In September 2026, LocalStack announced the acquisition of WonderTwin AI, a provider of local emulators for applications such as GitHub, Stripe, PostHog, Hubspot and others. This article was written by Tela Andrews, founder of WonderTwin AI, to introduce the vision for application emulation for the LocalStack community.
When you watch footage of a living cell under a microscope, you see something that never holds still: Constant motion, signals firing, structures forming and dissolving, everything reacting to everything else.
Contrast that with how most of the industry talks about commercial software–as a series of finished things. Version 4.2. The Q3 release. A product, shipped, sitting there until the next one replaces it.
Both pictures are true. But only one of them is the right size for the actual market. If you speed up the movie of the commercial software market, it looks just like living cells. AI agents are creating that acceleration, and demand a new paradigm.
Individually, most software products change at a pace that looks perfectly reasonable–a release here, a deprecation there, a small processing logic change, nothing dramatic. Add up every product, every API, every service that every company depends on, and the picture stops looking like a release train and starts looking like that cell. The software market, taken as a whole, is constantly moving. Not occasionally. Continuously. That aggregate motion, not any single company’s roadmap, is what actually sets the size of the software market, because it’s the real surface area of everything that could be built, tested, and depended on.
Engineering teams have never fully engaged with that surface area, for good reason. Every new dependency is a change-management liability, so most organizations deliberately limit how many they take on and how tightly they couple to them. That’s a rational constraint. It’s also a self-imposed ceiling: a decision to leave most of the moving, living market alone rather than risk it.
The Shift to Agentic Development
Agentic development is breaking that constraint, whether engineering teams are ready for it or not. Agents don’t wait for a quarterly review of new dependencies; they reach for whatever gets the job done, continuously, at a volume no change-management process was built to review.
While the dynamic view of software is an opportunity to engage with far more of the software ecosystem, it’s also the source of the risk. An agent accelerating past the constraints of the traditional model doesn’t necessarily accelerate velocity in a safe way. Speed without grounding just means you find out how a dependency actually behaves (its rate limits, its edge cases, its failure modes) in production, after the agent already shipped against a guess.
Knowing the shape of an interface was never the hard part. Knowing how it behaves is. I learned that the hard way while building integration programs and commercial API products long before agents made the problem urgent.
That’s the gap that application emulation closes. Providing a real, stateful, continuously calibrated model of how a dependency behaves ensures the automated workflow is grounded in reality instead of a guess. Ground it, and the traditional ceiling for change-management stops being necessary. Growth isn’t achieved incrementally, but by removing a constraint on engaging with the full, dynamic scope of what software already is.
Here’s the part I didn’t see fully until recently. Grounding acceleration only works if it covers the whole dependency graph, not just parts of it. An agent can be perfectly grounded in the behavior of every third-party application it calls and still run headlong into ungrounded cloud infrastructure, and vice versa. Partial grounding still leaves a gap wide enough to fall into.
Building that completeness alone, from application emulation outward, would have taken years I don’t have. LocalStack has already built the infrastructure half: Cloud emulation trusted by more than 1,500 organizations, pulled from Docker hundreds of millions of times, at a scale nobody else in this space is close to. Joining LocalStack is the strongest route to full coverage of the software development ecosystem. It turns two partial answers into a complete solution, on a foundation already proven at the scale this vision needs.
Joining LocalStack to Complete the Vision
So today, WonderTwin AI is joining LocalStack. What we’re building toward (grounding every layer an agent touches, application and infrastructure alike, in how it actually behaves) is something I’ll be writing a lot more about soon, under a name I’ve started calling agentic full-stack emulation. This post is the setup. The next one is the argument in full.
Today, WonderTwin AI supports local emulation for more than two dozen applications that AI-native software development teams build against frequently, with others still in development. We’re working behind the scenes to deliver a comprehensive experience that integrates local application emulation with local cloud development, and will share more information soon.
I’m proud of what I’ve built at WonderTwin AI, and excited to continue that journey as part of the LocalStack Team, leading our Application Emulator efforts. The details on today’s announcement are linked below. What comes next is bigger than a product update.
