In previous posts, I gave a high-level overview of personal and team-based AI tools. These are off-the-shelf tools that are easy to adopt, often free or inexpensive, and can deliver immediate productivity gains. However, they share one common barrier to delivering company-wide dramatic productivity gains: They are standalone tools that aren't directly integrated into your workflows or business.
This week, I'm going to take a brief detour into the way I use personal AI tools. I spend a great deal of money evaluating AI tools every month so that my clients don't need to, and I wanted to share my current AI tool stack with you as well.
Chatbots
I use chatbots every day, primarily ChatGPT but also Perplexity and Claude. They've largely replaced Google for my searches unless I need to search for something current (the models have a lag in how fast they ingest and process data).
A game-changer for me was organizing my usage by chat. All the chatbots will save your past queries under something like "Your Chats" or "Library". Instead of creating a new chat every time, I take a related previous chat and put my latest query in there. That provides the chatbot with context on your previous searches for this subject matter and greatly improves the quality of the responses.
The other key thing I do is prompt engineering. This topic could be a whole series of posts itself, but the take-away is that your prompts to chatbots are probably way too short and undetailed. The next time you're asking a chatbot for some information, leverage the chatbot to help you build an effective prompt. Start with a prompt like this: "I'm going to be looking for a vacation based on preferences that I will share. Craft a prompt to accomplish this search. Include a persona that will make you act as an expert in this field. This persona must ask me for my preferences and ask follow-up questions to clarify my intentions." I think you'll be surprised by the prompt that the chatbot gives you to use—and by the results!
A cautionary note: By default, all chatbots will take your data and ingest into their systems to use how they wish. I will not use a model that doesn't let me turn that off, so the first thing I do with any chatbot is look for the "Allow your content to be used to train our models..." and turn that setting off.
Automating the Web
Oftentimes I need do some tasks or research involving several different websites. Agentic browsers are a relatively recent development that let the AI system take over your browser and act as you on the web. That sounds dangerous, and it is, so I recommend supervising them if you're asking it to actually do things on your behalf.
I use these agents with prompts like "Check my Gmail and LinkedIn accounts for recent messages; give me a summary and let me know if there are any I need to follow up on ASAP" or even tasks like "Find the 5 highest ranked pizza shops in Hamilton based on 3rd party reviews. Compare their menus for a meat-lover's pizza; identify the best value, then create an order on their website that I can complete with my information." The agent will then work away (you can watch it or just let it do its thing in the background) and complete the task for you.
Claude has a browser plug-in that gives you an agent right from within your browser, while Comet (from Perplexity) is a standalone browser. I use the Claude plug-in the most simply because it is right there in my default browser. However, because this technology is still young and can provide uneven results, it isn't unusual for me to give the same task to both Claude and Comet to compare the results.
Data Organization and Retrieval
I typically have a lot of different projects going on at any one time, and it's always a challenge to stay on top of them. There are several tools that let you organize documents and then connect them together, acting as an AI research analyst that sits on top of your private material. I use NotebookLM (from Google) for this.
Within NotebookLM, you create notebooks and then upload your documents or link to related websites. I create one notebook per project and upload all my project-related documents into it.
NotebookLM has tons of features, including mind maps, audio and video overviews, and (particularly useful for students), flashcards and quizzes. However, I use the query functionality the most. That's where I ask a question and it provides an answer based on the full corpus of knowledge in that notebook. The free version of NotebookLM is good enough to get started with (100 notebooks, 50 documents per notebook), but if you have a lot of material to organize, you'll find that you need a paid plan.
As a free plug, if you're looking for an enterprise-class AI expert for your business's policies, procedures, and other documents, I have an app that does this on an enterprise scale. It acts as the go-to expert for the corporate knowledge you want to make available, providing fully cited responses to queries along with full analytics, access controls, unlimited documents, audit logs, and more. Feel free to ask for more information!
Coding and other IT tasks
For coding and other IT-related tasks, there are a couple of major categories. The first category is app builders (often referred to as "Vibe Coding Platforms") that can generate full apps from prompts. Some of the leaders in this space include Lovable and Replit AI. I want a lot more control over the process than they offer, so I use products from the second category, AI coding assistants.
AI coding assistants are typically command-line based tools that can generate code for large multi-file projects in all major coding languages. I use Claude Code the most, although for some specific use cases I will use Codex and/or Cursor.
I want to highlight that these AI coding assistants are not just useful for writing code. If you spend a lot of time at a command prompt, you should be using an AI coding assistant. They are invaluable for all the things you do at the command line including scripting, administrating servers, and managing Internet domains.
Images
A picture speaks a thousand words, and sometimes you just need a picture. AI image generators have come a long way recently and are able to generate useable images from just a simple prompt. At this point in time, I think there's only one choice for image generation—the interestingly named "Nano Banana", accessed by going to the Gemini chatbot and typing in a prompt such as "Generate a picture of someone using an impressionist style. The person should be at a computer, hunched over typing. Add some kind of element to the picture to make it AI relevant."
Conclusion
The state of the art in AI is changing at a terrific pace, certainly the fastest I've ever seen any technology change in my 30+ years in the business. It's going to be interesting to review this post in a month and a year and five years to see how my tool set has changed.
In my next post I'll return to looking at how you can use AI in your business as part of a toolkit to deliver productivity gains and cost savings, starting with simple automation tools that offer a practical way to implement real process change.