Is AI ‘vibe coding’ too good to be true?
AI means companies can now ‘vibe code’ their own software products on the cheap – but they may soon discover the risks, writes Paul Armstrong
Corporate leaders are about to discover software has become dangerously easy to build. Artificial intelligence tools now allow any employee to describe an application in plain English and generate working code within minutes. Silicon Valley is calling this vibe coding and treating it as the next gold rush. Most boardrooms have not yet priced in what comes after.
Investment money is already moving fast. Markets tied to AI-assisted development are projected to approach £365bn by 2040 according to Roots Analysis, as venture capital backs companies building these tools. Products such as Lovable, Cursor and Replit, alongside AI coding environments inside Claude and ChatGPT, promise to squash development cycles that once required months into hours. A working prototype can now open more VC wallets than a polished deck. The shift this creates is less about faster programming and more about how the economics of software itself is about to change for businesses around the world.
The numbers behind these tools tell their own story. Cursor is reportedly in talks to raise at a valuation of up to $60bn, more than double its November 2025 mark. Lovable added $100m in annualised revenue in a single month with just 146 employees. Replit tripled its valuation in six months, raising at $9bn in March after closing at $3bn in September. By any measure, capital in this category is, rightly or wrongly, moving with extraordinary conviction.
Cheap software changes the economics
For decades the ability to write complex code created defendable businesses with solid moats. Companies invested heavily in engineering teams because building reliable software required specialised expertise and long development cycles. AI is smashing all these elements and forcing a deeper shift in where value and, to some degree, power, now sits across the technology industry.
When applications can be generated with a decent prompt, code itself stops being the moat that protects your product and any durable advantage begins to diminish or move. The things that do become more valuable are things like proprietary data, distribution channels and integration with existing systems. Software features themselves become easier for competitors to reproduce and that should terrify any exec, even with good lawyers.
Investors are beginning to confront an uncomfortable reality at least in the short term. AI coding tools seem to be extremely valuable businesses, but companies built entirely with those tools may struggle to defend their economics once competitors can generate similar products within a matter of hours. Private markets analyst Paul Smalera sees less obvious moats with vibe coding, and believes the companies that look like platforms today may look like features tomorrow. Once competitors can generate similar products within hours, feature parity stops being a competitive advantage and becomes a baseline expectation.
Inside companies a double-edged impact may arrive even faster. Employees now have the ability to generate working applications without waiting for central IT teams. Marketing departments can build customer dashboards. Operations teams can automate reporting tools. Product managers can prototype entirely new services without assembling a development team first. The software industry has long assumed that writing code represents the hardest part of building technology businesses, yet AI-assisted development suggests the harder challenge may soon be controlling the systems created once code becomes cheap.
Another new wave of shadow AI
Speed introduces a new category of risk. Much of the code AI produces is generated by prediction rather than written deliberately line by line. Developers often receive working outputs without fully understanding how the system produced them or how they behave in unusual situations. Early prototypes frequently perform well while systems remain small, but problems tend to appear later once those applications become embedded in real workflows. Perhaps this is enough to get proper money, but there’s a ton of risk involved and due diligence to be done.
Reliability, edge cases and security issues often require extensive human checking. So much so that there’s already a cottage industry popping up ready to deal with that gap. Services such as Humans Fix AI specialise in reviewing, repairing and stabilising AI-generated systems that businesses rushed into production.
Governance therefore becomes the first challenge leadership teams will face. When employees begin generating internal software independently, organisations lose visibility over what code actually exists inside the business. Security vulnerabilities, data exposure and compliance risks multiply quickly when applications appear outside established review processes. Technology leaders already recognise the pattern from earlier waves of “shadow IT” created by spreadsheets, automation tools and low-code platforms, although AI-assisted coding dramatically accelerates the speed at which these unofficial systems can appear.
Control becomes the real challenge
Cheap code also threatens the economic assumptions behind many software companies. Traditional SaaS businesses built stable advantages on the idea that software features were expensive and difficult to replicate. Artificial intelligence changes that equation by allowing competitors to generate similar functionality dramatically faster, making feature parity easier to achieve and long-term differentiation harder to sustain.
As a result, value begins shifting away from the code itself and toward the surrounding ecosystem. Control of proprietary data, ownership of customer relationships and distribution channels become more important than the underlying application features. Investors are already beginning to examine whether software companies built purely on application functionality retain defensible advantages once AI dramatically reduces the cost of building competing products.
None of this suggests businesses should avoid AI-assisted development. Ignoring the technology would be a mistake because organisations capable of experimenting quickly will discover opportunities that slower competitors miss. Leadership discipline therefore becomes the real differentiator. A sensible approach is to treat vibe coding primarily as a prototyping engine rather than a production environment.
AI-generated software works extremely well for internal tools, automation experiments and early product concepts, while systems that underpin core operations should still pass through experienced engineers capable of reviewing architecture, security and long-term maintainability before they reach customers. Software may be becoming dramatically cheaper and faster to create, yet the responsibility for governing the systems produced by that software is moving in the opposite direction.
Organisations that recognise this early will experiment more freely while maintaining control of the technology running their business. Companies that assume AI-generated code behaves like traditional software may discover that rapid creation without disciplined oversight produces operational risks that take far longer to fix than the systems took to build.
Paul Armstrong is founder of emerging tech advisory, TBD Group and its intelligence communityTBD+