Mise en place (MEP), as proposed in Mise en Place for Agentic Coding: Deliberate Preparation as Context Engineering Methodology, is a structured preparation methodology for agentic coding that borrows the French culinary term for “everything in its place”. Its premise, in the author’s words, is that the alignment problem in agentic coding is fundamentally a preparation problem: when domain expertise, design intent and task boundaries are externalized into structured artifacts before agents begin writing code, the resulting implementation is argued to be more aligned, more coherent and less costly to verify. MEP formalizes that preparation into three sequential, phase-gated steps — contextual grounding, collaborative specification and task decomposition — each producing artifacts that feed the next, all completed before implementation begins, after which parallel agents execute the decomposed work and their outputs converge in integration verification.
Usage
In contextual grounding, the practitioner externalizes domain expertise and tacit knowledge into briefing documents — markdown files encoding domain knowledge, competitive analysis and design philosophy — that form a persistent context layer agents consult throughout implementation; following backward design, this starts from outcomes rather than features. In collaborative specification, structured human-agent dialogue (the practitioner describes intent, the agent proposes details, the practitioner accepts, rejects or modifies) produces a design document; its key mechanism is encoding value judgments as specification constraints, including what to exclude, and recording the why as well as the what so that agents can make aligned micro-decisions without escalating. In task decomposition, the specification becomes dependency-aware task records with priorities and acceptance criteria — the author used Beads — so that multiple agents can work simultaneously; the coordination burden shifts from runtime to preparation time.
The paper positions MEP relative to Spec-driven development, which it extends with a phase for tacit, value-laden knowledge, and to Prompt engineering, from which it differs in scope: prompt engineering tunes individual model invocations, whereas MEP structures the workflow-level artifacts that precede and constrain them. It presents MEP as the opposite end of the spectrum from Vibe Coding, front-loading alignment work that iterative flows pay incrementally as rework.
When It Applies
The author argues that agentic workflows need this kind of preparation because agents lack the tacit context human collaborators carry and cannot iterate cheaply without expensive regeneration. MEP assumes a practitioner able to articulate the relevant domain knowledge and to decompose work into independently executable units with explicit dependencies, and it is aimed at setups where several agents implement in parallel. The paper does not name misapplication modes; among its open questions are what counts as “sufficient” context and when preparation should stop, and whether the method scales from prototyping to multi-month development.
Its evidential basis is thin by the author’s own account: a single five-hour hackathon with one practitioner and no control group, in which about two hours of preparation preceded parallel implementation by four agents with no structural refactoring needed afterward. The author acknowledges that this cannot establish causation and that their own dual background in education and software engineering confounds the result, and describes MEP as a starting framework that formalizes practices experienced practitioners converge on independently rather than a validated method.