AI should empowerpeople, not replace us.

I'm not anti-AI. I want to get that out of the way first, because what I'm about to say is going to sound like it's coming from someone who is. It's not.

I use AI every day. It's made my work faster, sharper, and more competitive. I've watched it get genuinely good at legal analysis, document synthesis, case research, things that used to take hours. I'm impressed by it, which is exactly why I can see where it falls apart.

New Article, possibly my last for a while.

I've spent two years figuring out how to make a two-person law firm compete with teams twenty times its size using AI. This is the closest I'll come to explaining how.

Also explains why I can type “plz fix” and get back work product that reads like I spent three hours on it, when really I spent three hundred hours building the system that did.

I built a full design system with AI agents as my team. At VibeSell we needed a design system. Brand colors, typography, surface hierarchy, component patterns. As a two person team with no designer, I was hitting a wall trying to do the research, make the creative decisions, and implement it all by myself. So I brought AI agents into every phase. Not just the code. The research, the brainstorming, the architecture, and the implementation. I made the calls. The agents did the heavy lifting. But it wasn’t added as chaos, it was added as a structured process. Here’s the setup. I use the BMAD method, an open source framework where specialized AI agents facilitate every phase of product development. The research phase pulls live data from the web. Market reports, competitor analysis, technical papers. Every claim is cross referenced across multiple sources, with confidence levels and URLs cited. This isn’t an LLM guessing from its training cutoff. It’s structured research with real sources. Then comes what BMAD calls Party Mode. Multiple AI agents, each with a distinct role, enter a group discussion. A product manager, architect, UX designer, and analyst debating design system choices from completely different angles. The system selects which 2 to 3 agents are most relevant to each topic, and they naturally build on each other’s points. The framework deliberately shifts creative domains every 10 ideas so you don’t get stuck in a local optimum. Every workflow has explicit choice points. I define the topic and goals. I confirm scope before detailed work begins. I choose which direction to explore next. The agents handle the tactical execution. But the strategic decisions stay with me. The process flows through structured phases. Discovery, planning, architecture, implementation. Each phase has defined inputs, outputs, and gates before moving forward. By the time we reach implementation, the agent has full context from every prior phase. The research findings, the design decisions, the architectural constraints. In the video below, you can see one of our agents using a browser to test and select colors for our background animation in real time. This is the implementation phase. The agent is making visual decisions informed by the entire design system spec that was built through the earlier phases. Two people. No designer. A full design system built through human judgment and AI execution working together. BMAD is open source. Link in the comments.

I see a lot on Linked In about AI being the end of junior software engineers. I don't think this has to be true.

It will depend on how and in what they are trained. The learning will need to be different- fundamentals will be more important, and we will have to go harder earlier at things like engineering principles and design patterns. As the code creation part becomes more automatic, the types of things that were learned over time mostly by experience and osmosis will have to become more explicit.

There will likely be a greater disconnect between what is taught in school or university and what the reality of the work environment is, and we will have to fill that. I don't see universities pivoting and adapting as quickly as work places are going to have to.

Some of our DI people are using AI tooling regularly within their client work, and some are not. We are running AI Enablement projects to get everyone to the same place. We started with our graduates, and it led to some interesting and varied outcomes, which informs us what we need to change and how we need to teach these different skills.

Juniors bring more to a team than just inexperience. They bring enthusiasm, drive, open mindedness, curiosity, and a different perspective- qualities that aren't always present in more seasoned engineers. The juniors with the right characteristics will always excel.

The best way to approach AI is to start by scratching your own itch. When I joined Sequel.io , we had a product marketing gap and no immediate plan to fill it. I was hired to lead marketing with a lean team, so I was effectively our new head of product marketing in addition to our new VP of Marketing. Rather than hire right away, I decided to see how far I could push AI before I hit a wall that required a human. Six weeks in to my tenure, we launched 12 products in 12 days. Every launch had a full suite of assets including a blog, video script, web copy, emails, battle cards, sales enablement, CS messaging, FAQs, the works - all created using the AI process I built to solve for my own workflow needs. It was an intense pace of launches and over those 12 days, I learned a lot about what AI could - and could not - do well, and how I needed to iterate on my process to produce results that felt useful. Today, I have reduced what used to be a week's worth of work to about 10 minutes - a huge win that has allowed me to keep pace with our product team's ambitious launch schedule. I shared this story a few weeks back on Lisa Sharapata 's AI Exchange thinking that it was basic, at best, and was surprised when many of the marketers in attendance reached out after to thank me and say they learned from my process. So, I thought I would break it down in more detail in this week's Code Meets Creed. And bonus - the process has evolved and gotten better since I first shared it, so what you'll see in this week's newsletter is the latest iteration. I'm still iterating and know there are a number of ways I can make this process even more efficient, but for now, if you want to replicate my process, check out the newsletter to get the full breakdown along with prompt templates and instructions for building your own product launch skill in Claude. DISCLAIMER: This is not about using AI as a replacement for product marketing. What great product marketers do - the positioning judgment, the customer empathy, the narrative instinct - is genuinely hard and genuinely irreplaceable. This is about automating the execution layer so that human judgment can be spent on the work that actually requires it. ALSO DISCLAIMER: I don't pretend to be a leading expert in AI or in product marketing. This is a story about how I'm solving a problem I'm facing weekly on the job. I'm sure there are marketers who've built more sophisticated systems using AI, and product marketers who are better at the job than I am. Please read the newsletter with this caveat in mind. Read this week's issue here: https://lnkd.in/ewKi-38m #codemeetscreed #AI #productmarketing #kathleenhq