Software stocks have taken a beating lately. Every few weeks another headline warns that artificial intelligence is about to eat the lunch of companies we once thought were untouchable. The anxiety is palpable — and honestly, after what I’ve seen recently, I understand why investors are nervous.
But is the fear overblown? Or are we watching the early chapters of a genuine restructuring of the entire software industry? To find out, I decided to stop reading speculation and actually test the threat myself. The result left me more thoughtful — and more cautious — than when I started.
The Vibe-Coding Experiment That Changed My Perspective
Let’s be clear right away: neither of us involved in this little project are developers. We don’t write production code for a living. We barely know JavaScript from TypeScript. Yet armed only with plain English and one of the leading AI coding assistants available today, we set out to recreate — in under sixty minutes — a functional version of a popular project management platform valued at several billion dollars.
We didn’t expect much. At best, maybe a toy demo. What actually happened exceeded even the most optimistic predictions.
The first prompt was deliberately basic: build a dashboard that looks and feels similar to a well-known work management tool — multiple boards, task assignments, status dropdowns, the usual suspects. Minutes later a working prototype appeared. Not beautiful, but undeniably functional.
Encouraged, we pushed further. We asked the AI to study the real product, identify its core features, and implement as many as possible. Calendar views appeared. Dependency tracking started working. Notifications fired. Each refinement took seconds rather than days.
When the AI Got Personal
The real “aha” moment came when we connected a personal email account. Suddenly the system wasn’t just a generic clone — it became strangely proactive. It scanned recent messages and created tasks automatically: reminders to buy a birthday gift, book flight tickets, sign a waiver for a school event. All without being explicitly told to do so. It simply inferred intent from context.
Cost? Somewhere between five and fifteen dollars in compute credits. Time? Less than an hour. Skill required? Basically none beyond the ability to write clear sentences.
The barrier to creating reasonably sophisticated software has collapsed faster than most people expected.
— Observation from someone who just watched it happen
That experience forced a hard look at which parts of the software ecosystem might actually be vulnerable.
Who’s Most Exposed? The “On Top of Work” Layer
Conversations with people deep inside the technology world point to a clear pattern. The applications that feel most fragile right now are the ones that “sit on top of the work” rather than being foundational to it.
These tend to be tools that organize, visualize, communicate or decorate the real work happening elsewhere. They are valuable today — sometimes extremely valuable — but their value often depends on user habits, integrations, and convenience rather than irreplaceable data moats or network effects.
- Project and task management platforms
- Marketing automation and CRM-lite solutions
- Design and creative collaboration tools
- Customer support ticketing systems
- Spreadsheet-like database interfaces
Companies in these categories have built impressive businesses by making knowledge work smoother. But when an AI agent can read your email, understand your projects, and recreate 80% of the interface in an afternoon, the convenience premium starts to look shaky.
The Relative Safety of Systems of Record
Not every software company faces the same level of danger. There’s an important distinction between tools that sit on top of the work and those that are the work — or at least house the canonical version of critical business data.
These “systems of record” accumulate years — sometimes decades — of irreplaceable company information. Replacing them isn’t just a matter of rebuilding a nicer interface. You’d also have to migrate massive, sensitive, highly customized datasets while maintaining perfect auditability and compliance. That’s a very different proposition from spinning up a new task board.
Even here though, complacency would be dangerous. AI agents are already showing the ability to interact with APIs, extract structured data, and write back changes. Over time those capabilities will only improve. The moat still exists — but it’s narrower than it used to be.
Areas That Still Feel Defensible (At Least for Now)
Some segments of the software world seem structurally harder to disrupt with current-generation AI tools.
- Cybersecurity platforms — especially those that rely on proprietary telemetry, network effects, and constant adversarial adaptation. Few people want an AI-rewritten version of their endpoint protection.
- Infrastructure and developer tools that sit very low in the stack and require deep performance optimization or hardware integration.
- Vertical-specific applications with heavy regulatory requirements, complex pricing logic, or intimate knowledge of niche industry workflows.
- Platforms whose primary value is the network itself rather than the interface (think certain collaboration suites with millions of daily active users).
Even in these areas, though, the pressure is building. AI is not standing still.
What This Means for Investors Right Now
The broad sell-off in software names has created real dislocations. Valuations that looked stretched a year ago now appear far more reasonable — sometimes even attractive — for the right businesses.
But broad strokes won’t cut it anymore. The market is starting to differentiate sharply between companies that could see meaningful revenue erosion in the next 24–36 months and those that are likely to remain essential even in an AI-saturated world.
I’ve come to believe the key question isn’t “Can AI replace this tool?” but rather:
How much of this company’s value depends on interface convenience rather than irreplaceable data or structural advantages?
Businesses where the answer is “a lot” face a tougher road ahead. Businesses where the answer is “not as much as you think” may actually benefit from the productivity tailwind AI is creating across the economy.
The Bigger Picture: A Once-in-a-Decade Re-Rating?
We may be witnessing the beginning of a major re-rating of software multiples. For roughly fifteen years, investors paid very high prices for growth, scalability, and recurring revenue — even when competitive moats were questionable.
AI agents are shining a very bright light on those moats. Some will hold up beautifully. Others will prove surprisingly thin once you can replicate core functionality with a few sentences and a small cloud bill.
The companies that survive — and thrive — will likely share a few characteristics:
- Deep ownership of mission-critical data
- Strong network effects or switching costs
- Ability to leverage AI internally to widen their lead rather than watching it erode their advantage
- Management teams that understand the moment and act decisively
The rest? They’ll need to adapt very quickly or risk becoming interesting acquisition targets at best — and irrelevant at worst.
My Take: Caution, Not Panic
After running the experiment myself, I’m more convinced than ever that AI will reshape software — but I’m not ready to declare the entire sector doomed.
Some businesses will indeed face existential questions in the next few years. Others will quietly become more valuable because AI makes their customers more productive and therefore more willing to pay for tools that coordinate that productivity.
The real winners will be the companies that embrace the change rather than fight it. And for investors willing to separate the need-to-haves from the nice-to-haves, this period of fear could eventually look like one of the more attractive entry points in recent memory.
One thing is certain: the old rules no longer apply exactly as they did. The next few years will reward clear thinking and punish complacency.
And that, perhaps, is the most valuable takeaway of all.
