines my preferred voice, grammatical structures, tone, and writing preferences. By pinning my articles to this project and providing detailed communication style guides, I ensure a consistent tone across my work. Claude helps me brainstorm new ideas, structure articles, maintain consistency and flow, generate AI image prompts for DALL-E, and craft SEO titles, descriptions, and post URL slugs, as well as associated social media posts.

ChatGPT (by OpenAI): Serves different purposes in my workflow. While Claude handles most of my coding, writing, and ideation, ChatGPT has evolved into my research tool and editorial voice. I use it to summarize articles, analyze text, validate ideas generated elsewhere, get consensus opinions, and critique my writing. The combination of search capabilities and the newer o3 models for reasoning make it particularly strong for research.

Example: I often use ChatGPT to explore new spaces, like when I wanted to learn about Solana contracts and how they compare to Ethereum. I was already familiar with Etheruem and it's smart contract language Solidity. I used ChatGPT to learn about the Sol ecosystem, Sol contracts, the Rust development chain, and the Anchor framework, comparing it with Etheruem along the way.

Other capture methods: I use whatever is handy (voice recorder, notepad, pen & paper, etc.) to capture random thoughts, generate meeting notes, and create content that I later refine through Claude. I find Superwhisper invaluable for quickly recording conversations, brain dumping thoughts, and capturing complex ideas quickly. I feed the transcripts generated from Superwhisper into Claude to generate structured notes, which I save in Notion for later reference.

Example: This article itself started with me just talking to Superwhisper as if I was talking to a friend and telling them what AI tools I use daily.

Development Tools

Cursor: This has completely replaced VS Code in my workflow. Cursor is an AI-first code editor that helps developers write, edit, and understand code faster. It integrates seamlessly with Claude models, which I find provide more accurate coding assistance. Cursor excels at visual, rapid iteration when working with individual files or components.

Claude Code: A newer addition but quickly became essential. I use Claude Code for analyzing entire codebases, documenting complex systems, and generating complete subsystems from high-level concepts. These are tasks I previously attempted with Cursor but found that Cursor doesn't handle as well as Claude Code. Cursor shines on single file or isolated modules. Where as, Claude Code can handle complex full codebase work.

I find Claude Code mind-blowingly capable. My interactions with it are reminiscent of working with a mid-level developer or capable entry-level team. It makes mistakes, but it can self-debug situations and think through errors it encounters, break down problems, and iterate on solutions. It even appears to hack through some options when unsure. It still needs guidance from an experienced architect. With clear direction on architecture, technology choices, and solution design plus the ability to debug emerging issues, you can greatly expand your productivity.

Claude Code's /compact command is an awesome way to capture notes from a long-running session. This feature summarizes the work and conversations, making it easy to document what was accomplished and retain key insights for future reference.

Example: I recently used Claude Code to generate an entire complex cloud infrastructure in modular AWS CDK code. This included a GitHub Action CI/CD pipeline and stubs for all the services, Dockerfiles, Lambda IAM permission requirements, API gateways, VPC needs, S3, SQS, route VPCs, CloudWatch alarms, and SES with appropriate SPF and DKIM setup. I knew the desired architecture, the specifics of what I wanted the CDK to produce, and my vision for how the platform would come together. I was able to give explicit direction to Claude Code to generate the code exactly as I wanted, verify it, and correct errors as they arose, continuously expanding the code. What's even more impressive is that I was able to do this in an evening while multi-tasking on 2 other projects!

Test-Driven Development: I've found that test-driven development is super advantageous with these AI coding tools. Building extensive tests for many edge cases, then instructing the tool to execute tests after changes and resolve issues until test suites run green creates an effective iterative loop. This approach helps produce higher quality code and improves the accuracy and results of the generator. As an added benefit, you're left with a well-documented, well-tested codebase, which is always a win for other developers (and AIs) who might work with the code later.

AI has greatly reduced the friction involved in working with unfamiliar code. I recently had to modify, refactor, and enhance an existing codebase. AI tools allowed me to quickly generate missing documentation, create diagrams from Terraform and CDK scripts, document database schemas from the code, and add unit tests for poorly covered files. I could even enhance the CI/CD setup before touching a single line of code. In essence, AI enabled me to address the project's lack of best practices, documentation, and automated testing before making any changes.

v0 (by Vercel): Has become invaluable for frontend development. Vercel's v0 is an AI-powered tool that accelerates UI design and frontend development. It allows developers to generate user interfaces rapidly using natural language prompts. By leveraging AI, v0 can create and iterate designs that align with modern UI standards and best practices, significantly reducing the time and effort typically required to design and prototype new interfaces.

v0 empowers me to produce higher quality front ends without a designer and a robust Figma design. It enables me to rapidly create clickable mockups that I can export directly to Cursor for backend integration. This streamlines the workflow from initial concept to a functional prototype.

Research & Analysis Tools

Notebook LM (by Google): Has become my solution for analyzing long-form content, such as multi-hour podcasts like Lex Fridman's interviews. Notebook LM is an AI-powered research tool that helps you understand and summarize documents. When listening to long podcasts, I’ll capture initial notes using whatever is handy (voice recorder, notepad, pen & paper, etc.) and later use Notebook LM to organize and expand on those notes, ask follow-up questions, and create a structured reference I can return to in the future or kickstart further research. This is particularly useful for content-dense discussions, like Friedman's 3-hour+ conversations that cover a wide breadth of subjects in depth, where someone might mention dozens of interesting tangents worth exploring.

Example: I recently used Notebook LM to analyze a Dwarkesh podcast with Jeff Dean And Noam Shazeer, Google AI leaders, pulling out individual topics like the evolution of AI systems and strategies for scaling large language models to start further research threads.

ChatGPT with search: Handles most of my web research needs. While I've experimented with dedicated research tools like Perplexity, ChatGPT's integrated search functionality has proven sufficient for most of my needs, reducing the need for additional tools.

Example: I use ChatGPT to research new prospects and partners, gathering up-to-date information on competitors, differentiators, market positioning, TAM, strengths/weaknesses, and recent press to inform my service offerings and initial conversations.

Auxiliary Tools

DALL-E (within ChatGPT) and Adobe Photoshop: For image generation, I primarily use DALL-E through ChatGPT to create initial images. However, DALL-E is rarely the end of my process. Adobe Photoshop plays a crucial role in refining these AI-generated images. While I don't heavily use Photoshop's own creative generative AI features, I frequently leverage its content-aware fills and similar AI-enhanced tools to polish and perfect images after their initial generation.

Gemini: While my use of Gemini has been primarily experimental, recent updates have drawn me in more and more. I'm already satisfied with Claude and ChatGPT and don't have much room for another chat assistant in my workflow. However, I deliberately use it from time to time since it's available on my phone, mostly to track how Google's AI offerings are advancing. Gemini 2.0 Flash is a great model. Their Deep Research is top-notch, and I now often run the same deep research request through both ChatGPT and Gemini and then merge the results. I have also started to explore Gemini's new Canvas features (rolled out this week). Gemini 2.0 is by far the best editor in the group (which makes sense given their Google Workspace offering). For example, this article was iterated, refined, and edited using Gemini Canvas after starting as a Superwhisper rant turned Claude Artifact and Notion page.

Eleven Labs and AWS Polly: For voice synthesis, I occasionally use these when needed, though these aren't daily tools. For example, I've used these tools to generate a voice mail message for my consulting business and voice overlays for instructional videos.

LLMs in Home Devices: I'm really looking forward to LLMs coming to home smart devices like Amazon Echo devices and Google Hubs. I have been an early adopter of this technology and my house is covered in smart devices and voice assistants. I'm eager to lean into Alexa+ when it rolls out.

Creative AI Workflows

Prompt Engineering with Claude: I often use Claude to generate prompts for other AIs, including itself. When preparing to use Claude Code, I'll brain dump my requirements (either by keyboard or verbally) and ask Claude to structure this into an effective prompt with clear objectives, verification steps, detailed instructions, and context. Claude is particularly good at breaking down a complex problem into a sequence of prompts that can be fed to other AI environments like Claude Code, Cursor, or v0. This meta-level use of AI dramatically improves the quality of my interactions with other AI systems.

For example, I've used Claude to brainstorm data models, entity definitions, and attributes. I then use Claude to generate prompts for creating SQL migrations, model layer code for ORMs, CRUD ReST APIs, and v0 admin screens, along with tests and associated infrastructure and project organization. This is all boilerplate stuff. Claude helps generate a series of prompts in steps, that I feed to Claude Code, Cursor and v0 to incrementally build out the application. I recently did this to initiate a new user management and account and team constructs for an existing prototype in order to scale the offering to new complex enterprise organizational structures.

AI-Powered Image Creation Pipeline: I used Claude to help me generate a series of prompts for DALL-E to create images for a presentation. I gave Claude a general description of the images I wanted, and it generated a series of detailed prompts that resulted in high-quality, relevant, and consistent images.

Example: For PseudorandomBits.io graphics, I start with an idea from Claude, which helps me formulate specific DALL-E prompts. Once generated, I refine these images in Photoshop. I've developed specific prompt templates for consistent aesthetic results across different articles.

Dual Editorial Team: I use ChatGPT and Claude as complementary editorial assistants. I'll ask ChatGPT to provide feedback on writing while emphasizing my preferred voice and tone. Its suggestions then get fed back to Claude for implementation. This creates a virtual editorial team with different strengths.

Consensus Modeling: For important questions, I feed the same prompt to multiple AI models, then ask Claude to synthesize the common elements and merge the feedback. This approach helps avoid individual model quirks and gives more balanced responses.

New Prospect Research Workflow: I recently explored a new domain for a prospective client. I used a combination of AI tools to accelerate and deepen my understanding. First, I used both ChatGPT and Gemini to conduct Deep Research on the domain. I then pulled the outputs from both into Notebook LM to quickly explore and analyze the research. This helped me identify key information and patterns. Next, I used Claude to synthesize the research findings into a variety of resources, including learning materials, FAQs, and lists of resources. Finally, I used Claude to leverage these insights to draft tailored proposals, strategic recommendations, product visions, and product development plans.

Meeting Intelligence: Gemini's note-taking feature works exceptionally well for Google Meet, but I've developed a habit of using Superwhisper for other meeting platforms like Zoom, Slack, and phone calls. I feed these transcripts to Claude to create structured meeting notes, add my thoughts, and then save the refined artifact in Notion.

Learning Through Conversation: When driving or walking, I engage in educational conversations using ChatGPT's voice mode. I specifically prompt it to avoid reading out code or complex technical terms, allowing for natural discussion similar to office hours with professors. At the end of the conversation, I'll ask it to summarize the entire conversation in a Canvas document so that I can add it to my Notion notebook.

Document Cycling: I export PDFs from Notion to attach to new chat windows or pull into Notebook LM when I want to iterate, search my own notes, or use them as context for transient logs. This is a little clunky, but effective. I'm on the look out for a good MCP (Model Context Protocol) server that can integrate Notion directly into Claude Desktop.

What I Don't Use

While I've experimented with many other tools, some haven't stuck in my routine:

Notion AI: Despite using Notion extensively for note-taking and documentation, I find their AI features unnecessary. I prefer to create content in Claude and paste the markdown into Notion when finished. I use occasionally to generate a summary for a document, but even that I'd prefer to do in Claude. Most importantly, I just don't want to pay for something else that I can already do with tools I already pay for. Claude copy markdown and Notion paste markdown work so well together as is.

Integrated AI assistants: Integrated AI assistants in platforms like Slack or email clients haven't provided enough value to be part of my workflow. However, I'm increasingly using MCP to interoperate with these apps through Claude Desktop, which offers more flexibility and power than the native integrations. I had used Shortwave for a year, but I couldn't justify the renewal. I still found myself using Gmail most of the time and/or drafting emails in Claude or inline with Gemini. Shortwave bundles are great and I do miss them. Fingers crossed they come to Gmail soon.

Perplexity: I like what they have built and find it an amazing application and service. It just hasn't hooked me in the same way other tools have. ChatGPT remains good enough on these fronts. That said, I plan to give Claude + Perplexity a spin together now that Perplexity offers an MCP service. I'll continue to keep an eye on Perplexity as it advances.

What's Next: The Broader AI Ecosystem

This article focused primarily on my productivity tools, creativity accelerants, and research assistants. I haven't touched application building libraries, frameworks, and subsystems such as Ollama for local models, AWS Bedrock, Vercel AI SDK, and others. That's a subject for another day.

Conclusion

My daily AI interactions center primarily around Claude, ChatGPT, Cursor, and Claude Code. These tools have transformed how I think, write, code, and research. While I continue to experiment with new tools as they emerge, I've learned to be selective about which ones earn a permanent place in my workflow, focusing on those that offer unique capabilities or significant productivity improvements.

The most valuable tools aren't necessarily those with the most features, but those that integrate seamlessly into existing workflows and solve real problems. As AI tools continue to evolve, I expect this list will change, but the core principles will remain the same: Does this tool help me think better, work faster, or create something I couldn't otherwise?

Call to Action

What are your favorite AI tools and how do you use them in your daily life? I'd love to hear your thoughts and experiences in the comments below.