When I started this series, most people in my network had never heard of Claude! 🤯
Thirteen posts later, I’ve watched colleagues install Claude Code, build CLAUDE.md files for their projects, set up MCP servers to connect to PubMed, fork my custom Skills for their own workflows, and start actually enjoying version control! That’s the best outcome I could have asked for! 🧬
This final post is a retrospective - what we covered, what I learned writing it, and where I think all of this is heading! 🚀
The series at a glance 🗺️ Link to heading
Here’s the complete roadmap, organized by the progression that made the most sense for researchers coming in cold:
Understanding Claude (Posts 1-2)
The series started with the basics because I realized most people in bioinformatics and computational biology hadn’t even tried Claude yet! Post 1 introduced Anthropic’s approach - Constitutional AI, the 200K token context window, and why this matters for technical work! Post 2 put Claude head-to-head with ChatGPT and Gemini specifically for coding and scientific reasoning tasks! The takeaway wasn’t “Claude is always better” but rather “Claude excels at exactly the things researchers need most: long-context understanding, instruction following, and honest uncertainty!” 🎯
Claude Code fundamentals (Posts 3-4)
This is where the series shifted from “interesting AI tool” to “this changes how I work!” Claude Code isn’t a chat interface - it’s an agentic coding partner that lives in your terminal, reads your files, writes code, and runs commands! Post 3 introduced the concept, and Post 4 showed real workflows: debugging pipelines, refactoring messy scripts, generating documentation, and building entire analysis workflows through conversation! The key insight? Claude Code doesn’t do autocomplete - it understands your entire codebase! 🔬
Project context (Posts 5-6)
Posts 5 and 6 were about making Claude Code smarter for YOUR specific work! CLAUDE.md (Post 5) is a markdown file in your project root that gives Claude persistent context - your project structure, coding conventions, preferred packages, common pitfalls! Think of it as onboarding documentation for your AI collaborator! 📋
Skills (Post 6) took this further - reusable instruction packages that teach Claude HOW to do specific tasks! The magic is auto-detection: Claude reads your SKILL.md files and picks the right expertise without you asking! Together, CLAUDE.md and Skills transform Claude from a generic assistant into a domain specialist! 🧠
External connectivity (Posts 7-8)
This pair was about breaking Claude out of the terminal! MCPs - Model Context Protocol (Post 7) - are the universal standard for connecting Claude to external tools! PubMed, GitHub, Slack, Google Drive, databases - one protocol, endless connections! I described it as USB-C for AI! 🔌
Subagents (Post 8) flipped the script entirely - instead of connecting outward, Claude learns to spawn copies of ITSELF! Each subagent gets its own context window and focus! One reads your codebase while another writes tests while another reviews code - all in parallel! For bioinformatics, this means a QC agent checking data quality while a methods agent writes your pipeline while a documentation agent captures everything! ⚡
The ecosystem (Posts 9-10)
Post 9 revealed the plugin ecosystem - 9,000+ community-built extensions that bundle skills, agents, hooks, and MCP servers into one-command installs! This is where Claude Code stopped being a solo tool and became a platform! 📦
Post 10 spotlighted the single most impactful plugin: Superpowers by obra! With 40,000+ GitHub stars, Superpowers enforces a complete software development methodology - brainstorming before coding, structured planning, test-driven development, continuous code review, and clean git workflows! It transforms Claude from a code generator into a senior developer who REFUSES to skip steps! 🦸
Making it personal (Posts 11-13)
The last three posts shifted from “what exists” to “how I actually use it!” Post 11 broke the tutorial format with an honest opinion piece about how Claude Code’s git and gh integration made me genuinely love version control - something I’d been doing poorly for years despite knowing better! That post resonated differently because it was authentic, not a feature walkthrough! 🔥
Post 12 was the natural evolution of Post 6 (Skills) - instead of explaining what Skills ARE, I showed what happens when you actually build your own! Five custom bioinformatics Skills covering R coding conventions, project scaffolding, Quarto scientific notebooks, lab documentation, and R package development - all interconnected, all with BadranSeq as the default visualization layer, and all open-sourced for the community to fork! 🛠️
Post 13 tackled the most frustrating limitation of Claude Code: session amnesia! Every new session starts from zero, and for researchers iteratively tuning analysis pipelines over days or weeks, that’s a real productivity killer! claude-mem fixes this with automatic context capture, AI compression, and token-efficient progressive disclosure search - all running passively in the background! 🧠
What I learned writing this series 💡 Link to heading
Three things stand out:
The AI isn’t the hard part anymore!
The most valuable thing I could give my audience wasn’t “here’s how to prompt Claude better.” It was “here’s what tools exist and when to reach for each one.” The ecosystem moves so fast that most researchers don’t know half of what’s available! Discovery is the bottleneck, not capability! 🎯
Bridging audiences is the real challenge!
Every post had to work for both computational biologists who live in the terminal and experimental researchers who are terminal-curious! The analogies that landed best were the ones rooted in lab culture - PI and postdocs for subagents, USB-C for MCPs, app stores for plugins! Technical accuracy without jargon walls is hard, but it’s the only way to actually expand who uses these tools! 🧬
Opinion pieces and personal builds hit different!
Posts 11, 12, and 13 got noticeably different engagement than the tutorials! People connected with the honesty about bad version control habits, the transparency of open-sourcing my own Skills, and the practical problem of session amnesia more than they connected with feature lists! There’s a clear lesson about mixing authentic editorial content and real-world implementations into technical series! 🤝
The complete toolkit - quick reference ⚡ Link to heading
For anyone who wants the one-paragraph version of the entire series:
Install Claude Code in your terminal! Create a CLAUDE.md in your project root describing your codebase and conventions! Add Skills for domain-specific tasks you repeat (or fork mine from GitHub)! Set up MCPs to connect Claude to PubMed, GitHub, and your databases! Create custom Subagents for specialized parallel work! Install Plugins from the marketplace for community-built functionality (start with Superpowers and claude-mem)! Use git and gh integration for painless version control! And write a CLAUDE.md for every new project from day one - your future self will thank you! 🚀
Where this is all heading 🔮 Link to heading
I started this series when Claude Code was still relatively unknown in the research community! In the months since, I’ve watched the ecosystem explode - new plugins weekly, MCP servers for everything from NCBI databases to institutional LIMS, community-built skills for domain after domain! 📈
The trajectory is clear: AI-assisted research isn’t a novelty anymore! It’s infrastructure! The researchers who invest time now in building their Claude Code workflows - CLAUDE.md files, custom skills, domain-specific agents - are building compound advantages that will accelerate everything they do going forward! 💪
This wraps the Claude series, but the conversation doesn’t stop! New tools, new integrations, new series - there’s always more to explore! 🧬
Thank you to every single person who followed along, shared posts, asked questions in the comments, and tried these tools in their own labs and pipelines! You made this series worth writing! 🙏
The full series archive lives at badran-elshenawy.netlify.app 🌐