First thoughts for AI integration with my notes and dev workflow

Reflections on my scattered tools after a month of travel, and a plan to build a low-friction notes and development workflow around Obsidian, Git, Positron, and Claude.

Today is the first workday back after a month of travelling. It has been a productive month, with lots of new thoughts, connections, and opportunities. Especially after the Sharks International 2026 conference, I have a ridiculous amount of information to process and connections to try to hang onto as best as I can.

Currently, I have stuff scattered across Google Docs, Notion, handwritten in my notebook, GitHub repos, emails, etc. Over the past half a year or so, I’ve also done an enormous amount of work across multiple projects. The main ones being OceanIQ, Bowen Biodiversity Plan, and sharkabc3d. In the desperate struggle to get things done, while being basically nomadic as we moved across Japan from Taragi, Kumamoto to Kesennuma, Miyagi, a lot of information is just all over the place.

It’s tempting to jump right back into work (I already started having meetings before even getting home), but I think this is a good time to take a moment and get a system in place. With that much pressure to get things done, I’ve realized a lot of friction points and what tends to get lost / left undone. I’ve begun using Claude a lot more for my work, and started paying for the Claude Pro Max plan. This has definitely helped a lot with my productivity, but it’s a tool that needs some structure / constraints or it just accelerates the creation of mess. These AI tools are getting really good, but they’ll rapidly overwhelm poorly structured / documented systems with slop (which also gets created more when there are unclear constraints).

What I like about my current tools

Obsidian

I like the markdown file format that isn’t proprietary, easy to view and edit in many different software. I also like how clean and simple the UI is for idea generation, reviewing notes, etc. Clear priority towards documenting / capturing knowledge and facilitating distraction-free writing.

Google Docs

Easy collaboration, many people familiar with the application now so it’s become the universal standard for sending documents for review / comment. Gemini integration works pretty well, makes it easy to review individual changes.

Notion

Powerful platform, lots of choices for how to configure and use. Good for collaboration as well.

Claude

Becoming a very powerful tool, it can create a ton of output in a very short amount of time and effort. Great for interactively brainstorming or generating outputs, since it’s pretty good about incorporating feedback. It integrates well with my development flow as well, with extensions in Positron IDE.

Handwritten notes

Pure ideation with no support. In an ever more distracting world, this provides a refuge for listening to my own thoughts. No squiggly lines telling me that I am bad at spelling or grammar. No pushing AI to help me write what’s on my mind. No advertisements. Just me and my thoughts. Very essential to be part of thinking and mental health in general.

Git / GitHub

I combined these two together (even though technically different tools), since I use them in conjunction with each other. This might be an obvious tool for the longtime software devs, but I definitely didn’t learn how to use this in biological research from academia. It had some learning curve, but it has transformed how organized my work has become and truly enabled incremental progress. The commits track small, incremental changes over time. The issues and pull requests are great for managing work, making it easier to stay focused.

Positron IDE

Relatively new IDE, optimized for data science workflows in R and Python. Built from a fork of Visual Studio Code IDE. I like that it uses an established open source IDE that is very popular, so it can take advantage of the plugins available there. While it’s tailored for data science, it’s better featured for web development and other programming fields compared to RStudio. Good for my work that is increasingly cutting across multiple fields. The AI integration is good too, whereas it’s nonexistent in RStudio. Very active development community.

Zotero

Open-source reference manager. I like the browser extension and online backup service, making it easy to keep all of the literature that I have read and taken notes on in one place. Pretty decent application for reading and reviewing papers.

In general

I think I’ve been improving my ability to document my process. Having AI helps a lot to finish notes / thoughts, since I tend to get distracted and move on from a task once complete or from an idea once I’ve applied it, before fully documenting the thought.

What I don’t like about my tools (or how I currently use them)

Obsidian

Collaboration is a bit tougher, since most people are not used to using the software. Base Obsidian (without plugins) doesn’t have a clear way to use AI. There are definitely plugins and ways to integrate AI well with Obsidian, but I just haven’t done that yet.

Google Docs

So freaking distracting, so many AI tool integrations, pushing of new features, changes to the UI.

Notion

Too many choices and distractions, similar to Google that way. It doesn’t work well offline, so I can’t use it while on a plane or when I intentionally disconnect. Maybe related to its always-online characteristic, but I find it somewhat slow and not as responsive as Obsidian.

Claude

Too much output that is difficult to review by humans, since it will outpace us in terms of quantity many times over.

Handwritten notes

I don’t have a system for absorbing these thoughts into digital form, so it’s hard to query or review what I wrote.

Git / GitHub

Most scientists (whom I work a lot with) don’t use these tools. They definitely don’t use it well. Even just having their code / work on a repo at all is miles ahead of most scientists. They definitely are not using it for project management and active development - more so a place just to store and document code. I digress, this does mean though that it is significant friction to get collaborators on board with using version control like Git.

Positron IDE

Definitely buggier than RStudio, which makes sense because it’s in beta still, but does mean it’s frustrating to work with sometimes with random problems.

Zotero

I haven’t really cracked how to integrate Zotero references into Obsidian well. I’ve tried before during my Master’s degree for literature review, but it was quite janky and too much friction to use consistently. As a result, I’m using it standalone separate from my Obsidian notes.

In general

I have a ton of half-written and incomplete notes all over the place that are difficult to make useful. These are especially true from work / notes from before AI times.

Designing new workflow

Workflow needs to do the following:

1. Reduce friction for writing, thinking, creating, as much as possible. As few distractions as possible.
   1. One goal for this workflow would be to increase my output to document and share my work on my own blog, as well as platforms like LinkedIn.
2. Track and backup references for literature, books, videos, etc. for where I learnt each idea or knowledge “piece”. Traceability for information essentially.
3. Integrate AI tools with clear constraints, and be able to switch between AI model providers (local, cloud).
   1. I want AI for proof-reading (like an advanced spellchecker), high-level search (not for new claims, more like generally accepted knowledge for a given field), summarizing specific pieces of work (ex. scientific literature).
   2. I also want AI for development work, that can read from my notes to inform development decisions and help build a body of knowledge over time that I can review.
4. Sync across my devices (2 laptops, 1 smartphone) seamlessly.
   1. Back up so losing devices is not catastrophic.
   2. Be able to work in transit, in small breaks, squeeze out work in small chunks.
5. Track progress on work across projects
   1. Reduce duplication of effort for tracking tasks. Writing about the task should only happen once, writing updates to the task only once, reporting on the completed task only once, etc.
   2. Single source of truth. Easy to go back to working after a long break, looping in collaborators, handing off to clients.
   3. Enable AI to help with admin work, task management, hours tracking, etc.
6. Track contacts / collaborators / clients / people
   1. Clear record of contacts, so I can track how I’ve met and worked with various collaborators over time.
   2. Identify pending work, be able to revisit contacts to keep projects moving or clearly finish projects.
   3. Clear system for working with connections (potential contact, had contact, active discussions, current collaboration, completed collaboration, etc.)
   4. Use AI to look through emails to identify loose ends.

Choice of tools

Obsidian

I really enjoy how it feels to work in Obsidian. The uncluttered and simple workspace. No random popups asking you to use some new AI tool or to upgrade your plan. Much easier to focus on writing. Obsidian Sync makes it easy to keep in sync with my mobile too, so I can write on-the-go and add photos from my phone. The plugins extend the functionality a lot, so I can add all kinds of features. This also means that if there were a feature I truly needed but was not available, I can develop my own plugins. This has become a lot easier too with the improvement in AI tools. The markdown file format used by Obsidian to save content makes it much easier to integrate with a variety of AI tools, compared to being locked-in on one.

With that rationale, I’ll go with using Obsidian for the most part for holding the majority of the content / notes / diagrams / etc. for capturing my general knowledge and thoughts over time.

Obsidian also has encrypted sync, so I can feel more secure about holding a range of data with them.

Handwritten Notes

Essential part of my workflow. This should be a tool for organic thought, so I would benefit from creating space for writing notes as a clear component of the system. In addition, some process for rewriting notes into Obsidian from my handwritten notes would be good. Especially for knowledge / notes from conferences, presentations, workshops.

Git / GitHub

I’ve found Git / GitHub ideal for specific projects, that have defined scope, rather than general knowledge. Even with the challenges of getting scientist collaborators on board with using it fully, it’s worth the effort as it unlocks so much in terms of improving code quality, tracking changes, understanding development history, and more. I’ll strive to be more disciplined about writing issues and using PRs for my work. It’s easy to fall out of habit and just commit onto main when working alone, but it’s much more understandable for future collaborators or handing off projects when I use Git properly.

I would like to integrate GitHub and Obsidian somehow, especially being able to track and write Issues in Obsidian, rather than the GitHub website.

Positron IDE

Main development environment for me. There is a lot I could say about it, especially in contrast to using RStudio (the dominant IDE for R programming). It really does open up the world in terms of what is possible with programming, compared to using RStudio. One challenge with this is that the IDE is not as mature as RStudio is, but that is rapidly changing. Great integration with AI, enables remote access, lots of plugins. I could also create, edit, view markdown well from here, so more seamless to transfer information back and forth with Obsidian, especially compared to using something like Notion.

Claude

Very powerful AI tool, with lots of user friendly features. It’s super popular, so easy to find workflows and tutorials from other people to integrate into whatever tools. I use it all the time within Positron IDE for powering my development work. It seriously accelerates my development speed to a scary extent. It reads in markdown files the best, so that integrates well across Obsidian / Positron IDE / GitHub. With working in Git repositories within Positron IDE (or any other IDE or terminal for that matter), Claude can read all of the files and use as context for development, making it much better than the entry-level way of using AI through some chat interface through a website. Similar with Obsidian Vaults for notes.

Claude has integrations with like everything, and it’s pretty good with working with messy data. I can use it as the glue across different tools. A lot of the time sink for using a new tool is in the learning curve, so using Claude with MCP (Model Context Protocol) would accelerate which tools I can incorporate into my workflow, particularly if I just need it for some one-off situation.

General Considerations

I have failed at fully integrating Obsidian into my workflow in the past, with my last serious attempt being during my Master’s degree. Since then, Obsidian has improved a lot with more plugins and wider usage. Another game changer has been how much better AI is becoming. I don’t want to go as far as outsourcing my thinking and writing fully to AI, but it would help A LOT with automating the links and structuring information. These are the crunchy bits of using Obsidian and the “Second Brain” system in general that I found to be too much friction for my earlier attempts. I’m hoping that with this attempt, with more mature Obsidian, AI, and also my own knowledge, that I can be successful at building a true “Second Brain”.

Obsidian and Claude

This integration between these two tools in particular is key to my planned setup to work well. I read this blog post https://blog.starmorph.com/blog/obsidian-claude-code-integration-guide for some guidance on how to do this. Out of the multiple strategies that are outlined, I think “Strategy 2: Vault IS the Claude Code Working Directory” seems to be the best for what I want to do.

I really considered “Strategy 3: MCP Bridge”, but I haven’t done anything with MCP servers before. For the sake of getting up and running quicker, I’ll do Strategy 2 and maybe do Strategy 3 down the line.

I am doing a lot of development work too, so Strategy 1: Dedicated Developer Vault with Symlinks does seem very useful as well. Maybe I’ll combine these strategies later and gradually with new projects and see how it goes, rather than try to integrate all my existing projects at once.

Conclusion: Start with Strategy 2.

Implementation Tasks

1. Writing this piece in my existing Base Obsidian setup, to exhaust my thoughts independent of AI tools as a starting point.
2. Export notes from Notion and Google Docs, to archive into Obsidian Vault.
3. Access this Obsidian Vault from Claude via terminal, use vibe-coding to create essentially a CRM for my clients and collaborators, as well as a project management system. Specifically, I’ll use Claude to create templates for inputting new contacts, new organizations, track communications with each contact from meeting notes, link people to projects. With the Obsidian graph features, I can create connections across all of these components with AI to assist with automating this step with manual review.
4. Figure out some GitHub integration into Obsidian, so that I can see the pending issues and PRs for each project that I am working with. I should also be able to write and reference GitHub issues from Obsidian. This would be especially useful for recurring bugs or features that I implement across different projects. Another reason for this would be the ability to track which issues came from a particular meeting or discussion, and track that task from meeting to PR.

Closing Thoughts

This was a good exercise in reflecting on what has worked and what has been a growing pain point in organizing my knowledge and outputs. If I didn’t do this now, I definitely would not have done it when I was neck deep in five different projects with deadlines to meet.

The intention for this dump of thinking at this stage was to lay a solid foundation for future work. I want to increase my throughput and efficiency, to meet the growing demand for my output.

I want to be able to help more people and do more towards building sustainable futures, whether doing shark and ray research or creating biodiversity plans for communities or building machine learning algorithms for seafloor habitat mapping. I hope this structure will help me a lot with doing more, and documenting this will help others in their work, especially in the ecology and environment fields.

This has been quite a ramble, so thanks for reading to this point! I’ll update again once I’ve implemented and worked with this system.