Stop mixing concrete

Once upon a time, building something great meant doing it the hard way.

If you were building a house, you had to do everything by hand. Hauling bags of concrete, mixing it with water in a wheelbarrow, and pouring it yourself. That’s just how things were done.

But things have changed.

Just like modern contractors don’t mix concrete by hand anymore, we don’t need to do every task manually anymore. In this new era of work, AI is the construction crew. We’re the architects. Our job isn’t to do the grunt work. It’s to design better systems, make smarter decisions, and deliver real results.

Here’s how I’m approaching the process of building an AI-native, AI-first team.

1. Stop mixing concrete

If you find yourself doing manual work that a tool could do faster. Stop.

Mixing concrete is spending time on things like writing summaries, formatting docs, or building first drafts from scratch. For example, if you’re formatting a spreadsheet or doc by hand, pause and ask if AI can do it instead. I can assure you, it can.

Here are some specific ways I delegate “mixing concrete”:

• My team and I use Fathom.video to record most meetings, then feed the transcript to GPT and interrogate it for ideas, actions, summaries, requirements, etc. It’s crazy how good it is and how much faster we can move collectively because of this small shift.

• I will often open the Otter AI mobile app while I’m driving and do a brainstorm or debrief out loud. Once I’m stationary, I’ll feed the transcript to GPT to distill into a concise document that I can paste into Notion and share with others. What’s really crazy is you can also use Notion AI to generate ideas, actions, and summaries across all pages/content to get even more ideas and insights.

• One of my favorite time-saving workflows right now is to do deep research on an idea with GPT’s reasoning model, then switch models and give GPT prompts to turn the insights from the research into product requirements. From there I ask GPT to turn the requirements into a series of prompts that I can feed to Lovable.dev to build a working proof of concept to quickly vet the idea.

2. Identify loops/circuits & figure out which parts you can delegate to AI

Tools are changing faster than ever but patterns last.

Much of the work we do everyday are circuits or loops. Work flows between them. I’ve been challenging my team to think critically about their circuits and identify which parts can be AI-powered, and build a workflow flow of tools to optimize/enhance/accelerate.

For each type of task, experiment quickly with tools and define the best combination to get the most value. Challenge your team to share what they’re learning daily and iterate together to evolve fast.

Here’s a brain dump of some practical ways to put this into action:

• Break down a project into steps. For each step, ask yourself, “What’s feasible by AI? What’s human-essential?” Keep a shared doc or whiteboard with each workflow and tool flow.

  • ex. Create a “circuit” for journey mapping: GPT (deep research) → FigJam/Miro (mapping) → GPT (summarization) → Notion → (store/share outputs).

  • ex. Create a “circuit” for prototyping: Lovable (UI generation) → Anything.to.design (copy into Figma) → (Figma (refine UI) → GPT (specs for development).

  • ex. Create a circuit for onboarding. This might be GPT to generate a welcome email → N8N for automation → Notion for storing/sharing.

  • ex. Create a CX circuit. GPT to shape a proto-persona and journey → Lovable to create a quick wireframe or prototype → Gamma.app I to build a high-fidelity version.

• Host weekly jam sessions to share circuits and learnings. Make sure your team learns how to think, not just click.

• Keep a prompt library and update it as things evolve.

• Run “deconstruction” sessions. Take a finished project and ask, “What could we have automated here?”

3. Decompose everything

Big prompts lead to bad outputs.

I see a lot of folks give AI ambiguous, big prompts and then get frustrated with the results. You have to break work down into chunks and give the prompts one piece at a time. If you’re a parent, it’s a lot like talking your child through how to do something.

I often start with “Here’s what I’m thinking.. what am I missing?” and let GPT poke holes. I love to give it a rough brain dump and say “Make it clearer. Identify issues or problems in my thinking. Elaborate on key points.”

For example, if you want to quickly prototype an app idea with Lovable.dev or Bolt.new, don’t say “Build me an app for ______.” Outline the vision for the app, the problem(s) it will solve, who it’s for, and a breakdown of the main pieces. Give context like you would give to another person. Think through the target state user journey with flows, pages, and sections. I like to quickly sketch things out on paper, take a pic with my phone, and give it to GPT to fill in the gaps and outline a full sitemap and information architecture.

Fun fact, you can actually ask GPT to decompose big prompts into smaller chunks, which you can then paste one-by-one into Lovable/Bolt. If you’re writing prompts by hand, here’s an example of a bad prompt vs a good prompt:

• Bad prompt: “Add a navigation bar to the app.”

• Good prompt: “Build a responsive global navigation bar across the entire app. The links should be [list of links] that navigate to [list of pages]. Mimic the style of the attached screenshot using [hex value] as the primary color and [hex value] for hover states.”

Wrapping up

AI is like a teammate who never sleeps. Continually ask yourself what concrete-mixing tasks you and your team could/should be delegating to it from your day-to-day work.

Something I’ve noticed is a lot of folks feel mildly guilty by skipping the hard stuff. I think this is probably the hardest part of adopting an AI-first mindset because our culture glorifies hard work. You have to drop the guilt. This is the new way and the faster you adopt the mindset, the more effective you’ll be.

Catch yourself when you think, “But I didn’t do this by hand.”

Replace the thought with: “Did I deliver value faster?”

Celebrate when 2-hour tasks take 20-minutes because of AI. Use that extra time to create even more value.

You don’t need to build the whole house by hand anymore. You just need to know what you’re building and orchestrate the tools to get it done.

Let the machines mix the concrete. Focus on what matters.

More coming soon,

Drew

Bonus

I thought I’d share some illustrative CX prompts you can use as thought starters to help you think and execute faster. Feel free to copy and use however you’d like. If you find them useful - let me know! I’d love to hear 😊 

CX vision & strategy summary

Map current state experience

Identify experience gaps from data

Design a target state journey

Turn strategy into quick wins

Internal alignment & communication

Executive readout for CX initiative

Write a CX playbook section

Retro on a broken experience