These are my AI daily drivers and how I use them. For context, I have a CS undergrad but haven't written code in decades. I'm technical but work as a product manager, focusing more on roadmapping, data gathering, outcome analysis, and plan sharing than on semi-colon placement, technical architecture, HTML, color palettes, or designs.

My AI Journey

With my first exposures to this cycle of AI, I was as focused on the tricks and surprises as everyone else. I started by playing with generative images, curious about what "Santa Batman" might look like. I explored ChatGPT's advice and even sought novel recipes like Caribbean-inspired shortbread.

In recent months, I've made a concerted effort to integrate these tools into my workflow. I've written code, built apps, conducted research, had code explained to me, created diagrams, and I'm currently on day 9 of a commitment to publish 30 pieces of writing in 30 days. There's nothing like a project to help you flex in a new space, and I've built several.

My AI Toolkit

This is what I'm working with on a daily basis, things I’m looking to play with as well as things that are on their way out to pasture.

General LLM: Anthropic's Claude

• Excellent "projects" functionality and code previews

• ChatGPT for voice interaction but transitioning here as well

I pay for both ChatGPT and Claude, but my usage is primarily Claude these days. The scaffolding they've built via their "projects" functionality and code previews is amazing. A key to success with these tools is providing enough context for specific needs, and Claude's scaffolding facilitates this well.

I still use ChatGPT for voice mode interaction, which is helpful for talking through problems or speaking stream of consciousness while driving or walking. It transcribes, organizes, and bullet points what I want to convey. I'm pretty sure Claude can do this too, and I'll find out when I discontinue my ChatGPT subscription.

Answer Machine: Perplexity

• AI-enhanced search engine with up-to-date information

• Provides citations for responses

• Free tier gets the job done

Perplexity's free tier is a FREQUENTLY used tool during my days to just find the answers for things. Claude and ChatGPT leverage their training data which is a decent number of months in the past but Perplexity's responses are based on web searches and provides citations for its responses. I use this to help confirm my notions of how perplexity is different from ChatGPT and Claude, to do company research, and even to get help writing code that is using libraries or APIs that have been more recently updated. I recently had it explain to me why everyone on the internet was talking about being "demure" all of a sudden.

API Tools

• API keys for ChatGPT, Claude, and Perplexity

• Allows pay-per-use when hitting subscription limits

• Uses big-agi.com as a chat interface

I have API keys for ChatGPT, Claude, and Perplexity, allowing me to pay à la carte when I hit usage constraints on "pro" subscriptions. There are open-source tools where you can plug in these API keys for chat interfaces and other features, like querying multiple models and synthesizing responses. I use big-agi.com but I've also heard good things about libre chat.

Context Gatherer

• Recon: Gathers info from various sources into a buffer for LLM input

• AI-digest: Focused on getting your codebase into a compressed markdown file for LLM input

• Helps provide sufficient context for better AI responses

As I have mentioned, providing sufficient context for your queries helps improve the value of responses. I've used recon which can gather info into a buffer in order for it to be passed on to an LLM. I've used the clipboard to ferry info but you can also pipe it to a CLI LLM front end as well. You can query dbs, pass whole directories of files, or even pass the contents of urls.

I've recently heard of ai-digest which I'm eager to try out. This seems focused and taking the contents of your code base and concatenating it into a single markdown file. This will be particularly helpful in using claude projects for coding. Keeping the project updated with the state of assorted files is a point of friction that

I'll also say here that I've heard lots of good things about cursor.com, which is an AI powered IDE which, because it has your whole project open, also manages that up to date context for queries. I'm trying to balance experimenting with all the tools while also not allowing the number of subscriptions I have in play to get absolutely ridiculous. Constant struggle.

Note Taking: Granola.so

• Mac app for transcribing and organizing notes

• Listens to audio piped through your Mac

• Customizable note structure

There are lots of little bots that have to be integrated with your calendar and company. This isn't that. It is a Mac app that listens to the audio being piped through your Mac. It transcribes and organizes notes, taking cues from any structure you would like or things you may tap out quickly as the call proceeds. Still on the trial for this but have every intention of subscribing later.

Images: Midjourney

• Generate blog post cover images

• Experimented with generating culturally-specific art

• Cautious about accuracy with complex cultural imagery

I use Midjourney to generate the cover images for these blog posts. I have to say, I don't think I'm as aware yet of how to deploy this segment of tools in a material way. My cover images are low stakes. It's helpful there, but it could be more. I've played with it to try and generate collectible art in country specific styles. I got a little leery of using it to generate something I couldn't objectively assess the outputs of. Sure, I can tell if the fingers on the hands look alright or if the letters on a sign are garbled, but when touching upon culturally significant imagery and symbolism… I got scared off. On that Mexican inspired art, is that a puma, a quetzal, or is it just an AI hallucination? I didn't have the expertise and I couldn't count on it.

What's in your bag?

The AI space is evolving rapidly, and so are our tools. I'd love to hear from others about their discoveries or challenges with existing tools. If you're having trouble unlocking the potential of any of these tools, I'd be happy to help!