My journey from AI Neophyte to Apprentice, Part 1

My journey from AI Neophyte to Apprentice, Part 1

Several months ago I started actively engaging with AI.  Not…assigning someone else to explore and implement something or reading artificial intelligence focused posts and watching videos.  Not relying on an expert to explain the capabilities and ramifications of these solutions. Actual honest to goodness leveraging them myself.  There was a goal - my partner and I had come up with an idea a few years ago that we were dying to get off the ground.  I figured that AI might help us get started.  As an added benefit, this might prevent me from being left behind in the shift presently hitting our industry.  Lastly…it sounded fun.

Because of this post’s length, we will break this up into parts:

  • The first focused on some background information and the more generalized AI that have impacted my project as well as how I’m using them.
  • The second sharing the solutions that are more “technology enhanced with AI” that have been helpful as well as a lessons learned and thoughts on how Sales Engineers can take advantage of my experiences

A little background

Our goal, as in what I’m personally building, would require a great deal of web development.  For those that don’t know - I am not a web developer.  Back when coding was my living, I despised JavaScript.  Too unstructured…too wishy washy.  And don’t get me started on that multiple versions of different browsers implement things in different ways BS.  Give me a good, complex algorithm to develop, ETL to manipulate, or automation to implement and I am there with gusto. But web dev… <shiver>

When we’d originally had the idea, I’d tried to learn React and Next to build out our goal but found the task daunting.  Eventually the craziness of life caused me to shelve it.  We still talked about the idea and how cool it would be on occasion but had accepted that it wasn’t in the cards without something else to assist.  We’d even talked about hiring an outsourcer to get the development started since I was lacking.

That brings us to AI.  With the capabilities that modern AI brings, I thought it might bridge the gap.  Spoiler alert - it has.  In fact, the more I leverage these tools the more they accelerate my ability to create.  The more I create, the more AI impacts my daily life. 

AI has accelerated my ability to write code by 20x as well as enabled me to build in arenas I am painfully inept at (Javascript).  It has also sped up my ability to learn, research, create document, write blogs, etc.  Want to learn more?

Keep reading.

So what am I playing with?  

Let’s break this down into two categories: “pure AI” and “technology enhanced with AI”.  Why is this an important distinction? Every software application on the planet is now claiming to use AI.  Some of the “technology enhanced with AI” are incredibly powerful.  Others are… windows dressings for marketing. 

The AI making headlines tends to be the Pure AI such as ChatGPT rather than the “tech + AI”.  That coverage gap can imply that the Pure AI is applicable for everything when it’s not. They are like a sedan as a general family hauler vs a sports car for track day: one is broadly applicable to many challenges and very very good in some fashions, the other is focused in specific areas.  It's a little more complicated than that but hopefully that analogy provides some context.  Let’s treat all these solutions as tools in our tool belt and move forward with this categorization. 

While I’ve experimented with twenty odd tools over the last few months, here are the ones that have been most useful for my project:

Pure AI

Claude 

Claude has become my primary development partner.  Through our conversations, Jarvis (which is what I call my instance of Claude) and I rapidly develop stuff that previously would take me days or weeks.  There are limitations, of course, but Claude excels at complex reasoning tasks such as coding.  It’s a great partner when writing.  It can provide feedback on existing content (such as this very blog post, which it rightly says is too long) and can even be used to adjust the voice to drafts (“Jarvis, please make this blog post read more professionally”). 

By the way, Claude / Jarvis is not just smart in what it produces, it is also extensible in how it interacts with its users. When I asked it whether it minded if I referred to it as Jarvis, not only was the AI intelligent enough to understand the reference but also noticeably changed its intonation and sentence structure from then on to match how Paul Bettany speaks in the movies.  I now hear his voice in my mind every time Jarvis and I interact.  It’s wild.

With respect to coding, Claude can take ideas from concept to inception incredibly fast and does a pretty reasonable job at debugging.  There are limitations, potholes and dangers that can jump out and grab you, but even with them the productivity gains are giant.  By the way, that “potholes” callout is foreshadowing for my next blog post.

Why Claude over OpenAI?  Both work well.  I found OpenAI to be better foundationally (helping one get started with a project) and Claude better conversationally (steadily building on and improving an existing project).  My impressions could be wrong but what ultimately led me towards 100% Claude was my subconscious.  My thumb and pinky just developed a tendency to go to Claude’s tab on their own over time.  There is something there even if I can’t put my finger on it.  

For clarity - right now I'm working closely with Claude 3.5 Sonnet though occasionally using 3.0 Opus for non-coding tasks. My trialing with OpenAI was with GPT-4o.

Perplexity 

The answerer of all questions.  Seriously.  

Perplexity provides answers to any questions you ask with up to date data points.  It provides citations for everything. Most of the AI models build off of data sets that are not recent.  GPT is presently working off of data from December 2023, while Claude’s is from August of that same year.  That’s not bad when what you want is to take advantage of the AI’s training but is less useful when looking for collated information. Perplexity leverages multiple other AI models for its processing and analysis with a focus on being fact or information based.  

Lastly, much of the time it anticipates your next question and prompts you to ask it.  So even when you think you’re done there is a breadcrumb that might take you further down the rabbit hole.  As a five whys kind of guy, this is excellent and informative.  And powerful. 

Perplexity has completely replaced Google.com in my daily life. 

Also, less ads. 

Copilot 

Copilot brings a ton of value to the table if you’re living in the Microsoft Office / Visual Studio world, which I’m not (we’ll get to what IDE I’m using eventually).  Much of its capabilities focus on productivity enhancements or gathering and analyzing conversations, emails, and other data sources to handle inquiries.  Again, not what I’m doing most of the time but it is pretty solid at taking content from diverse sources to build out entire presentations or documents.  

Like Perplexity, it can also be used for general, real time queries, though I’ve found Perplexity’s results more valuable.  While this could be how I’m using it, Copilot tends to work better in single engagements rather than a conversation.  It does have a feature that has been useful for me - image generation.  It’s awesome at that, especially images in a theme.  For example, here are some icons Copilot generated for me:

In actuality, Copilot sits on top of Dalle to generate image content from prompts.  It’s not as powerful or flexible as Dalle directly but the interweaving of a short conversation to figure out what I want followed quickly by the creation of the artifact is fairly seamless and easy.  

That may sound “unuseful” but think about how many times we SEs need to build presentations. Would it be more valuable to spruce them up with contextual imagery and provide a theme or just use that three year old template our marketing team dreamed up for a conference once?

One of my former SEs is a pro this.  Earlier this year in about fifteen minutes, he augmented an entire hour-long training session with images of ninjas doing various tasks related to the content he was presenting.  That minor sounding tweak added tremendous impact to the visual content he was presenting.  There are other AI image generators out there, but this is the one I’m leaning on at the moment.  

That's all for this post! Click on next to continue with more specific AIs that have been useful as well as lessons learned.