Here is the pattern we have watched for 25 years, playing out on a shorter and shorter loop: someone builds something genuinely good. It solves a real problem for real people. A community grows around it. Then a bigger company buys it, folds it into a product suite, and eventually kills it. The people who depended on it scramble to find something else, and the cycle begins again.
Petal exists because this happened to CoScreen. CoScreen was a pair programming tool people actually loved. Datadog acquired it. Then Datadog shut it down on July 31st. Four engineers from the original team spent months rebuilding something equivalent from scratch, open-source, because they knew the tool was worth saving. That kind of commitment tells you something. It tells you the original product had real value. It also tells you that real value is not enough to protect something inside a larger company's portfolio.
This is not a CoScreen problem. This is a structural problem with how tools get built and distributed. The most useful specialized software almost always gets built by a small team obsessed with a narrow problem. Then someone with money and distribution buys it, not because they love the problem, but because the acquirer wants the team, the users, or the positioning. The product itself is often secondary. And once the strategic rationale shifts, the product goes dark.
Now look at what is happening on the AI side. Agents are being pushed into every corner of knowledge work, and the framing is always capability-first: look what it can do. And it is genuinely impressive. Researchers are predicting AI as the default approach to theoretical problems across entire scientific disciplines.Specialized financial NLP systems are hitting near-perfect accuracy on narrow benchmarks. The capability curve is real. We are not here to dismiss it.
But here is what nobody is talking about loudly enough: almost every one of these AI capabilities is locked inside a product owned by someone who does not share your interests. The model is rented. The API is rented. The workflow automation is rented. The agent memory is rented. You are building your operations on infrastructure that can be repriced, restructured, deprecated, or sold on a quarterly earnings call you were not invited to.
We have seen this before. Not just with CoScreen. With Parse, which Facebook shut down and left a hundred thousand mobile apps scrambling. With Google Reader, which still has people angry about it thirteen years later. With every beloved dev tool that got acqui-hired into silence. The lesson is not "don't use third-party tools." That would be paralyzing. The lesson is: understand what you actually own versus what you are renting, and act accordingly when the stakes are high.
There is a version of your stack where the pieces you depend on most are also the pieces you have the least control over. Your CRM, your analytics, your payment routing, your AI layer, your communication infrastructure. Every one of those is a subscription to a company whose incentives are not yours. That is fine for commodity functionality. It is a serious risk for anything that is core to how you deliver value to customers.
The Petal story is actually the optimistic version. A team cared enough to rebuild the thing open-source and give it away free. That almost never happens. Most of the time, the tool just disappears and you are left rebuilding workflows under pressure, at exactly the wrong moment, with no runway to do it cleanly. The last thing you want is to discover your dependence on a tool the week it announces a sunset.
So what do you do? A few things, and none of them require becoming a paranoid infrastructure purist.
First, audit your stack once a year for single points of dependency. Not for everything, but for the tools that, if they disappeared in 90 days, would genuinely break your operations. If that list is longer than three items, you have work to do. Not necessarily now, but on the roadmap.
Second, bias toward open-source or self-hostable options for the tools closest to your core product. Not because open-source is always better, but because it cannot be sunset by an acquirer. The code does not go dark. Petal exists today precisely because the original team built something the community could carry forward without permission from a corporate parent.
Third, when you are evaluating a new tool that sits deep in your stack, ask one question before you ask anything about features: who owns this, and what happens to us if they sell or shut down? A startup with no business model and a growing user base is a beautiful acquisition target. That beauty is a risk for you, not a feature.
The AI wave is going to make this harder, not easier. The capability gains are real and the productivity arguments are often legitimate. But the consolidation pressure is enormous. A handful of companies control the foundational models. The agent layer is getting locked up by the same players. If your business depends on AI functionality three layers deep inside someone else's platform, you are not building a capability. You are renting one with no lease.
We are not saying avoid AI. We are saying: know what you own. Build the parts that are yours to keep. Rent the commodity. Own the edge. Because the tools you love have a habit of being acquired, and the ones that get acquired have a habit of disappearing, and the ones that disappear take your workflows with them.
The question is not whether it will happen to something in your stack. It is whether you will notice in time to do something about it.