Show Notes: Inside the 90™ Episode #39
This episode is part one of a two-part conversation about AI — starting with vibe coding, the practice of describing what you want in plain English and having AI write the working software.
Neither host claims to be an expert here. This is two entrepreneurs sharing what they've actually built over the last five months, and why they think the pace of change is worth paying attention to.
Three Camps, and Why This Conversation Matters Now
Most people fall into one of three groups: already excited and building, curious but unsure where to start, or simply unaware of how fast things have moved.
That third group is the real concern. Plenty of entrepreneurial leaders are genuinely detached from just how much has changed in the last several months, and that gap tends to widen quietly until it's suddenly very visible.
For anyone wanting a more structured, background-level understanding before diving in, Jeff Yelton's books, Agentic AI for Leaders and Responsible AI Governance in Practice, are worth a look. Neither host claims expertise in AI governance, but Yelton's work fills that gap well.
The pace itself is the headline. Software is being built differently between Q1 and Q4 of this same year — not gradually, but in a way that makes standing still a real risk.
From AI Studio to Vibe Coding: How the Fire Got Lit
Watching two friends in a co-working space build real, functioning tools in Google AI Studio was the spark — actually publishing one over a single weekend is what made it undeniable.
The tool in question, an internal practice-management app nicknamed Wonder Bread, replaced a tangle of spreadsheets used to track work value across client scopes. From a rough functional spec to a working prototype took about four hours on a Saturday night. By Monday morning, the whole team was using it.
A technical sales and marketing background — writing functional specs and business requirements without ever touching a keyboard as a developer — turned out to be exactly the right preparation. Understanding how software should be structured, paired with a naturally high risk tolerance, meant going from idea to working feature almost immediately.
The tools themselves matter less than the mindset. Lovable is the current tool of choice simply because it works and hasn't needed replacing, but the real skill being built is delegation — describing what's needed clearly enough for AI to build it well.
Start Small: The Power of Single-Use Apps
The best early projects aren't ambitious platforms — they're small, single-purpose tools that plug an obvious gap.
A stress assessment tool for a coaching business, built in about two hours after a first over-engineered attempt was scrapped and rebuilt with a much simpler prompt.
A core values grader that scores a company's stated values against Lencioni's "permission to play" filter and displays the strongest results on a public wall of fame.
A personal to-do app built in about an hour because no existing task tool matched how its builder's brain actually organizes work.
A rapid wireframing tool that replaced physical note cards for planning website structure and copy in about an hour instead of days.
There's a useful way to frame these projects: "I want to do blank, but blank." I want to wireframe faster, but I don't want to buy another SaaS subscription. Filling in those two blanks tends to point straight at a good first project.
None of this replaces the actual value a team provides. It replaces the friction around it — the spreadsheet gymnastics, the underused software paid for in full and used at twenty percent, the manual grunt work standing between an idea and a result.
When It Gets Bigger: Multi-Tenant Apps and Real Guardrails
Once a single-use tool proves itself, the natural next step is opening it up to more than one person — and that's where a few real guardrails matter.
An internal practice-management tool built for one person's own use became multi-tenant simply by asking the AI to make it so — separate logins, separate data sets, one shared codebase. A recruiting CRM built the same way now runs a whole hiring pipeline, complete with assessment scores and interview rubrics.
Clients are starting to take this further than expected. One manufacturing client built a quoting tool that reads uploaded CAD drawings and uses an LLM to estimate cost per piece and turnaround, with a human able to correct and retrain it over time.
The guardrails that matter here are simple but real: keep mission-critical work and client data out of early experiments, watch for active agents or scheduled jobs that quietly burn credits around the clock, and know when a project's stakes call for bringing in an actual developer instead of pushing further solo.
Final Takeaway: Look at Your Spreadsheets, Find Your Community
The best place to start is wherever a spreadsheet is currently duct-taped to a process it was never built for — those gray areas are usually the highest-value first projects available.
Finding a community that's already experimenting, even something as informal as a recurring dinner where the only rule is no talking about anything except what's being built, makes the learning curve far less lonely.
This isn't about replacing the work that makes a business valuable — it's about removing the friction standing in front of it.
Recent Posts









