# Wolverine Enables AI-Friendly Vertical Slice Development > Source: > Published: 2026-06-04 20:55:20.103415+00:00 # Wolverine Enables AI-Friendly Vertical Slice Development On June 4, 2026, Jeremy D. Miller, creator of the .NET "Critter Stack" (the Wolverine framework and Marten library), published the blog post "The Codebase Is the Prompt: Wolverine, Vertical Slices, and AI-Assisted Development" arguing that because a coding agent's context window is finite, the structure of a codebase effectively becomes part of the prompt (jeremydmiller.com). Miller makes the case that Wolverine's take on vertical slice architecture is AI-friendly because it aggressively compresses the code written and read per feature, removing the controller-mediator-handler ceremony of layered designs, while its "A-Frame" testability keeps slices easy to verify (jeremydmiller.com). Editorial analysis: the piece is an opinionated, partly promotional argument rather than a benchmark, but it reflects a real industry pattern in which code locality and consistent idioms reduce the tokens and irrelevant context an LLM assistant must process. ### What happened On June 4, 2026, Jeremy D. Miller published "The Codebase Is the Prompt: Wolverine, Vertical Slices, and AI-Assisted Development" on jeremydmiller.com. Miller is the creator of the .NET "Critter Stack" - the Wolverine messaging and HTTP framework and the Marten document and event-store library. He frames the post as part of a forthcoming strategy document on the Critter Stack's approach to AI, and argues that because an AI coding agent has a finite context window, the structure of a codebase effectively becomes part of the prompt (jeremydmiller.com). ### The argument Miller contrasts traditional layered .NET designs - where one feature is spread across controllers, mediator request and handler types, repositories, DTOs, and mapping code - with vertical slice architecture, which organizes code by feature so a single slice owns its whole pathway (jeremydmiller.com). He credits Jimmy Bogard with popularizing the pattern and notes the .NET community often conflates it with MediatR (jeremydmiller.com). His core claim is that Wolverine compresses a slice "about as far as the language allows": with Wolverine.Http the endpoint is the handler, with no controller calling a mediator calling a handler, and its "A-Frame" approach to testability keeps each slice easy to verify (jeremydmiller.com). Fewer artifacts, he argues, mean less boilerplate for an agent to generate correctly and less surface area to get wrong. ### Why it matters Editorial analysis: the post reflects a genuine and increasingly discussed industry pattern - that code locality, smaller per-feature surface area, and consistent idioms materially affect how well LLM coding assistants perform, because a context window imposes a practical budget on the files and symbols an agent can load. Teams adopting feature-local layouts often report lower friction with repository-indexing and retrieval-augmented tooling. ### Worth noting Editorial analysis: this is a single-source, opinionated engineering argument from the tool's own author, not an independent benchmark, and it is partly promotional - Miller points readers to JasperFx's paid "AI Skills" for the Critter Stack and a forthcoming "CritterWatch" visualization tool. The underlying adoption signal is real, though: Miller's June 3 post reports that Wolverine's main NuGet package passed five million downloads. ## Scoring Rationale An opinionated, single-source blog post from Wolverine's own creator arguing that vertical slice architecture makes .NET codebases more amenable to AI coding agents. The "codebase is the prompt" framing is timely and relevant to practitioners, and Wolverine is an established tool (5M+ NuGet downloads), but the piece is niche to .NET and partly promotional rather than a benchmark, release, or independent finding. Solid niche practitioner content rather than an industry-moving event. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)