July 28, 2025
Most people use Claude Code for, well, coding. But I discovered something interesting: it's incredibly effective for non-coding tasks that require deep research and a structured approach. I was putting together a research grant application and needed to describe what we'd been working on. In the past, I've struggled with GPT API integrations and CustomGPTs, which have produced mediocre results for research and writing tasks. Naturally, I was skeptical if Claude Code would do a better job, but I decided to give it a try.
I got this idea from Ben Mann, Anthropic's co-founder, who mentioned on Lenny's podcast that Anthropic already uses Claude Code internally for legal and finance teams - non-coding applications where precision and structured thinking matter.
This year, as every year, we wanted to apply for a research grant. That meant I had to turn a mountain of user research data, technical documentation, and survey results from our work at Mimo into a formal German academic research document. I did a lot of writing at my university a decade ago, but in the last ten years, writing academic papers hasn't exactly been my day job.
The following is what I did for every grant application for every project I worked on. Let me walk you through the first one: our persistent coding environment project.
First, I gathered all the materials I thought Claude Code would need and provided them systematically:
├── Historical Examples
│ └── Previous research grant applications (2021-2023)
├── Grant Application Template
│ └── Required structure for the grant application
├── Project Description
│ └── An overview of the project
└── Research Data
└── Internal user studies, technical analyses, results of A/B tests, etc.Claude Code analyzed all these materials and created a CLAUDE.md file that captured our documentation standards and the application requirements, which meant it could maintain consistency with the needed format.
I provided Claude Code with an overview of our project and asked it to draft a research grant application plan, incorporating our existing research, previous applications, and documentation. When I requested to include external research, Claude Code initially focused on market data and growth metrics, but I had to correct course - I needed external academic research specifically about project-based learning effectiveness.
After my clarification, Claude Code came up with a much better plan. It would focus on how our coding environments enable effective project-based learning, addressing the research gap between theoretical instruction and practical application in programming education. This was exactly what I needed.
Claude Code effectively pre-selected potentially relevant papers, and already told me what parts it would use. I accepted some papers and rejected others.
Claude Code drafted the research grant application by synthesizing our research data, my project overview, and the previously approved research. It seamlessly wove these external citations together with our internal survey data, test results, and technical documentation.
A core part of this workflow was my instruction to have Claude Code load all referenced papers and save them as PDFs. It efficiently located and downloaded all the academic papers it had cited and then made them available for my review. Having them readily available to read was a tremendous help. Since I'm skeptical of LLMs citing fabricated sources or making up parts of the quotes, double-checking and reading the papers was crucial.
The initial document was solid but too long for our application requirements. I made some manual adjustments where I thought we should formulate things differently, then asked Claude Code to shorten it while preserving all external references and core quantitative data. Claude Code did a great job at this, and I only had to make a few additional adjustments.
There was one small issue: Sometimes, Claude Code's edits didn't save properly to the file, as it kept an older version in its context. I had to prompt it to reread the document to work with the current version. This was a minor hiccup in an otherwise smooth process, and Claude Code handled the correction well once I pointed it out.
What started as scattered user research data and technical specs became a polished academic research documentation that:
The final document positioned our technical innovation within the broader context of educational research, showing how our project addresses real pedagogical challenges identified in the academic literature.
What I loved about this process was that I remained in complete control throughout. Claude Code's planning mode let me fine-tune the approach (e.g., selecting which papers to include) before any writing began. After the initial draft, I could easily request changes, whether finding different sources or revising how information was presented, if something didn't capture the right essence. And for minor edits, I simply made changes directly in the document. The collaborative experience with Claude Code was truly seamless and delightful.
I spent a total of $36.84 on this project and 4 similar research documentation tasks. That's roughly $7.37 per comprehensive research document, including academic citations, data integration, and well-structured formatting.
This was incredibly cost-effective compared to hiring a research assistant or doing all the research and writing myself. More importantly, I maintained full control over the direction and could iterate in real time based on my specific requirements.
I got a research assistant who could write academic German, find and download relevant papers, integrate external research with internal data, and produce documentation that met our grant application standards.
The time savings were substantial. Researching, writing, and formatting this would have required a lot of manual work. Instead, it took a few hours of collaborative iteration to get exactly what I needed.
Would I use Claude Code for academic writing again? Absolutely. The combination of research capabilities, writing skills, and file management made it an incredibly effective collaborator for transforming raw research data into polished academic documentation.