From Flashlight Beams to Persistent Context: A Docs-First Approach to Human-AI Collaboration
How to stop rebuilding context from scratch every session and start building something that compounds. Transform scattered AI collaboration into systematic partnership.
import { Image } from βastro:assetsβ; import flashlightBeamProblem from β../../assets/images/collaborations/docs-first-approach/flashlight-beam-problem.jpgβ; import collaborativeWorkspace from β../../assets/images/collaborations/docs-first-approach/collaborative-workspace.jpgβ; import organizedWorkspace from β../../assets/images/collaborations/docs-first-approach/organized-workspace.jpgβ;
From Flashlight Beams to Persistent Context: A Docs-First Approach to Human-AI Collaboration
The Flashlight Beam Problem
Picture this: Youβre collaborating with an AI assistant on a project. Youβre showing it bits and piecesβa code snippet here, a design idea there, some half-formed requirements scattered across messages.
Itβs like navigating a dark cave with a tiny flashlight beam, illuminating just small patches while the bigger picture stays hidden in the shadows.
The AI builds up context as you go, getting excited about your ideas, understanding your goals, helping solve problems. Youβre both in flow, building momentum, making real progressβ¦
Then the session ends. π₯
Next time you chat, itβs like meeting someone with amnesia. Youβre back to square one, re-explaining the project, rebuilding that hard-won context, watching the AI rediscover insights youβd already worked through together.
Both of you end up more scattered than when you startedβa perfect recipe for digital ADD amplification.
π§ Click to reveal the brutal reality of context amnesia...
Session 1: βI want to build a time-tracking app with some interesting featuresβ¦β
[30 minutes of exploration, AI builds understanding, suggests cool ideas]
Session 2 (two weeks later): βHey, remember that time-tracking thing we discussed?β
[AI has no memory, you spend 20 minutes re-explaining everything]
Session 3: βLet me just paste some of the code we talked aboutβ¦β
[More context rebuilding, less actual progress]
Session 4: βActually, let me just figure this out myself.β
[Collaboration abandoned]
The pattern is brutal: Every session starts from zero. All that collaborative momentum gets lost. You end up treating AI like a stateless tool instead of a creative partner.
Sound familiar? Welcome to the most frustrating part of AI collaboration in 2025.
When Context Dies, Creativity Dies With It
π Average project restarts: 3-4 sessions
β±οΈ Context rebuilding time: 20-40 minutes per session
π Momentum retention: ~15% after 1 week
π€ Frustration level: Maximum
π Project completion rate: You donβt want to know
Hereβs what typically happens in scattered AI collaboration:
π’ Session 1: The Honeymoon
"This is amazing! The AI totally gets what I'm trying to build. We're making real progress!"
π‘ Session 2: The Reality Check
"Wait, it doesn't remember anything? Let me re-explain... okay, we're back on track."
π Session 3: The Struggle
"I'm spending more time explaining than building. This is getting frustrating."
π΄ Session 4: The Abandonment
"I'll just do this myself. It's faster than constantly re-teaching the AI."
But what if we flipped the script entirely?
The Docs-First Discovery
I stumbled into a better approach while working on what became The Living Artifact Checklist. Instead of diving straight into writing, I found myself naturally documenting the methodology firstβthe story arc, key arguments, examples, reader journey.
It felt weird at first. Why spend time βplanning to writeβ when I could justβ¦ write?
But then something clicked: The planning wasnβt separate from the writingβit was collaborative thinking made persistent.
Instead of my usual scattered approach (start writing, realize Iβm missing something, restart, lose momentum), the docs created a foundation that both human and AI could build on across multiple sessions.
The breakthrough moment came when I realized: What if every project started with creating its own βuser manualβ for collaboration?
The Methodology: Vibe-Write Your Way to Persistent Context
Instead of jumping straight into implementation, we create the documentation firstβbut not boring technical docs. Weβre talking about collaborative vibe-writing that captures the vision, energy, and decision-making context in a way that survives between sessions.
Hereβs how docs-first collaboration actually works:
Step 0: Context Dump π₯
The Magic Phrase
This immediately signals that you want to collaborate on structure before implementation. The AI knows to help you document the vision rather than jump straight into building.
Diataxis is a documentation framework with four types: Tutorials (learning-oriented), How-to guides (problem-oriented), Reference (information-oriented), and Explanation (understanding-oriented). For our purposes, it just means "think about different types of documentation needs" rather than dumping everything in one place.
Step 1: Vibe-Document the Vision π
Instead of jumping into code or content, collaboratively write the βdocsβ for what youβre building:
π¬ Story Arc
What journey are you taking people on? From problem recognition to solution implementation.
π‘ Key Arguments
What core ideas are you communicating? The foundational insights that drive everything else.
π Examples
Which stories, use cases, or scenarios illustrate your points? Concrete beats abstract.
πΊοΈ User Journey
How do people go from problem to solution? What's the ideal path through your creation?
π¨ Voice & Structure
What's the vibe and logical flow? How does it feel to interact with what you're building?
Step 2: Let Implementation Flow π
With comprehensive docs as your foundation, the actual code/content/design flows naturally. Youβre implementing against a clear spec rather than figuring it out as you go.
No more βwhat was I trying to build again?β moments.
Step 3: Persistent Collaboration π
Just point the AI at your docs. Instant context restoration. No rebuilding, no re-explainingβjust continuing where you left off.
Itβs like having a project memory that survives between collaborations.
Real Example: This Very Post (Meta Alert! π¨)
π― Context Dump
Started with the "flashlight beam" observation and walked through the problem space together.
π Vibe-Documentation
Spent time mapping out the story arc (problem recognition β methodology β examples β practical framework), identifying key arguments, and planning the reader journey.
π Reference Integration
Used the Living Artifact Checklist post as concrete proof the methodology works in practice.
β¨ Meta Moment
Realized we were demonstrating the exact approach we were documenting. The AI didn't have to rebuild understandingβit was all there in our collaborative docs.
Why This Changes Everything
π For Complex Projects
βοΈ For Creative Work
π For Learning Projects
The Different Flavors
π» Software Projects
Write API docs and user flows before writing code. Define the interface before building the implementation.
π Content Creation
Document story structure and key messages before writing. Create the skeleton before adding the meat.
π¬ Research Projects
Create research frameworks before diving into sources. Know what you're looking for before you start looking.
π Organizational Tasks
Build systematic checklists before executing (like the Living Artifact approach).
Implementation Strategy
Create Your Project Artifact π
Every docs-first project needs its own living checklist. Hereβs a battle-tested template:
# [Project Name] - Docs-First Collaboration
> **Current Phase:** [Where you are now]
> **Last Session:** [Date and what you accomplished]
> **Next:** [Immediate next step]
> **Energy Level:** [High/Medium/Low - track momentum]
## Project Vision
**Goal:** [What you're building and why]
**Timeline:** [Rough estimate]
**Collaborators:** [You + AI partner]
**Success Looks Like:** [Clear completion criteria]
## Documentation Status
β Story arc / user journey mapped
π Key features / arguments identified (need user research)
β Examples and references gathered
β Structure and flow planned
β Voice and tone decided
## Implementation Phases
β Phase 1: [First major milestone]
π Phase 2: [Second milestone - needs technical investigation]
β Phase 3: [Third milestone]
## Context Preservation
**Key Decisions:** [Important choices made and why]
**Lessons Learned:** [Insights and course corrections]
**Next Session Prep:** [What to review before continuing]
## Resources & References
[Links, examples, inspiration, tools needed]
## Status Legend
β Not started
π Needs investigation/research
β
Complete
Signal Intent Clearly π―
Magic Phrases That Work:
- "Docs-first approach" - Immediately tells AI you want structure before implementation
- "Let's vibe-write this into existence" - Signals collaborative, conversational documentation
- "Using Diataxis style" - References the documentation framework (tutorials, how-tos, reference, explanation)
- "Living artifact checklist" - Indicates you want a resumable, evolving document
Start Small, Build Momentum π
DONβT: Create rigid, formal documentation that kills creativity
DO: Use conversational, living documents that capture energy and decision-making context
DONβT: Abandon the approach if the first attempt isnβt perfect
DO: Iterate and improve - your documentation gets better as you learn what kind of context actually helps
Docs should take 10-20% of your total project time. For a 4-hour project, spend 30-45 minutes on docs. For a weekend project, maybe 2-3 hours of planning.
The documentation time pays for itself by preventing context rebuilding and scope creep.
β οΈ Common Pitfalls (And How to Avoid Them)
π₯ Over-documenting
π€· Under-documenting
β¨ Perfectionism
π Ignoring the artifact
When NOT to Use Docs-First
β Perfect for docs-first:
- Multi-session projects - Anything that spans multiple days/weeks
- Collaborative work - When others (AI or human) need to understand the vision
- Complex or novel tasks - Projects where youβre figuring things out as you go
- Important outcomes - When failure would be costly or frustrating
The Meta-Layer: When AI Helps Document AI Collaboration
Advanced Techniques for Power Users
π Version Control Your Collaboration
Use git to track every change in your docs-first projects. This transforms collaboration from a scary, irreversible process into a documented, experimental journey where mistakes are just interesting detours.
git commit -m "Phase 2 complete: User journey mapped with 3 key decision points
AI suggested focusing on mobile-first flow after analyzing user research.
Human added emotional journey considerations for signup experience.
Next: Build wireframes based on documented user flow."
π― Project-Specific Templates
Create reusable documentation templates for your common project types:
- Web Development: API docs + user flows + component hierarchy
- Content Creation: Story arc + key messages + target audience + distribution plan
- Research Projects: Question framework + source strategy + synthesis methodology
- Learning Goals: Concept map + practice exercises + progress tracking
π€ Multi-Session Context Bridging
Develop rituals for session transitions:
Session End Ritual:
- Update project status and next steps
- Document key decisions made and why
- Note any unresolved questions or blockers
- Set up context for next session
Session Start Ritual:
- Review last sessionβs progress and decisions
- Confirm current phase and immediate goals
- Check for any context updates needed
- Dive into work with full context restored
From Scattered to Systematic
The docs-first approach transforms AI collaboration from:
π΅βπ« Scattered
Each session starts from zero, energy gets dissipated across repeated context building
β‘ Systematic
Context compounds across sessions, energy builds momentum rather than starting over
π² Reactive
Figuring out what you're building while you build it, leading to scope creep and confusion
π― Intentional
Clear vision guides implementation decisions, reducing wasted effort and rework
π₯ Fragile
Lost momentum kills projects when life interrupts or motivation wanes
πͺ Resilient
Projects survive interruptions and pick up where they left off, preserving progress
π§ Individual
You carry all context in your head, creating single points of failure
π€ Collaborative
Shared artifacts create true partnership with persistent shared understanding
π― Try It Right Now (5-Minute Test)
Pick the smallest project on your mind right now - maybe organizing one folder, writing one blog post, or building one small feature.
Set a timer for 5 minutes and write its βdocsβ using our template above. Donβt overthink it. Just capture:
- What youβre trying to accomplish
- Why it matters
- What success looks like
- The basic steps to get there
Then notice how much clearer the actual work feels when you start.
Thatβs the docs-first effect in microcosm. β‘
If it works, scale it up to bigger projects. If it doesnβt, you only lost 5 minutes figuring out itβs not for you.
What happens after the test? If the 5-minute docs made your work clearer, try the full template on your next multi-session project. Start tracking progress, update the artifact as you learn, and see how much easier it becomes to resume work after interruptions.
π¨ Pro tip: Notice how this post uses visual elements, interactive sections, and rich formatting? Thatβs intentional - docs-first artifacts work better when theyβre engaging and easy to scan. Plain text works fine, but a little visual organization helps both humans and AI parse the information more effectively.
The Bottom Line
Weβve all experienced that frustrating cycle: exciting AI collaboration that fizzles out because context doesnβt persist. The flashlight beam problem isnβt just annoyingβit actively sabotages creative partnerships.
Docs-first collaboration solves this by frontloading the exploratory energy into something persistent. Instead of scattering creative insights across forgotten sessions, youβre building a foundation that both human and AI can return to and build upon.
Your next project doesnβt have to start from zero.
Try spending 20 minutes documenting what you want to build before building it. You might be surprised how much clearer the path becomes when both collaborators can see the map.
Pick your next project and start with: βWassup, letβs take a docs-first approach to thisβ¦β
Then watch the magic happen. β¨
Ready to escape the flashlight beam problem forever?