Startup Folsom Vibe Coding Panel Discussion
Rachid Abadli didn’t start building software because he was looking for a cool AI project.
He owned a couple of multifamily properties and was paying a property manager to handle them. Once he dug into the numbers, he realized the real cost of management was much higher than he expected. So he started asking a very founder-like question:
What would it take to solve this myself?
Rachid already had a software background, but AI dramatically shortened the path from problem to product. He built a property management platform initially for himself, then realized other landlords had the same problem and turned it into a SaaS business.
That story captured much of what we discussed at our recent Startup Folsom AI Meetup.
On August 18, Startup Folsom brought together three people approaching AI-assisted software development from very different directions: Orion Burdick, a longtime software and product developer; Justin Mendez, Head of AI at GreyDot Marketing; and Rachid Abadli, an AI-native engineering leader and founder of LeaseBase.
The topic was vibe coding.
But the conversation quickly became much bigger than coding.
It became a discussion about what happens to startups when building software is no longer the expensive part.
Here are some of my biggest takeaways.
1. Vibe Coding Is a Spectrum
The simplest definition of vibe coding is telling AI what you want and allowing it to write the code.
At one extreme, you might know almost nothing about programming. You describe an application, keep prompting the AI until it works, and never look under the hood.
At the other end is what the panel described more like agentic engineering: an experienced developer uses AI to perform much of the coding but still directs the architecture, reviews the output, tests the product and makes the important decisions.
Orion has moved significantly in that direction.
After decades of writing software, he no longer feels the need to personally inspect every line the AI generates. If the tests pass, the application works and the result looks right, he is increasingly comfortable allowing the AI to handle the implementation.
That distinction matters.
I think too many discussions about vibe coding assume you are either a professional developer writing everything manually or somebody typing “build me Airbnb” into a prompt box and hoping for the best.
There is a huge amount of territory between those two extremes.
And that middle ground is probably where most serious founders will operate.
2. The Productivity Gain Can Be Enormous
The most interesting part of the panel wasn't that AI could write code. We've known that for a while.
It was hearing what happens when someone incorporates it deeply into their workflow.
Orion estimated that he is producing roughly five times as much software as he did previously.
Instead of spending all his time writing code, he increasingly spends his time directing AI, designing, testing and deciding what should be built next. He described the experience as almost having ten engineers standing in his garage waiting for something to do.
That changes your thinking.
Things that used to sit permanently on the “nice to have someday” list suddenly become reasonable.
More automated tests?
Do it.
Translate the application into Spanish?
Why not?
Clean up an old AWS configuration you've been avoiding for two years?
Give an agent the job.
The interesting productivity gain isn't just doing your existing work faster. It is being able to do work you previously would never have justified doing at all.
3. The Best Founder Advantage May Be Knowing the Problem
For years, aspiring founders would come to startup events and say something along the lines of:
“I have a great software idea, but I need to find a technical co-founder.”
Sometimes what they really meant was, “I need somebody to spend the next year building my idea for free.”
That constraint is starting to disappear.
Rachid's experience is a good example. He understood his property-management problem because he personally experienced it. AI helped reduce the cost and time required to turn that knowledge into software.
One of the strongest conclusions from the panel was this:
Don't start by asking, “What app should I build?”
Start by asking, “What problem do I understand unusually well?”
The panelists repeatedly came back to the idea that the person closest to a problem has an advantage because they can define what a good solution actually looks like.
That may become increasingly important.
AI knows a lot about software.
It doesn't know what drives your customers crazy every Tuesday afternoon.
You might.
That domain knowledge becomes valuable.
4. AI Can Do the Middle. Humans Still Need to Own the Ends.
One of my favorite frameworks from the discussion was the “5/90/5 sandwich.”
The idea is roughly:
Humans own the first 5%: defining the problem, requirements and what success looks like.
AI can perform much of the messy 90% in the middle.
Humans own the final 5%: testing, reviewing, judging and deciding whether the result is actually good.
The panel emphasized that the middle only works well when the human has clearly defined what “correct” means and built testing into the process.
I think that's a much healthier way to think about AI than either “AI is replacing programmers” or “AI is just autocomplete.”
AI is leverage.
A forklift can help one person move a lot more boxes. You still need somebody who knows which boxes belong on which truck.
5. Faster Development Can Also Mean Faster Mistakes
There is a dangerous side to all this speed.
Rachid shared an example of getting a new paying customer who immediately encountered a workflow he hadn't properly tested. The customer hit a dead end and was gone.
AI had helped him build faster.
It hadn't removed his responsibility to test what he built.
That lesson came up repeatedly during the panel.
When an application grows beyond a simple prototype, decisions about architecture, scale, security, databases, APIs and deployment still matter.
Rachid described AI as writing much of his code while he retained responsibility for the architecture and decisions. He knew why his system was structured the way it was, how it was deployed and how its services fit together.
That's a critical distinction for founders.
You can vibe code a prototype surprisingly quickly.
A product with paying customers is different.
6. Security Cannot Be an Afterthought
The lower the barrier to creating software becomes, the more people will build software without necessarily understanding its vulnerabilities.
That is already creating problems.
During the discussion, Justin talked about malicious packages designed with names similar to legitimate software dependencies. An AI coding tool can potentially select the wrong package, inadvertently bringing malicious code into an application. He also warned about developers accidentally publishing passwords, API keys and other secrets while asking AI tools for help.
This is probably one of the least sexy parts of vibe coding.
It is also one of the most important.
If you're experimenting on your laptop, mistakes might be annoying.
If customers are trusting you with their information, those mistakes become a business problem.
Speed doesn't eliminate responsibility.
7. Ideas Are Becoming Cheaper. Distribution Is Becoming More Valuable.
This may be the biggest startup implication of vibe coding.
If almost anyone can build an MVP, having an idea for an app becomes less impressive.
A competitor may be able to reproduce many of your features surprisingly quickly.
The panel made the point that the value isn't simply the code. It's whether people actually use the product, whether it works well, whether you've built partnerships and whether you can get it in front of customers.
We've seen this before.
Orion compared AI coding to what happened to music production.
Decades ago, recording an album could require an expensive studio, engineers and specialized equipment. Today somebody can record professional-quality music on a laptop.
That dramatically democratized music production.
It did not turn everybody with a laptop into a great songwriter.
Software may be heading in the same direction.
Building becomes easier.
Creating something people want remains difficult.
And getting people to discover, trust and pay for it may become even harder as the number of products explodes.
Founders who understand customer acquisition, sales, positioning and community may actually become more valuable in an AI-native world, not less.
8. The One-Person Startup Is Becoming Much More Realistic
There was a time when launching a software startup meant assembling a team before you could build much of anything.
Today, one person can increasingly perform pieces of product development, design, research, marketing, customer support and operations with help from AI.
Rachid described how difficult that would have been with his earlier startup attempts. Today, the combination of AI and lower-cost infrastructure makes the one-person or two-person startup far more viable.
That creates an interesting opportunity for smaller companies.
Large organizations may have more money and people, but they also have legacy systems, approval processes, technical debt and layers of management.
A two-person startup can change direction Tuesday morning and ship Tuesday afternoon.
A 50,000-person corporation sometimes needs a committee to schedule the meeting where somebody will eventually discuss changing direction.
Small has always had a speed advantage.
AI potentially multiplies it.
9. Your Scarce Resource May Become Attention
There is another problem that comes with nearly unlimited development capacity.
You can build almost anything.
Which means you will be tempted to build everything.
That can be dangerous for founders.
When creating another feature only costs another prompt, it's very easy to have five agents working on seven ideas while you're halfway through three products.
The constraint moves from development resources to founder attention.
The question is no longer:
“Can we build this?”
Increasingly, the question becomes:
“Should we build this?”
That's a much harder question.
10. Don't Drink the AI Kool-Aid. Don't Throw It Away Either.
The panel ended with advice I think is particularly useful right now.
Don't believe either extreme.
There are people who try AI once, get a bad answer and decide the entire technology is useless.
There are others who assume AI will soon build entire companies autonomously while we all sit by the pool drinking margaritas.
Both positions make good social media content.
Neither is particularly useful for running a business.
The better approach is experimentation.
Understand where AI fails. Expect mistakes. Test the output. Keep humans responsible for important decisions.
But also recognize that these tools are improving quickly enough that ignoring them creates its own risk. The panel's closing advice was to understand the business problem, build around something meaningful, and avoid both blind faith and automatic dismissal of AI.
What This Means for Founders
My biggest takeaway from the discussion is that vibe coding isn't really about coding.
It's about what happens when the cost of turning an idea into software drops dramatically.
For founders, I think that changes the order of operations.
Instead of spending six months raising money to build a product nobody has tested, you may be able to build something useful, put it in front of customers and learn what they think in a fraction of the time.
But the fundamentals haven't disappeared.
You still need to understand the customer.
You still need to solve a real problem.
You still need to decide what gets built.
You still need to test it.
You still need to sell it.
And somebody still has to convince the first customer to pull out a credit card.
AI can write a lot of code.
It can't make your customers care.
At least not yet.
About Orion Burdick
Orion Burdick is a technology and product development leader with a background in software UI/UX design, engineering, and digital product development. He brings hands-on experience across mobile and web technologies, with a strong focus on building user-centered products that are practical, intuitive, and effective.
About Rachid Abadli
Rachid Abadli is an AI-native engineering leader and founder of LeaseBase. With more than 10 years of experience building and modernizing production systems across SaaS, payments, cybersecurity, and enterprise platforms, he focuses on using AI, cloud architecture, DevSecOps, and platform engineering to help teams build scalable, secure, and reliable software systems.
About Justin Mendez
Justin Mendez is Head of AI at GreyDot Marketing, where he helps business owners deploy AI in practical, revenue-focused ways. An Afghanistan veteran, father, and husband, Justin brings a service-minded approach to helping companies use technology to simplify operations, improve marketing, and move faster.
