My journey from AI Neophyte to Apprentice, Part 2
In my previous post, we looked into the background of the project that I've been building with AI as well as some of the tools which I landed on to build it. Specifically - more general AI. If you missed it and somehow found this link, it can be found here.
In this post we'll dig into what I'm calling "tech + AI" - that is solutions that already did great things but have been enhanced with artificial intelligence. We will close out with my lessons learning and thoughts on how Sales Engineers can take advantage of these capabilities today.
Technology Enhanced with AI
Before rolling into this list, a word of warning - don’t sleep on this category. There are absolutely examples of AI features added on to existing technologies that kinda suck and seem to exist strictly to impress “The Street”. These are not them. The tools in this list are incredibly powerful at what they do. The one at the bottom is mind blowingly awesome. They do as much if not more to my productivity building out our goal than the Pure AI in the last.
Firebase Gemini
Yes, I know this is more of a general AI but that's not what I'm using it for.
For my development, I’m using Firebase as a data store. Easy, extensible, fairly cheap (or free). It also has Gemeni which is a lifesaver for people such as myself who have never been fans of little things like “security” or “firewall rules”. Great conversationally and can provide code and rules snippets. The thing I like the most about Gemini is it is an expert in its space. It replaces searching reddit or stack overflow in an arena that I’ll never have much knowledge in. I’ve found it to be much more accurate than the pure AIs in my previous post on topics related to its sphere and quicker to get me the answers I needed. Obviously Gemini can do so much more than what it's being leveraged for in my project but where it has positively impacted me, it just makes it too easy:
“Gemeni, my project is designed for <stuff here>. Right now, we are in development but eventually it needs to be able to <more stuff here>. I’m seeing <roadblock>. How should I approach <ask here>?”.
By the way, it can also be used to help build content in Google Docs. Just haven’t explored that much yet.
Replit
Replit is a cloud IDE that makes it simple to develop and deploy solutions. It has easy integrations to a number of other platforms such as Github. It will launch dev instances locally and does production deployment with ease as well as having its own Claude and GPT integrations. Frankly, I’m not touching the periphery of its capabilities. It is so far advanced from what was in use back in my day it could be considered science fiction. More Star Trek than Dune though, thank goodness.
Replit has two different AI tools: Assistant and Agent.
Assistant is a great way to get quick ROI from AI on existing codebases. It’s focused on short iterations and improvements of projects and can participate in troubleshooting or feature enhancement. It’s also knowledgeable about its own IDE and capabilities akin to where Gemini has been invaluable, with the ability to answer questions that otherwise would result in alt-tabbing. It’s great at helping me hunt and peck around my environment as that has steadily has become more complex. And it is fantastic at serving up to me (still within the IDE) valuable information from the docs customized specifically around my requests. Finally, it can also rapidly turn small requests from natural language into code that can be slotted into the base. Consider it your lightweight assistant to help in a project.
Agent is new and to be fully transparent, I haven’t played with it nearly as much as I’d like. Presently, it must be embedded in a Repl (the term for a code repository built in Replit) from the beginning and my current projects were underway before the Agent started hitting its stride. However, what I’ve seen makes me want to explore more. It can set up environments, check and install dependencies and supposedly even write automated test cases. And it is intended to be fully writing, deploying, and rolling back code akin to the capabilities that pure AIs provide. This technology is brand new and became accessible within the last month but has significant potential.
Simply put - Replit is excellent without the AI and amazing with it.
Better Dictation
Let’s take a hard left turn here on the application of technology because this tool has nothing directly to do with coding.
What it does do is turn your voice into text with context, punctuation, and emphasis that is far more accurate than other voice dictation tools I’ve messed with. To continue the earlier sci-fi reference, it is like that scene in Star Trek 4 where Captain Scott just expects the Apple 2c to understand his spoken language. Not in that it will respond and hold a conversation. But it can consume, understand, and (in our case) replicate that thought process onto the screen. Lightyears ahead of Siri and able to be leveraged in so many apps. They claim “all apps” but I can only speak to my use of it. It handles dictating to Google Docs, Chrome, Safari, Replit, Studio, Mail, etc like a champ.
It does require a Mac with an M chip though. So if this description piques your interest it might be time to add something onto your Christmas list.
Don’t sleep on this one if you have a newer Mac. It’s also stupid cheap for a permanent license. Like $25.
Privacy, unfortunately, is a concern. While most of these tools have some type of implication that you should totally engage with your legal team before using for professional purpose, tools like Better Dictation may have access to your entire computer instead of just the information you give it. One of its competitors recently was front and center in a tiff on Reddit that absolutely didn't inspire confidence. Better Dictation's privacy policy is pretty awesome as far as modern technology goes but it's worth a callout. Props to my editor and close friend Conor for this callout.
Look into it, draw your own conclusion. BD is A+ in my book.
Cursor
Best for last here, but let me start with a bit of a divergence on how development with Gen AI progresses from my experience. At first the AI is dictating entire files for you, sometimes even multiple files based on your prompts. As your codebase grows and your files become more complex, the AI wants to start providing chunks of code to be slotted into the appropriate file, relying on you to place them at the right place in your code as well as remove the right chunks. This is useful for two reasons -
- It saves effort on the AI’s side, which is in turn quicker for you
- It forces you to review and independently apply the changes, thus validating the accurate implementation of your needs (should you be paying attention)
By itself, GenAI is a jump to lightspeed in the development cycle (though there are dangers that become more pronounced as one’s codebase grows and becomes more complex).
At its core, Cursor provides some impressive code generation and code completion capabilities, often predicting sizable chunks of code for me rather than just anticipating a line or two as other IDEs may. It does a great job at refactoring and automatically fixing “loose nut between the chair and the keyboard” issues that can be time vampires especially in a loosely defined language such as Javascript.
What Cursor does that has brought my development from lightspeed to ludicrous speed is its Composer AI, which sits on top of Claude to not only generate code changes based on requests from its human but also apply it in the correct spot, thus removing that element of human error from the equation. It has impressive change management which can (hopefully) ensure the human is still reviewing, editing, and approving the changes.
Like the Spaceballs reference, it does rely on the human to understand what it's asking for compared to what it got. But when human and machine are in sync the results are incredible. This tool is money and I’ve fallen head over heels for it in just the weeks we’ve been together.
A few useful general lessons learned
Before we close on out with how all of this can be useful specifically for Sales Engineers, here are three general lessons learned:
AI works best when treated as a partner. You probably picked it up in my Claude commentary. Some of them can absolutely generate sizable chunks of content on their own but where they can truly excel is in co-creating “stuff”. Your knowledge and background is just as key as the AI's training. Have a conversation with it. This results in building finite, verifiable elements that are slotted in and validated, then tinkering with until it is just right. At least for now - never ask an AI for something and just assume it got the right answer. The results can be bad. Leveraged correctly though, the results are excellent.
Contrary to the advice in the previous paragraph, AI tends to be billed by transaction. This encourages one to submit fewer, more complicated requests. Ergo, balance is important - asking for multiple similar items when generating code tends to be less error prone and easier to validate as a human. But dissimilar or complex requests can be problematic.
Example - I once submitted a request to resolve a pretty straightforward bug while also asking for additional feature development. The results were not great and took me a little while to unravel. For clarity - the issues were totally my fault. I try not to think about the costs (sometimes $0.10 per request or more). What works for me is to approach my requests strategically, trying to think three or more steps into the future so as to better structure them. Sort of akin to a game of chess.
And lastly…while everyone learns differently, learning how to leverage AI has worked best for me with specific problems to solve and things to build rather than just general tinkering. Coming to understand the value of each individual tool by using them and later its limitations has guided me to discover new tools. Without application almost all AI solutions and technologies have a tendency to blend together. If you’re at a loss finding an AI tool to support a particular need, just wander over to Perplexity and ask it for an AI tool that can help with “X”.
Why is this important for Sales Engineers?
But this is an SE blog, Paul! It is not supposed to be for developers! Heck, I got into sales engineering because I was tired of coding!
(me too btw)
For starters, notice that about half of the tools mentioned aren’t actually coding? Not to oversell, but Perplexity has literally changed my personal habits. Half my commentary on Claude had more to do with writing than developing. Go back and look.
Also - maybe you don’t want to be a developer but these tools can enable you to be when it is valuable. If I, a people manager for almost twenty years now (ugh), can quickly make real progress on building something using modern technology, imagine what you - a real engineer - can do? Y’all are all more intelligent than me to begin with. Many of you grew up with iPhones, adopting tech is in your blood. I had the indestructible Nokia brick, but that wasn’t until after college.
Seriously though - a few key thoughts just for SEs, applicable to our jobs:
There is literally no excuse not to do customer research any more. What Bing Chat could do 18 months ago and Perplexity, CoPilot, and Claude can do today empowers us to learn so much so quickly in a focused fashion about our customers, their market, their competitors, their needs and problems, a new type of technology or breakthrough…or anything really. It takes minutes, or less. The only thing preventing us from walking into that conference room (or Zoom) truly knowledgeable on the key topics at hand is ourselves.
Next, while most of this post has been about my journey to develop something, the modern tools can quickly make our presentations and documents better as well as provide feedback on our work to rapidly accelerate our productivity on the non technical elements of our job. Or make cute dog icons to throw into a post. Generating and sharing documents is as integral to our roles as SEs as looking smart and saying smart things. I also believe that Claude provides a more reliable feedback loop than my peers who are just as busy as I am and is significantly more timely.
Moving on - leveraged correctly, AI can do magic things for when we do need to code or develop. With what I’ve learned (and again, I’m just a dumb people manager), developing customer tailored demos or giving customers a quick head start on their proof of concept / proof of value enables our customers to more quickly see and experience the value of our products and services. It can get many technical hurdles out of the way in conversations and activities so we can focus on understanding and solving their problems which is where the real fun is anyways.
Lastly - don’t miss this boat. I’m not sure how long the Island of Pre-AI Toys will be above water, but the tide is rising quickly. AI is bringing a paradigm shift of epic proportion to our industry. It’s not like Big Data was a decade ago (cool stuff, but mostly lots of smoke and little fire). This is real and it is coming fast. I know we’re all busy. Frankly, we’re all stupid busy and overworked in this market with steadily increasing demands that we do more with less. Seriously - hop on the boat and grab an oar. After some initial investment your return will be worth it in increased performance and less repetitive tasks. And I promise it will be fun.
In Conclusion
This post was loooooong but hopefully a worthy read for you. These solutions are just the ones I’ve found valuable and are now leveraging almost daily.
What AI tools have you found valuable? What are you using them for? I’d love to hear from you and learn. Also - what did I get wrong in this diatribe? As implied in the title of the post, my skills make me still only worthy of the title "Apprentice".
A few quick notes…
First, apologies for the delay in getting more content out. Life overtook me and I’m only just now coming up for air. Raising our two boys while being fully invested in their lives has been challenging for two parents with full time jobs. We prioritized that over other activities, such as having a blog. Also - Baldur’s Gate 3 happened and ate giant chunks of my free time.
In the future, I will make a concerted effort to have content at a reasonable frequency. That frequency should be good for a little while as there are presently 5 topics at different stages of development for my blog.
Next - the AI landscape is evolving faster than a speeding bullet. This post was finalized on 12 December 2024. If you read it more than a few months from then, some of the content could be wrong or outdated.
Finally, with AI and this content I am totally standing on the shoulders of giants. My acceleration on this topic has been absolutely driven by what others have shared. The list would be too long to reference them all but thank you to the podcasters, youtubers, and vloggers who are out there building content to help folks level up. I wish you all success.