Single Conversation Methodology: A Human-Centered Protocol for AI-Assisted Software Development
Salvador D. Escobedo · 2025
Read Single Conversation Methodology: A Human-Centered Protocol for AI-Assisted Software Development by Salvador D. Escobedo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-12665.txt.
Introduction
Salvador Escobedo puts forth a plan to do a whole software job in one long chat with a language model. Loose prompts scatter a job's memory across lots of chats with no links. His plan, the Single Conversation Methodology, keeps all steps in one long chat. Needs, frame, and code build all sit in that one thread. Long context turns into the job's live log.
Four rules ground the plan: clear thought, full trace, parts that split clean, and good docs. The coder stays chief and boss of the tool, not a slack taker of code it makes. That fix aims at a real drift: the tool can take the wheel in a quiet way. Escobedo gives firm rules for AI-aided code work, with human calls in charge by design.
Catalog record
| CREATOR | Salvador D. Escobedo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-12665
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Single Conversation Methodology: A Human-Centered Protocol for AI-Assisted Software Development Salvador D. Escobedo Abstract We propose the Single Conversation Methodology (SCM), a novel and pragmatic approach to software development using large language models (LLMs). In contrast to ad hoc interactions with generative AI, SCM emphasizes a structured and persistent development dialogue, where all stages of a project—from requirements to architecture and implementation—unfold within a single, long-context conversation. The methodology is grounded on principles of cognitive clarity, traceability, modularity, and documentation. We define its phases, best practices, and philosophical stance, while arguing that SCM offers a necessary correction to the passive reliance on LLMs prevalent in current practices. We aim to reassert the active role of the developer as architect and supervisor of the intelligent tool. 1 Introduction The integration of large language models (LLMs) into software development has introduced significant changes in how programmers interact with code [ 7 , 9 ] . These models offer unprecedented capabilities in code generation, refactoring, and debugging, reshaping not only development workflows but also the cognitive practices of developers [ 16 , 18 ] . However, their ease of use and on-demand nature have also led to engagement patterns that can undermine key aspects of sound software engineering [ 11 , 5 ] . A dominant pattern has emerged: developers engage in short, goal-oriented prompts with minimal contextual continuity. A typical session consists of requesting a function, copying the output, and terminating the interaction. This piecemeal approach, while expedient, promotes what we may call a fragmented development mindset . Code is accepted based on surface-level utility rather than on architectural soundness, long-term maintainability, or conceptual clarity [ 8 , 11 ] . This trend echoes what has been described as vibe coding —a style of programming characterized by intuition, speed, and exploratory trial-and-error rather than deliberate design [ 3 , 14 , 5 ] . While useful in early prototyping or creative experimentation, vibe coding becomes problematic when applied to complex systems or production-grade software. Its informal, improvisational nature often leads to opaque logic, weak modular boundaries, and fragile interdependencies. When LLMs are used in this mode, they tend to amplify these weaknesses by encouraging rapid iteration without reflection. The central risk is not merely technical debt, but cognitive erosion. As developers outsource more of the planning and reasoning to the model, they may lose their grip on the architectural vision of the project [ 1 , 10 ] . Over time, this can result in software systems that are difficult to reason about, evolve, or even explain. The Single Conversation Methodology (SCM) was conceived as a principled alternative to such practices. Rather than reducing the human developer to a passive prompter, SCM proposes a development protocol in which the conversation with the LLM is treated as a coherent, structured, and persistent workspace. The developer remains in full control of the design direction, using the model as a context-rich assistant rather than a disposable code generator. SCM restores a sense of continuity and intentionality to AI-assisted development. It encourages disciplined reasoning, architectural awareness, and modular design through a methodical use of dialogue. In this way, it addresses both the practical shortcomings of ad hoc prompting and the philosophical concern of maintaining human agency in an increasingly automated workflow [ 1 , 17 ] . In practice, the Single Conversation Methodology has been successfully applied using Claude 3.7, whose extended context window enables coherent long-form development sessions. However, the methodology is model-agnostic in principle: any sufficiently capable language model with a large enough context window to maintain architectural continuity and recall prior interactions can support SCM. The key requirement is not a specific model, but the ability to sustain a persistent, structured dialogue where design reasoning, implementation steps, and revisions remain accessible throughout the project lifecycle. 2 Methodology Overview The Single Conversation Methodology (SCM) reimagines the process of software development in the age of large language models by treating the LLM interaction itself as the primary development environment. Unlike conventional workflows fragmented across multiple tools and prompts, SCM unfolds entirely within a single, persistent conversation thread. This conversation becomes the developer’s workspace, memory, and reasoning context. The model assists not merely as a code emitter but as a high-context collaborator throughout the lifecycle of the project. 2.1 Core Principle The conversation is the development environment. The developer remains the architect; the LLM serves as a high-context assistant. This principle underlines a clear division of roles: the human retains strategic control, while the model contributes local insights, structured outputs, and memory continuity. The use of conversational programming interfaces as assistant personas has been explored in recent work [ 15 ] , validating the idea that dialogic interaction enables richer forms of collaboration than traditional code generation. SCM restores intentionality and coherence to the development process by grounding every action within a shared, evolving context. At the same time, critical oversight remains necessary to ensure correctness and trust [ 17 ] . 2.2 Phases of the Methodology SCM unfolds through a structured sequence of phases, each governed by explicit transitions and all anchored in the same conversational thread. This design ensures that architectural decisions, rationales, and implementation history are preserved within a unified dialogue. The overall structure recommended by SCM has three stages: (1) The grounding phase , where the generalities of the project and setup are discussed and defined. No code is generated in this phase. (2) The code generation phase, where the code is actually generated, following certain cyclic patterns, and (3) the documentation phase, where design concepts and project general documentation artifacts are generated. (see figure 1 ) Code generation phase usually takes the largest part of the dialogue, but grounding phase can also be extensive; it depends on the project and the capacity of the model window. Grounding phase Code Generation Documentation General SCM structure Temporal Flow of the Conversation Figure 1: The SCM has three main stages: Grounding phase, Code Generation and Documentation. 2.2.1 Grounding Phase (No Code) In this first stage, the project begins with a conceptual grounding discussion. No code is written during this phase. Instead, the goal is to establish a mutual understanding between the developer and the model. • Define functional and non-functional requirements. • Describe the overall architectural vision and technological stack. • Clarify domain boundaries, design patterns, and architectural constraints. • Align expectations and clarify terminology to avoid downstream ambiguity. Only after a coherent conceptual base is established should the development of actual code begin. This avoids premature generation and ensures that subsequent outputs are contextually grounded. Recent studies highlight the advantages of long-context interaction with LLMs [ 18 ] , confirming the value of sustained, coherent dialogue in complex tasks such as architectural planning. Project Goals and Vision Technology Stack Selection Architectural Overview Security Considerations Pattern design and Interfaces Setup and repository structure Grounding Phase Conversation Overview Temporal Flow of the Conversation Figure 2: Illustration of distinct thematic blocks in an LLM-based Grounding Phase. Each section corresponds to a focused design discussion. The schematic conversation in the figure 2 exemplifies the purpose of the Grounding Phase: establishing shared vocabulary, validating requirements, and framing the architectural vision—without writing a single line of code. Only after