Value Creation When AI Makes Prototypes Cheap
Generative AI has dramatically lowered the cost of creating software prototypes, but it has not eliminated the work required to create durable startup value. The value is shifting away from simply producing working screens and toward building reliable, secure, domain-specific systems that solve real customer problems, fit existing workflows, and can be sold, supported, and improved over time.
1. The Question: Has Software Creation Lost Its Value?
I have been thinking a lot about value creation and valuation in a world where revenue-generating software can be built with far less traditional engineering investment. A few years ago, a working prototype was often enough to clear the bar for an angel investment conversation. Today, a non-technical entrepreneur can create a prototype, MVP, or even a seemingly functional product in days. That makes it tempting to conclude that the value of software creation has fallen close to zero.
My experience building an internal operating system for Harkavest suggests a more nuanced answer: the prototype has become easier, but the durable product is still hard. The moat is not the first demo. The moat is the ability to turn a demo into a dependable workflow that customers trust and becomes indispensable to their business. The proof of the moat is not the product features, but the adoption rate of your customers.
2. The Experiment: Building the Harkavest Operating System
To test the theory, I decided to build my own operating system for Harkavest, the investment company sister organization to my advisory and coaching company, Harkador Partners. The application sounded simple on the surface: account for operating expenses, organize investment documents, issue occasional LP capital calls, ensure vendors get paid, and track payments to portfolio companies. Easy, right?
The exercise became a practical way to see the good, the bad, and the ugly of generative AI-enabled software development. It also became a useful case study in where AI accelerates product creation and where traditional product discipline, architecture, testing, and domain expertise still matter.
3. Stage One: Fast Prototype, Fragile Foundation
I started with Google AI Studio and had something that looked like it might work within a few days. It was good at ingesting PDFs, understanding the basic document flow, and producing a decent early user interface. As a prototype, it was impressive.
But the limits showed up quickly. The app frequently lost data that was already in the system, and improvements in one part of the software sometimes removed functionality elsewhere. The first architecture relied too heavily on the browser. After talking with my (trained computer sciencetist) daughter, I realized I needed a persistent database-as-a-service and moved to Firestore/Firebase. That improved the foundation, but the development process still produced inconsistent results.
4. Stage Two: More Durable Build, New Production Problems
I then used the Google AI Studio work as a prototype and rebuilt a more durable version with Cursor, while keeping Firestore/Firebase as the database layer. This approach produced more consistent incremental progress and more usable working software. But it also revealed a different category of problems: the user interface was clunky, PDFs were sometimes misinterpreted, signs were flipped in calculations, and data was misclassified.
That distinction is important. AI made it easier to generate working functionality, but it did not eliminate the need for systematic testing, data validation, edge-case handling, and architectural judgment. A product can look useful long before it is dependable.
5. Stage Three: Domain Expertise Became the Real Bottleneck
After several months of part-time work, I finally spent time reconciling bank statements and accounting for the investments Harkavest had made over the prior year. That is when the real product gap became clear. Working with an accounting domain expert, my wife, I realized that the app needed an integrated double-entry accounting module, full general ledger, P&L, and balance sheet.
The modules were easy to add in the narrow coding sense; they took roughly an hour to generate. But they took several days to debug and test. More importantly, I would not have been able to get the accounting right without multiple working sessions with someone who deeply understood the domain.
That experience clarified the central lesson: in B2B applications, value is not created by software that merely appears to work. Value is created by software that correctly captures the details of a domain, handles real-world exceptions, and earns the trust of users who depend on it.
6. The Result: Working Software, But Not a Finished Solution
After one month in Google AI Studio and four paid months of Cursor, I now have stable working software that I can use to run Harkavest operations. That is a remarkable outcome for a side project. It also proves the larger point: generative AI can dramatically accelerate the path from idea to usable internal tool.
At the same time, I expect I will eventually migrate the accounting functions to a more established application such as Zoho or QuickBooks. The AI-built accounting layer helped me learn and operate, but a more mature accounting framework will likely be the better long-term solution for that part of the system.
7. What does this mean for funding B2B Startups
Prototyping Is Now Cheap and Democratic
It is now easy to create a prototype, and every product manager and entrepreneur should be able to prototype their vision. AI allows teams to move from explanation to demonstration much faster than before.
But a prototype should not be confused with a company. The easier it becomes to produce a polished demo/MVP, the less defensible the MVP itself becomes.
Production Software Is Still Hard
It is much harder to build a fully tested, cybersecure, domain-specific application that customers can rely on. AI can create software that looks good in the same way an Excel model can produce numbers that look reasonable. But robustness requires structured product design, testing, controls, support systems, and careful attention to failure modes.
The better you can tell the AI what you want to build, the better the result will be. A crisp product definition, informed by customer discovery, remains essential. Good planning on the front end still produces a much better product on the back end.
Domain Knowledge Is a Durable Source of Value
Startup value creation continues to sit at the intersection of truly understanding the customer’s domain and delivering a solution that works the way the customer works. The more niche the domain, the more opportunity there is for a startup to create sustainable value by deeply understanding workflows, exceptions, terminology, and trust requirements.
The actual software code may deliver less sustainable enterprise value than it used to. But a business that can continuously refine its product around real customer pain points, supported by domain expertise and operational discipline, has a much higher chance of lasting success.
Distribution and Trust Still Matter
Enterprise customers will increasingly build internal systems to glue together existing tools. That work is easier at the prototype level than it used to be, but internal bespoke systems will still face familiar challenges: inconsistent standards, weaker security, limited support, and feature sets that may lag competitive products.
For venture-backed startups, the implication is clear. Investors will be more interested in teams that can show they can sell, retain customers, and grow revenue than in teams that merely have an attractive generative AI prototype. Revenue growth, user growth, distribution, and customer trust remain powerful signals of value.
8. Conclusion: The Moat Has Moved
Generative AI has not made startup value disappear. It has changed where value accumulates. The moat has moved from the ability to create software at all to the ability to create software that is trusted, differentiated, embedded in real workflows, continuously improved, and distributed effectively.
In the AI era, a beautiful prototype is no longer enough. Sustainable value will come from domain understanding, product judgment, operational reliability, customer trust, and the ability to turn rapid software creation into a real business.