Paris-based SkipLabs has officially launched Skipper, a novel AI coding agent that promises to redefine the landscape of software development by shifting the industry’s primary focus from mere speed to inherent correctness and architectural integrity. The company’s founder and CEO, Julien Verlaguet, a prominent figure known for creating Hack, the gradually typed programming language that underpins Facebook’s extensive business logic across millions of lines of production code, argues that the current generation of AI coding tools has been addressing the wrong problem. Verlaguet posits that the core challenge in building robust software has always been an architectural one, masquerading as a coding issue, and that AI, rather than solving this, has only amplified its urgency.
Skipper, described by SkipLabs as the critical "substrate sitting between foundation models and shipped software," is designed to operate within a "closed-loop" system. This approach aims to deliver complete, validated backend services from plain-language descriptions or OpenAPI specifications without the need for extensive developer review cycles or iterative prompting. The agent is engineered to handle the entire development process internally, from initial prompt interpretation to the generation and validation of code, all before presenting a finished product to the user.
"Building correct software has always been an architecture problem disguised as a coding problem," Verlaguet stated in a release. "AI did not change that; it just made the problem more urgent." This philosophy directly challenges the prevailing narrative in AI coding tools, which often tout increased developer velocity as their primary benefit.
The "Closed-Loop" Paradigm: Beyond Iteration
The differentiating factor for Skipper lies in its "closed-loop" mechanism. Unlike tools such as Claude Code, Cursor, or Codex, which necessitate continuous developer intervention for review and refinement, Skipper internalizes this process. "Closed loop is not a feature," Verlaguet elaborated. "It is a different theory of what an AI coding tool is supposed to do. The current generation makes the developer faster. The next generation makes the developer’s involvement optional. Describe what you want, and Skipper builds it."
In practical terms, Skipper takes a user’s prompt, generates an OpenAPI specification, and then proceeds to construct the entire backend service. This includes defining routes, data mappers, validators, TypeScript types, and crucially, comprehensive unit tests. The generated service is then run within a Docker container. If the code fails type checks, Skipper is programmed to attempt corrections autonomously, with up to eight retries before deeming the process complete or signaling an issue. This entire validation and correction phase occurs without direct human oversight, aiming for a seamless transition from concept to a functional service. The analogy provided by SkipLabs is akin to defining an input, running Skipper, and receiving a working service, freeing developers from the minutiae of generated code.
Architectural Bet: Tackling State and Concurrency
At the heart of Skipper’s design is an architectural bet on the inherent difficulties AI-generated code faces with state management and concurrency. Verlaguet contends that these are the areas where AI models most frequently falter, and that the solution requires a structural, rather than a prompting-based, approach.
This stance is not new for Verlaguet. In prior discussions, he expressed skepticism about existing "guardrail" solutions for AI, observing that they often amounted to more sophisticated prompting techniques rather than fundamental architectural innovations. "Every time I look closer at people who claim that they are bringing guardrails to AI, I see more the same," he remarked previously. "I see more prompting – and I don’t see anybody who is trying to build real guardrails and real tooling from scratch."
Skipper’s solution is a reactive runtime, drawing inspiration from Skip, the programming language Verlaguet developed at Facebook. This reactive model defines programs as declarative graphs of computation, mapping inputs to outputs. The underlying runtime automatically manages state, cache invalidation, and concurrency, thereby absolving the AI model of the complex task of reasoning about how changes in one part of the state graph might affect dependent parts.
"I think it’s because it’s a lot of work," Verlaguet attributed the lack of similar approaches from competitors to the sheer effort involved. "And so it’s much easier to make those big claims that you’re going to do these things when, in the end, you don’t." He draws a parallel to the impact of React on UI development, suggesting Skipper applies a similar declarative, component-based philosophy to backend services. In this model, JavaScript manages the computational links between graph nodes, while the Skip runtime handles the underlying complexities.
Launch Details and Future Trajectory
The initial launch of Skipper on Monday delivers on the core promise of the closed-loop experience: transforming a single prompt into a deployed, running service. Skipper is designed to be multi-model, intelligently routing tasks to various foundation models based on their suitability. Claude Opus is set as the default, with Anthropic’s Sonnet and Haiku also integrated. SkipLabs emphasizes that this model selection is an architectural choice, not a product dependency, indicating flexibility and an openness to integrating other models.
"We don’t do any AI here," Verlaguet had previously stated, underscoring the company’s view of AI models as a commodity. "We treat the models as a commodity. To us, the model is just an API that we call with a context; it comes back with the result."
Also included in the launch is the capability for Skipper-generated services to integrate with external services. This allows the produced software to interact with external APIs, fetch live data, and communicate with other systems, ensuring that the output is not confined to a sandbox but is immediately functional within a real-world stack.
Looking ahead, two key capabilities were slated for release shortly after the initial launch but were not ready for the default shipping: an incremental TypeScript implementation, referred to as SKJS, and an incremental update mode. The latter would enable developers to modify a specification and have Skipper update the running service without requiring a full rebuild. While these features were close to completion and included in the product roadmap, SkipLabs opted for a cautious rollout to ensure stability. "We’re not going to make it the default quite yet because we need to iron out a couple of things," Verlaguet noted in a pre-launch briefing, "but we should be able to release the type system [quickly after launch]."
The Infrastructure Thesis: Anticipating the Bottleneck
Beyond the immediate product offering, Verlaguet’s broader argument with Skipper addresses the evolving bottleneck in AI-assisted software development. He contends that the rapid advancement of AI models is already outpacing the development of tooling capable of effectively validating their output.
"In the next few years, it’s not going to be okay to wait for CI to take half an hour to an hour to validate a diff," Verlaguet predicts. "AI is getting faster and faster, and so you will need the tooling to guardrail the AI, and that tooling will need to be incremental." This foresight underpins the strategic development of the incremental TypeScript reimplementation (SKJS), designed to provide fast, reliable feedback to the AI during the generation process, rather than solely relying on end-of-process validation.
The financial backing for SkipLabs underscores the perceived potential of this vision. The company secured an $8 million seed round led by Amplify Partners, which, in its investment memo, characterized Verlaguet as "one of the top two to three programming language designers in the world." The round also attracted notable angel investors, including Yann LeCun, a Turing Award winner and former Meta Chief AI Scientist, and Spencer Kimball, co-founder and CEO of Cockroach Labs.
Skipper is now available for exploration and use at skipperai.dev, inviting developers to experience this new paradigm in AI-assisted software creation. The company’s approach signals a significant shift, moving beyond the immediate gratification of faster code generation towards a more sustainable and architecturally sound future for software development.
