Post
JA EN

Solo Development in the AI Era and the Bus Factor Problem: Do You Need a System Where Reviewers Can Take Over Immediately?

Solo Development in the AI Era and the Bus Factor Problem: Do You Need a System Where Reviewers Can Take Over Immediately?
  • Target Audience: Engineering managers, tech leads, engineers working solo or in small teams
  • Prerequisites: Basic knowledge of Git, code review, and CI/CD
  • Reading Time: Approximately 12 minutes

Overview

The evolution of AI coding assistants has ushered in an era where individuals can achieve high productivity on their own. However, if the DORA 2025 Report is correct that “AI doesn’t fix a team; it amplifies what’s already there”1, doesn’t this mean that the vulnerabilities of solo development setups are also amplified?

This article examines solo development practices from the perspective of “bus factor” as an organizational risk, building on the SDLC transformation discussed in Building AI-Native Engineering Teams and the code review frameworks presented in Organizational Challenges and Solutions.


The Light and Shadow of “AI as an Amplifier” in Solo Development

Productivity Amplification

As explained in Building AI-Native Engineering Teams, AI coding agents support developers across all phases of the SDLC. OpenAI internally (with 92% of engineers using it) reported a 70% increase in weekly PR merges2, and a controlled experiment with GitHub Copilot showed a 55.8% speed improvement3.

This has made it possible for development work that previously required a team to be completed by a single person.

Risk Amplification

However, recall the core message of the DORA 2025 Report:

“AI doesn’t fix a team; it amplifies what’s already there.”1

This means that the vulnerabilities of solo development setups are also amplified.

flowchart TB
    subgraph Amplification["AI Amplification Effect"]
        direction TB
        Good["Productivity Amplification<br>(One person, many outcomes)"]
        Bad["Risk Amplification<br>(Increased dependency on one person)"]
    end

    subgraph Result["Result"]
        direction TB
        R1["High Individual Productivity"]
        R2["Bus Factor of 1"]
        R1 --- R2
    end

    Good --> R1
    Bad --> R2

The Bus Factor Problem: New Dimensions in the AI Era

The Bus Factor refers to “the number of team members who would need to suddenly disappear before the project is in jeopardy”4. A bus factor of 1 means there is a single point of failure (SPOF) within the team.

In the AI era, the following risks are added to traditional key person dependency risks:

New RiskDescription
Siloed AI KnowledgeIndividually optimized prompts and AGENTS.md configurations aren’t shared5
Overconfidence in AI OutputThe psychological distance of “AI wrote it, so it must be fine”1
The High Productivity TrapBecause one person can do so much, there’s less perceived need for knowledge sharing

Common to All Levels: Building the Shared Foundation

For any project, establish the following “shared foundation.” This forms the basis for handover capability.

flowchart TB
    subgraph Foundation["Shared Foundation (Common to All Levels)"]
        direction TB
        F1["README.md<br>Environment Setup"]
        F2["AGENTS.md/CLAUDE.md<br>Standardized AI Config"]
        F3["Tech Stack Documentation"]
        F1 --- F2 --- F3
    end

    Foundation --> L1["Level 1"]
    Foundation --> L2["Level 2"]
    Foundation --> L3["Level 3"]

Minimum Required Documentation

DocumentContentEstimated Effort
README.mdEnvironment setup, basic architecture2-3 hours
AGENTS.md/CLAUDE.mdAI settings, commands, project structure1-2 hours
Tech Stack ListLanguages, frameworks, infrastructure30 minutes

Basic AGENTS.md Structure

The AGENTS.md explained in the AI-Native article also functions as a tool that improves handover capability6:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
# AGENTS.md

## Commands
npm test              # Run tests
npm run build         # Build
npm run deploy:staging # Deploy to staging

## Project Structure
src/
  ├── components/   # UI components
  ├── services/     # Business logic
  └── api/          # API layer

## Code Style
- Naming: camelCase (variables), PascalCase (components)

Choosing the Right Structure Based on Project Scale

Once you’ve established the shared foundation, select additional measures based on the nature of your project.

flowchart TB
    Start["Project Start"] --> Base["Build Shared Foundation<br>(Required for All Projects)"]
    Base --> Q1{"Duration<br>> 3 months?"}
    Q1 -->|No| L1["Level 1<br>Minimal Measures"]
    Q1 -->|Yes| Q2{"Business<br>Critical?"}
    Q2 -->|No| L2["Level 2<br>Standard Structure"]
    Q2 -->|Yes| L3["Level 3<br>Full Structure"]

Level 1: Minimal Measures (Short-term / Experimental Projects)

Target: Under 3 months, prototypes, experimental development, systems that can be quickly rebuilt

In addition to the shared foundation:

MeasureEffortContent
Code commentsAs neededExplanations for complex logic

Structure Image:

1
Developer (1) + AI → Shared Foundation → Accessible to anyone

Estimated Handover Time: 1-2 weeks


Level 2: Standard Structure (Medium-term / Regular Business Systems)

Target: 3 months to 1 year, internal systems, moderate importance

In addition to Level 1:

MeasureEffortContent
Assign Regular Reviewer1-2 hours/weekCan be part-time from another project
ADR (Architecture Decision Records)15-30 min eachDocument important decisions
Monthly Code Walkthrough30 min/monthKnowledge sharing with reviewer

Structure Image:

flowchart TD
    Dev["Developer + AI"] -->|"PR"| R1["Reviewer<br>(Part-time OK)"]
    R1 -->|"Approve"| Dev
    Dev --> Shared["Shared Foundation<br>README / AGENTS.md / ADR"]
    R1 -.-> Shared

Key Point: Position the reviewer not as an “approver” but as a “potential successor,” deepening their understanding of the code.

Estimated Handover Time: 3-5 days


Level 3: Full Structure (Long-term / Business Critical)

Target: Over 1 year of operation, business critical, regulated industries (finance, healthcare, etc.)

In addition to Level 2:

MeasureEffortContent
Multiple Reviewer Structure2-4 hours/week2+ reviewers for redundancy
CI/CD Documentation Auto-generationInitial setupAPI docs, changelog
Quarterly Handover Simulation1-2 days/quarterPractical verification
Regular Bus Factor Score MeasurementMonthlyContinuous improvement

Structure Image:

flowchart TD
    Dev["Developer + AI"]
    Dev -->|"PR"| R1["Reviewer A"]
    Dev -->|"PR"| R2["Reviewer B"]
    R1 -->|"Approve"| Dev
    R2 -->|"Approve"| Dev
    Dev --> Shared["Shared Foundation (Auto-updated)<br>README / AGENTS.md / ADR<br>API Docs / Changelog"]
    R1 -.->|"Can take over"| Shared
    R2 -.->|"Can take over"| Shared

Estimated Handover Time: 1-2 days


Level Comparison

 Level 1Level 2Level 3
Initial Effort3-5 hours5-10 hours1-2 days
Ongoing EffortAlmost none2-3 hours/month2-4 hours/week
ReviewersNone1 (part-time)2 or more
Handover Time1-2 weeks3-5 days1-2 days
Risk ToleranceHighMediumLow

To objectively measure “always ready for handover” status, you can use the following scoring system.

Bus Factor Score (5 points × 5 items = 25 points):

Aspect1 Point (Danger)3 Points (Caution)5 Points (Good)
Environment SetupRequires verbal explanationMostly possible with README.mdFully automated
Code UnderstandingOnly developer understandsComments/ADR availableReviewer also understands
DeploymentManual/key person dependentScriptedFully automated CI/CD
Incident ResponseOnly developer can respondRunbook availableReviewer can also respond
Design DecisionsTacit knowledge onlyPartial ADRComplete ADR

Target Scores:

  • Level 2: 15+ points (3+ points per item)
  • Level 3: 20+ points (4+ points per item)

Conduct a simulated handover drill quarterly to verify whether handover is truly possible.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
# Handover Simulation Checklist

## Phase 1: Preparation (Day 1)
- [ ] Assume the developer is "absent" for 1 week
- [ ] Reviewer acts as primary developer

## Phase 2: Practice (Days 2-5)
- [ ] Environment Setup: Build dev environment using only documentation
- [ ] Code Understanding: Understand changes from the past month
- [ ] Bug Fix: Fix 1 minor bug
- [ ] Deployment: Execute deployment to staging environment

## Phase 3: Retrospective (Day 5)
- [ ] Identify missing documentation
- [ ] Document tacit knowledge
- [ ] Define improvement actions

Summary

Development in the AI era has made high productivity possible for small elite teams or individuals. However, the DORA Report’s principle that “AI is an amplifier” means that the vulnerabilities of solo development setups are also amplified.

Choose a structure based on the nature of your project:

ProjectRecommended LevelMinimum Requirements
Short-term / ExperimentalLevel 1Shared Foundation (README.md, AGENTS.md)
Medium-term / Regular BusinessLevel 2+ 1 Reviewer + ADR
Long-term / CriticalLevel 3+ Multiple Reviewers + Handover Simulation

The key insight is that what you “can do alone” and what you “should do alone” are different. Establish a “shared foundation” at every level, and add measures according to the importance of the project.


Articles covering related themes:


References

References corresponding to citation numbers in the main text, listed in numerical order.

Additional References (Not Cited by Number in Main Text)


On Citation Accuracy: The research cited in this article has been verified through academic databases (arXiv, Google Scholar, etc.) and official websites.

On Research Limitations:

  • GitHub Copilot paper (Peng et al., 2023): Controlled experiment with 95 professional developers. Limited to HTTP server construction tasks in JavaScript.
  • OpenAI Codex: Internal OpenAI usage data. Reproducibility in external companies is unverified.
  • DORA 2025: Self-reported survey of approximately 5,000 people.
  1. Announcing the 2025 DORA Report - Google Cloud Blog (2025). [Reliability: High] ↩︎ ↩︎2 ↩︎3

  2. Announced at OpenAI DevDay 2025. Reported in: Everything OpenAI Released on DevDay 2025, Explained - The Neuron (2025). [Reliability: Medium-High] ↩︎

  3. The Impact of AI on Developer Productivity: Evidence from GitHub Copilot - Peng, S., et al. (2023). arXiv (pre-peer review preprint). [Reliability: Medium-High] *Controlled experiment with 95 professional developers. Task completion time reduced by 55.8%. ↩︎

  4. Bus factor - Wikipedia - Wikipedia. [Reliability: Medium-High] ↩︎

  5. Generative AI Accelerates Personal Optimization—The Collapse of Knowledge Sharing Has Begun - CIO (2024). [Reliability: Medium-High] ↩︎

  6. How to write a great agents.md: Lessons from over 2,500 repositories - GitHub Blog (2025). [Reliability: Medium-High] ↩︎

This post is licensed under CC BY 4.0 by the author.