My Current AI Workflow
AiSometimes, after rebutting someone’s extreme exuberance over AI or cracking a joke, I feel like people may assume that I don’t like AI or I never use it. That’s not the case at all! I do use it and I integrate it (or advocate for its use) in my CTO role at Ten2 and when consulting but I do take care to use it in situations that make sense, such as text generation, and not places where it doesn’t make sense, such as decision-making.
Because of all that, I thought it was long past time that I share my AI workflow. Specifically, what tools and services I use, how I use them, and what I think they do for me.
Services, Tools, & Apps
- GitHub Copilot - $8.33/month
- ChatGPT Plus - $20/month
- API Usage - ~$1/month
- Google Gemini - $0/month * Used through Ten2’s corporate Google Workspace or Google’s free tier personally.
- Zed
- ChatGPT App
- OpenAI Codex
- MindMac\
- Ollama
- Web Interfaces (ChatGPT, Gemini, etc.)
- Custom Apps
Yes, I am very OpenAI-heavy. That’s not really a choice about the models themselves, as I think by and large they are all extremely close together in functionality and abilities, but more of a choice of how to consume it. I do use other models, Claude in particular, through GitHub Copilot, but the integrations OpenAI has created fit nicely with the way I want to use the models.
A couple of other things to call out:
- Zed is my code editor of choice, and it’s great outside of any AI features, but it does now integrate some nice AI features. Primarily this is in the form of auto-complete/suggestions, which I have wired to Copilot, and “agentic” coding via an assistant window. I’m using API keys to power that, but I haven’t gone full “vibe coding” for any project yet. Zed also has an AI service that I could see myself moving to if Copilot ever pushed me away.
- MindMac is a macOS application that has a solid idea behind it: Plug in your API keys and house all of your conversations in one place. It has been great and actually earned a place in my macOS dock…but…updates have slowed dramatically lately, which is not a good sign. I find myself using the ChatGPT app more since I started paying for their Plus service anyway.
- Ollama is used to run models locally. I don’t do a ton of that now that things are a bit more mature, but it’s helpful when I want to experiment with something.
- My API usage, outside of the occasional test or project, is for my home-brew JARVIS voice assistant in my home office powered by Home Assistant.
Uses & Benefits
I suck at Photoshop. I’ve learned to be barely passable with Pixelmater on my Mac. AI’s greatest gift to me personally is the ability for me to describe something and it can make generate something close enough that I can bring it to someone (or occasionally I can do it myself) and take it the last mile into something usable. The new image model from OpenAI is great and the $20/month is almost worth it for me to not have to wait as long for my image alone.
I don’t use AI for general search. I like it when a prompt can utilize live web results rather than just data from when it was trained a year ago, but I don’t want an LLM filtering results or making any kind of “decision” regarding web search results, and neither should you.
I do use it for search-ish low-risk tasks. For example: “How do I make a Paloma?”
For work, I will use it to flesh out a paragraph into something like a product spec for a task, and it’s amazing for generating test data, or for text processing. I don’t have a lot of these tasks, but I also want to spend as little time as possible on them when I do, so it’s a win there.
OK, let’s talk about coding.
I use multiple services and apps for different parts of the coding process, but it’s best to group all that usage under the idea that AI is my Jr. Developer assistant. This Jr. Dev Assistant:
- Is good at some things, OK at others, and downright awful at a list of things. As the “manager," you need to know what those things are!
- I don’t have to mentor or teach it anything. This saves a lot of time, but obviously would have major downsides if I was in the position where I needed to deal with staffing churn.
- Like all technical managers, I am constantly working on letting myself delegate tasks to my Jr. Dev Assistant.
Currently, AI handles tasks for me like the following:
- Figure out that I’m setting up this very long switch statement and type it out ahead of me. – Smarter auto-complete seems so old now, but it’s still as useful as anything AI does.
- Go look around the repo and submit fixes to any bugs you see. – Mostly they are superficial bugs, but they are still bugs!
- This gives me this error: . Can you find it in this file?
- Make this “bottle” function that I can drop into the project myself. – Greenfield, new coding work for simple goals is pretty consistently good, so I will set up situations for “new” code when I can drop it into a more complex project.
The overarching summary is: It’s helpful, and I could definitely use it more, but I’m not worried about the livelihoods of developers.
I haven’t done much of anything in the way of full agentic “vibe coding,” mostly because I can code, but also because I haven’t needed any of those small, great for demos-type projects. If/when I do, I will give it a try, but for now I think that’s a lot of smoke, in that there’s definitely a fire, but it’s not as big as it seems from the other side of the hill.
Conclusion
AI is a powerful tool that can be used in a multitude of ways, from text generation and coding assistance to image creation. While it offers significant benefits, it’s crucial to use it judiciously and understand its limitations. It’s clear to me that the hype is outpacing what it can actually do, and that we’re already seeing the limitations of LLMs, so will I continue to call out and poke at that breathless hype monster? Absolutely. But that doesn’t mean there aren’t valid uses, and I will continue to use them, experiment to find others, and enjoy the benefits of this approach for a long time.
If you have any questions, comments, or AI-generated love notes, please send them my way!
Related Posts
Software Development is Dead, Long Live Developers!
“If you don’t learn to code, in ten years it will be like being illiterate!” That was what someone exclaimed on a panel discussion I was on in 2013. It was a talk about bringing technology and entertainment together held in Beverly Hills and hosted by a coding bootcamp company. Two of the people on the panel were from a different bootcamp company, and then there was me, an actual technologist working in entertainment.
Read moreRethinking My Email Workflow
My previous email workflow sucked and I’ve been working to find a new system in 2022. I knew the old workflow sucked as it was based entirely on unread status, a notoriously shaky foundation. In short, if I forgot to get back to you about something last year, I tried my best but it was probably my workflow’s fault.
Read moreLinkedIn is Going Great
🚨🔥AI will obviously solve everything because hype is never ever wrong. Do you want to know a thing that will help you?! I have an AI thing that does stuff like, I don’t know, let’s say sales or marketing…whatever. It’s amazing though and it won’t be replaced in a matter of days by a new model.
Read more