"What is the Reason You're Even Creating Documents in Excel?"──Breaking the 'Convention' That Stalled AI Utilization.
OM Network Co., Ltd. (OMN) has implemented Markdown conversion for design documents in its leading development team to overcome the "Excel barrier" that was hindering AI utilization. Full-scale operation began in April 2026, speeding up coding and internal knowledge sharing, and establishing a new operational flow where Markdown files used for coding also function as design specifications. This project spearheads OMN's acceleration towards true "co-creation with AI."
📋 Article Processing Timeline
- 📰 Published: May 8, 2026 at 18:10
- 🔍 Collected: May 8, 2026 at 09:31
- 🤖 AI Analyzed: May 8, 2026 at 10:33 (1h 1m after Collected)
OM Network Co., Ltd. (Headquarters: Chuo-ku, Niigata City, Niigata Prefecture; Representative Director: Shinya Yamagishi; hereinafter OMN) has proactively implemented Markdown conversion for design documents within its frontline development team to break down the "Excel barrier" that was hindering AI utilization. Full-scale operation commenced in April 2026, leading to improved coding speed and internal team knowledge sharing, and establishing a new operational flow where Markdown files used for coding also function as design specifications. With this project as a starting point, OMN will accelerate towards true "co-creation with AI."
1. The "Optimization Opportunity" of Document Formats Revealed by AI Utilization
OMN introduced generative AI "Claude Code" as a standard tool in its development sites with the aim of improving development efficiency and quality.
Since its introduction, work speed in code generation and specification verification has significantly improved. However, as the company pursued higher utilization accuracy across the entire team, operational challenges emerged, such as rare cases where the AI misunderstood the content of specifications or couldn't correctly interpret information hierarchies. Initially, efforts were made to address this by refining prompts (instructions), but as instructions became more detailed, the prompts themselves became more complex, paradoxically increasing the difficulty of operation. An unexplainable sense of frustration permeated the team, suggesting a fundamental misalignment with AI.
2. Questioning the "Obvious"──A Fundamental Inquiry and Proposal from the AI Team
To overcome this situation, the development team consulted the internal AI specialized team. The initial objective at that time was purely technical: "how to accurately parse complex Excel files with AI," aiming to improve reading accuracy.
However, the AI specialized team posed a surprising and fundamental question to the development team, who had confessed their current troubles:
"What is the reason you're even creating documents in Excel in the first place?"
This single phrase was, for OMN's development team at the time, literally an "eye-opening" experience. The development team had been constrained by the preconception that "specifications are made in Excel," only considering ways to force AI to understand human-oriented formats.
However, the optimal format for AI should be one with a clear information structure, not a human-oriented layout. "Instead of making AI adapt to us, we should provide the format in which AI can perform best." This moment of questioning the "obvious" and shifting perspective by 180 degrees became the starting point for OMN's next-generation document DX.
3. Information Assetization and Validation of Effectiveness Through Markdown Refresh
Validation results clearly showed that "Markdown (md)" with a simple hierarchical structure is the most accurate and fastest format for AI (Claude Code) to parse. Excel files retain a large amount of metadata internally, such as formatting, cell positions, and styles, leading to inflated token consumption when AI reads the same content. In contrast, Markdown contains almost no redundant information other than structural symbols and the main text. Based on OMN's trial operation estimates, the token consumption for processing equivalent document content is felt to be reduced to approximately one-tenth compared to via Excel (※1).
With AI now understanding intentions as if they were "on the same wavelength," the wasteful correction instructions to AI (re-doing prompts) that traditionally bothered engineers have drastically decreased. Developers can now implement at the fastest speed without interruption. Based on these results, OMN's development team proceeded with the Markdown conversion for most major design documents and development materials by April 2026.
The scope of the refresh was not limited to formal specifications. Technical memos and knowledge bases, which tended to be scattered until now, were also consolidated in Markdown format. As a result, context sharing with AI became smoother, and discrepancies in understanding during information sharing among engineers decreased. The single decision to align information format with AI transformed on-site knowledge into a powerful asset for operating AI most intelligently.
■ Voices from the Field: How dialogue with AI has changed – a dramatic shift from "interpretation" to "implementation"
An engineer actually responsible for development in this project describes the changes brought about by this refresh as follows:
"With the previous Excel-based design documents, getting AI to understand the specifications was the biggest hurdle. Even when we tried to make the AI grasp our intentions...
1. The "Optimization Opportunity" of Document Formats Revealed by AI Utilization
OMN introduced generative AI "Claude Code" as a standard tool in its development sites with the aim of improving development efficiency and quality.
Since its introduction, work speed in code generation and specification verification has significantly improved. However, as the company pursued higher utilization accuracy across the entire team, operational challenges emerged, such as rare cases where the AI misunderstood the content of specifications or couldn't correctly interpret information hierarchies. Initially, efforts were made to address this by refining prompts (instructions), but as instructions became more detailed, the prompts themselves became more complex, paradoxically increasing the difficulty of operation. An unexplainable sense of frustration permeated the team, suggesting a fundamental misalignment with AI.
2. Questioning the "Obvious"──A Fundamental Inquiry and Proposal from the AI Team
To overcome this situation, the development team consulted the internal AI specialized team. The initial objective at that time was purely technical: "how to accurately parse complex Excel files with AI," aiming to improve reading accuracy.
However, the AI specialized team posed a surprising and fundamental question to the development team, who had confessed their current troubles:
"What is the reason you're even creating documents in Excel in the first place?"
This single phrase was, for OMN's development team at the time, literally an "eye-opening" experience. The development team had been constrained by the preconception that "specifications are made in Excel," only considering ways to force AI to understand human-oriented formats.
However, the optimal format for AI should be one with a clear information structure, not a human-oriented layout. "Instead of making AI adapt to us, we should provide the format in which AI can perform best." This moment of questioning the "obvious" and shifting perspective by 180 degrees became the starting point for OMN's next-generation document DX.
3. Information Assetization and Validation of Effectiveness Through Markdown Refresh
Validation results clearly showed that "Markdown (md)" with a simple hierarchical structure is the most accurate and fastest format for AI (Claude Code) to parse. Excel files retain a large amount of metadata internally, such as formatting, cell positions, and styles, leading to inflated token consumption when AI reads the same content. In contrast, Markdown contains almost no redundant information other than structural symbols and the main text. Based on OMN's trial operation estimates, the token consumption for processing equivalent document content is felt to be reduced to approximately one-tenth compared to via Excel (※1).
With AI now understanding intentions as if they were "on the same wavelength," the wasteful correction instructions to AI (re-doing prompts) that traditionally bothered engineers have drastically decreased. Developers can now implement at the fastest speed without interruption. Based on these results, OMN's development team proceeded with the Markdown conversion for most major design documents and development materials by April 2026.
The scope of the refresh was not limited to formal specifications. Technical memos and knowledge bases, which tended to be scattered until now, were also consolidated in Markdown format. As a result, context sharing with AI became smoother, and discrepancies in understanding during information sharing among engineers decreased. The single decision to align information format with AI transformed on-site knowledge into a powerful asset for operating AI most intelligently.
■ Voices from the Field: How dialogue with AI has changed – a dramatic shift from "interpretation" to "implementation"
An engineer actually responsible for development in this project describes the changes brought about by this refresh as follows:
"With the previous Excel-based design documents, getting AI to understand the specifications was the biggest hurdle. Even when we tried to make the AI grasp our intentions...