Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

When everyone can build software, judgment matters more

A framework for creating solutions that last

[Photo: Getty Images]

Profile photo of Geoff Smalling

Chief Technology Officer · Shippo

Within a single month, our team went from zero vibe-coded apps to dozens. No engineering background required. Marketing built a custom GPT trained on company voice and past content. Sales built customer intelligence tools pulling from our own and external sources. Operations automated workflows that had been manual for years.

Ramp reports similar patterns, with more than 1,500 internal apps shipped in six weeks by 800+ non-engineer builders. A 2026 survey of 200 enterprise CISOs by Kanopy Security puts a number on it: Business software builders now outnumber professional developers by 4 to 10 times across large organizations, and 98% of those CISOs expect that gap to keep widening. The barrier to building software has collapsed faster than most leaders can process.

Every business leader is now making software decisions, whether they realize it or not.

THOSE WHO UNDERSTAND CAN BUILD THE SOLUTION

For decades, there was a gap between the people who understood business problems and the people who could build software to solve them. The classic cartoon: A customer asks for a simple wooden swing and the engineering team delivers a slide. Software was what happened after you handed your problem to someone else and waited.

That wall is gone. The marketing manager who knows their needs can build it directly. The operations lead drowning in manual processes can automate them without filing a ticket. Frontline teams that would never win a prioritization argument against a product feature can now just build what they want.

This is mostly good. But when anyone can build, the bottleneck moves from creation to judgment.

ACCESS WITH NO FOUNDATION CREATES DEBT AND RISK

We learned this ourselves. As I wrote previously, our company accumulated 19 separate GenAI tools, each procured without knowing about the others. Our vibe-coded app ecosystem started the same way: duplication, abandoned projects, and security exposure nobody had mapped.

The research reflects what we saw. In a May 2026 account of scaling vibe-coded apps built by their marketing team, Thoughtworks cited industry data showing 25% of AI-generated code with confirmed vulnerabilities, a 44% year-on-year rise in attacks exploiting application vulnerabilities, and 1 in 5 enterprise breaches now linked to AI-generated code. Gartner, in a January 2026 report on agentic AI security, was equally direct: “A rapid influx of AI tools and agents adopted by employees and developers without central oversight has led to shadow attack surfaces.”

Speed is real. So is the exposure when basic questions haven’t been answered before the build starts.

BUILDING QUICKLY STILL REQUIRES A CONTROLLED FOUNDATION

Seeing the problem is only half the battle. As our ecosystem scaled, we implemented a controlled foundation: production-separated app zones, functional categorization to prevent duplication, role-based access, and a robust hosting environment for AI apps and AI agents treated as critical infrastructure.

That last piece matters more than most leaders expect. The hosting environment is where security, access, model versioning, and monitoring come together. Organizations treating it as an edge case will spend the next two years cleaning up what they built in the first six months.

EVERY LEADER SHOULD THINK LIKE A PRODUCT MANAGER

Good infrastructure handles the plumbing, not the judgment problem.

Once we understood the pattern, we ran a half-day product bootcamp and contest with our extended leadership team. We just had business leaders building, paired with product and engineering leaders as facilitators. A few hours, one framework.

We first brainstormed internal problems worth solving. Then we prioritized with RICE: reach (how many people does this affect?), impact (how much does it change their work?), confidence (how well do we understand the problem?), and effort (how long will this take to maintain, not just build?). Most people most want to skip the effort question, but it matters the most, as AI now makes creation feel nearly free.

The winning team automated shipment claims, a customer workflow with broad reach and clear financial impact. They created a documented process for each carrier and tested it against real transaction data using LLM-based evaluations. It wasn’t the flashiest application, but it solved a real problem, and reflected the product thinking we were trying to teach.

Teams also discovered that not every problem requires a vibe-coded app. Often, a shared prompt or a simple data interface suffices. Only a few scenarios justify the overhead of a full build and its maintenance. Our framework provided the clarity to distinguish between these needs before building began.

THE HARDEST QUESTIONS WERE NOT TECHNICAL

The leaders at that bootcamp walked out with five questions no coding tutorial would have taught them: Should we build this at all, or is there an existing solution? Who is this for, and how do we know their problem is what we think? How will we test it as models change over time? Who owns it after launch, including security patches and eventual deprecation? And what does it cost to maintain compared to what it solves?

Those are product questions, and now management questions too.

McKinsey’s CEO said in early 2026 that his firm runs 25,000 AI agents alongside its human workforce and expects parity by year-end. Microsoft’s 2026 Work Trend Index, surveying 20,000 knowledge workers, found that organizational factors like culture and management support drive more than twice the impact on AI adoption compared to what individual employees do on their own.

The companies that get the most from all of this won’t be the ones who handed employees the ability to build AI-powered software and hoped for the best. They’ll be the ones that paired that capability with governance and judgment. The coding barrier is down. The gap that remains is not technical.

Geoff Smalling is CTO of Shippo.