How to think about AI.

Almost daily I’m asked “Drew, seriously.. where do we start with AI? How should we be using it? How should we even be thinking about it?”

This is a question a lot of folks are tortured by with a mix of fear and FOMO, which makes sense. The ground is shifting under our feet daily. An earthquake of tectonic proportions. Tools, ideas, and content are plastered on every feed and mentioned in daily conversation. It’s overwhelming.

Like most things, you need to walk before you run. You can’t dive into building Agents if you don’t know the fundamentals.

Here’s how I’ve been approaching with our team and clients at StealthX.

First - You’ve got to hire, grow, and invest in people with high agency. This is a non-negotiable. If you have low performers on your team with low agency, you may need to make some hard decisions.. 😬

Second - Make sure you and your team know the core AI building blocks/capabilities (and continue to pay attention to what’s coming). You don’t need to know every tool that comes out. That’s literally impossible these days. BUT, you do need to be aware of the core capabilities that are available. For example, GPT-5 came out this week and marks a pretty major shift in artificial intelligence. You should know this (just saying).

Third - Build a habit of constant experimentation. Push yourself and your team to always be testing. No Skillpop or Udemy class is going to teach you more than hands-on testing/playing.

Fourth (and most importantly) - DO NOT OUTSOURCE YOUR BRAIN. Be unrelenting on this. You cannot afford to mess this one up.

Let’s dive in.

1. Hire for & instill high agency into your team.

High agency is the absolute most important trait I hire for and develop. It’s the bias to make things happen. It’s the person who believes with conviction that “I don’t know, but I’ll figure it out.”

People with high agency do not wait for perfect inputs or let perfect be the enemy of great. They don’t wait to be asked. They dive in and work the problem and figure out a bunch of different ways to tackle it. They move through blockers with creativity and ownership. They understand that constraints are normal and don’t make excuses.

In a fast-moving environment where things are changing daily, I cannot stress enough how important this element is.

If you want a great primer, read about High Agency.

Putting this into action

• Hire and coach for high agency. Ask candidates to walk you through a time they changed a decision with no authority. Look for verbs (e.g., “I built this thing. I tested XYZ. I tried these different options.. 2 didn’t work, but 1 did.”).

• Set outcomes, not tasks. For example, “Close the deal by end of the month.”

• Make constraints visible and ask questions like, “Given _______, what can we do by Friday?”

• Push for tangible prototypes and demos, not fluff. Track shipped experiments.

Learn & teach the core building blocks

Your team needs a shared map of what AI can do. You need to have a handle on the core capabilities so that you can more quickly/easily apply them to problems in your context.

Here’s a set of 12 simple building blocks that anyone can pick up and combine to solve a problem. The goal here is to get you and your team to internalize these so it’s practically muscle memory.

1. Search across your data/content/docs: Find the right info quickly across docs, tools, and conversations. This is how you crush tribal knowledge and make answers repeatable.

2. Summarization/synthesis: Collapse long calls, threads, and docs into what matters. This reduces clutter and improves decisions.

3. Extraction of structured fields: Pull clean fields from messy text. Names. Order numbers. Complaint types. Dates. This makes automation possible.

4. Classification and tagging: Sort content into buckets. Priority. Intent. Sentiment. Language. This helps routing and reporting.

5. Rewrite and translation: Tailor tone and language without losing meaning. This lifts quality at scale.

6. Question answering over knowledge: Ask natural questions and get cited answers from your team’s data/content. This is different from search. It returns a direct answer with sources.

7. Planning/outlining: Break big work into steps. Create drafts. Clarify scope. Help the team start faster and finish cleaner.

8. Code generation & scripting: Generate small tools and apps focused on specific problems your team has regularly.

9. Image understanding & text on images: Read screenshots, explain charts, transcribe/convert into text and transform into structured tables of information.

10. Speech to text & text to speech: Turn calls into clean notes. Create explainers you can listen to. This improves knowledge sharing.

11. Personalization/recommendation: Match the right message or offer to the right person. Even simple rules can drive outsized gains.

12. Multi step agents & workflow orchestration: Chain other building blocks and create agents that are capable of reasoning and function as AI employees focused on specific set of tasks in a workflow.

Schedule recurring jam sessions

AI rewards teams that learn in public. You need a system that makes tiny experiments fast and safe. I call it an innovation jam session. We run at least 1 every week at StealthX and have a long-form version bi-weekly. The rules are clear, the bar is low, and the cadence never slips.

Putting this into action

• Create a one page AI building blocks menu. Add your top examples next to each one.

• Run a weekly jam session. Pick mundane tasks and challenge the team to solve each in different ways.

• Keep a living backlog of what to automate next. Map by effort/impact.

• Budget small. Many bets. Kill fast. Celebrate what you learned, not just wins.

• Keep a running log of tools to test. Allocate budget purely for play and let the team tell you when they find something they want to try. If data/privacy is a concern, let them expense and use personally in a safe sandbox.

If you’re curious, I thought I’d share a list of the AI tools I’m using, testing, and excited to try next.

Tools I use constantly

• GPT, Claude, Gemini - Core models for all the basic building blocks above.

• Manus, Gamma - Use these to generate slides & 1-pagers.

• Lovable, Figma Make, Supabase - Vibe coding/designing ideas and quick prototypes.

• Fathom, Otter, Wispr Flow - Recording meetings and helping with quick voice-to-text notes and brainstorms.

• Notion AI - My team’s primary knowledge hub with quick access to everything in our ecosystem.

Tools I’m testing

• ClickUp Brain - ClickUp is a project management tool that recently rolled out Brain Max, which hooks into a variety of common team tools and gives you access to all the model providers in one space. So far I’m impressed!

• Warp - Warp claims it’s the fastest way to build with multiple AI agents with code in the terminal. Pretty intriguing so far, but not sure it beats Claude Code.

• Chorus - A single Mac app that gives you access to all models in 1 place. Definitely simplifies things, but still find myself opening each in the browser.

• Ilus - Generate custom illustrations for a brand. Pretty nifty for quickly concepting ideas for marketing assets and to use in product landing pages and onboarding flows.

• customgpt.ai - Pulls in data/content/docs from a bunch of sources including your website, Sharepoint, YouTube, Google Drive, etc. and allows you to create a white-labeled public facing chat experience.

• GPT agent vs n8n vs Zapier Agents vs Lindy - Constantly testing each of these to figure out which agent flow is best. Honestly, now that GPT agent is available (a general purpose agent) I find myself using it more for random thoughts/ideas and Zapier for basic workflow automation even though n8n is a lot more powerful/customizable.

Tools I’m planning to test next

• Perplexity Comet - On the waiting list for this one. Promises to be a new way to browse the internet. Veerrrryy intriguing.

• Akiflow - I’ve played with a bunch of the calendar/tasks AI tools and haven’t found one that quite nails it for me. Looking forward to testing this one out.

• Conveo - AI-managed interviews for qualitative research. If it does what it promises, it would save a ton of time during the research process (although I’m sure my researcher friends would hate me for saying this 😅).

• Outset - Similar functionality to Conveo. It’ll be interesting to see how they stack up side-by-side.

• V7 Labs - Another agent builder that promises to “label data at scale.” Very curious to see how this stacks up to the other agent tools out there.

• Onlook - The self-proclaimed “Cursor for designers,” a “visual code editor

  that lets designers and product managers craft web experiences with AI.” Interested to see how it stacks up to the other more common vibecoding tools on the market.

• TheyDo - A journey management platform with a “Journey AI” tool that looks very interesting for streamlining the typical journey mapping work that CX & UX teams do.

BUT… Do not outsource your brain

AI is a sharp tool, BUT it is not a substitute for critical thought. Brain dump into AI tools and use it to sharpen/challenge your ideas, eliminate noise, generate options, catch edge cases, and draft a straw man.

THEN apply your critical thinking/judgment/taste/craft. This is incredibly important and is the thing that will keep your team from falling down into chasm of AI slop.

When I review AI work, I ask a series of questions…

• What’s the ultimate outcome we’re after? What’s the real problem?

• Who’s this for? What are their needs/wants?

• Does this output deliver on these things?

• What’s missing? What will cause us to fail?

• What will make this UNIQUE, DIFFERENT, AND MEMORABLE??

Continually asking these questions through the process of creating anything with AI keeps us from making trash and focused on what’s most important.

Putting this into action

• Draft a brief first. Detail out the problem, target outcome, target audience, any constraints/dependencies. Then start riffing with AI while continually evaluating against the questions above.

• Check sources of anything generated by AI. Try the same prompt with a different model. Compare results. Do your own homework side-by-side. Vet with an expert.

• Keep a decision log. Record the choice, the reason, and the data. Revisit later.

• Ask AI to argue the opposite and look for blind spots.

Wrapping up

AI is already in the tools you use every day. It’s no longer a novelty, it’s the status quo.

Your edge will come from having high agency people who know the core building blocks, are constantly experimenting, and being a leader that refuses to let themselves or their team/s outsource their brain. Multiply these things together and you’ll get better experiences, faster cycle times, happier teams, and deliver real value you can point to.

Start now. Build the habits and skate to where the puck is going ⛸️

Onward & upward,
Drew

P.s. If we haven’t met yet, hello! I’m Drew Burdick, Founder and Managing Partner at StealthX. We work with brands to design & build great customer experiences that win. I share ideas weekly through this newsletter & over on the Building Great Experiences podcast. Have a question? Feel free to contact us, I’d love to hear from you.