brain.bfurg.com

brian furgason

ux, ai-native systems, deep-curious learner

captures
255
topics
491
days
90
range
mode
255 captures
01 / breadth

what does this mind care about?

ux auditdesignaiuxmvppgcronob1datavoicerulesuxauditci/cdtokenlayerdenocasestudypgnetuserdemoaiworkfigmatastetoolsbasesizeiconscodefixtddutf-8perlyagnicssuiagentlooptechdebtmetabrainactormodel
02 / focus

where has the last 90 days gone?

  • buildfeb 2026 — ongoing

    brain — personal memory + identity system

    v0 backend shipped

    OB1 + Supabase + MCP stack for capture and recall across surfaces. Two-tier auth (private/worksafe). 11 MCP tools live on bfurg.com/mcp.

    • ChatGPT export + 70K YouTube history backfilled
    • telegram bot for mobile capture
    • vaporwave-helical dashboard now in flight (this project)
  • buildmar — apr 2026

    helical.design — token system rebuild

    live

    Rebuilt the agency site as a fully tokenized stylesheet. WCAG ratios documented inline, dark mode via token override, no hardcoded values.

    • single-file html with inline css-vars
    • every contrast ratio verified against actual bg
    • design language now portable — see this dashboard
  • learnongoing

    claude api — caching, tool use, mcp

    applied in brain

    Prompt caching as an architectural concern, not an optimization. MCP as the contract between memory and any LLM session.

    • cache hit-rate matters more than tokens-per-call
    • tool definitions are interfaces — design them like apis
    • agents-as-roles beats agents-as-prompts
  • learnapr 2026

    d3 + react — buttery interactive viz

    shipping in radar + pulse

    d3-hierarchy.pack for cluster sizing, d3-zoom for pan, framer-motion for spring physics. Animate transform — never cx/cy.

    • transform-only animation = gpu-accelerated
    • pre-compute layouts, animate to cached positions
    • visx vs d3-direct is an authoring-style choice, not a perf one
  • learnmar 2026

    refactoring ui — wathan + schoger

    applied throughout brain dashboard

    Hierarchy via size + weight + color (not size alone). Color as meaning, never decoration. Whitespace is structural.

    • labels lighter and smaller; values bold and large
    • depth via contrast and 1px lines, not shadows
    • icons earn their place or get cut
  • buildapr 2026

    next.js 16 — cache components + app router

    stack of choice

    PPR + use cache + cacheLife as a new mental model. Server components by default; client components surgically applied.

    • tokens in globals.css, components in css modules
    • next/font for self-hosted geist + helvetica fallback
    • turbopack dev under 400ms ready
03 / rhythm

how consistent is the practice?

04 / momentum

what's live right now?

  • direct5d ago

    planning

    tagged planning · agentic layer · self-improvement

  • direct5d ago

    verification layers

    tagged verification layers · infrastructure builds · bug detection

  • direct6d ago

    health-check

    tagged health-check · detection layer · separation of concerns

  • direct6d ago

    build pipeline

    tagged build pipeline · CI/CD · testing

  • direct6d ago

    fluid design

    tagged fluid design · viewport management · code performance

  • direct6d ago

    Before building a discipline-enforcing skill, run the RED baseline first: put a…

    skill-authoring

  • direct6d ago

    Designing a health/diagnostic instrument surfaces a cluster of reusable gotchas…

    observability

05 / synthesis

how do these ideas connect?

  • Track B of the brain IP gate: how much work is it actually… ↗ supports What do I actually need to do to charge for brain, given it…

    • strategy

    Scopes Track B (de-vendor) of the IP gate: ~2-4 days, small isolated footprint.

  • Wanted a portable, privacy-tiered AI memory layer that surv… → evolved into Yesterday's session shipped the Attention layer with 70K ro…

    • engineering

    A designed and shipped the foundational privacy-scoped schema with thinking_records table; B extends that same system by adding cross-layer query logic (correlate_attention_with_thinking) that reuses A's substrate and visibility filters without schema changes, advancing the architecture day-over-day.

  • What's the actual decision rule for when to take the templa… ↗ supports What's the single operating instinct underneath my judgment…

    • design
    • personal

    The design hardness gate is one expression of the unifying substance-over-surface instinct.

  • What does it feel like — and what's the one-time vs. compou… ↗ supports What's the single operating instinct underneath my judgment…

    • strategy
    • personal

    The daily practice is the mechanism that builds and sharpens the self-model over time.

  • What is the Health pillar of brain, and what's broken about… ≈ relates to Where does brain's first dollar come from, and what's the p…

    • strategy

    The Health pillar is one of the three life pillars in the product architecture, with brain as the nervous system aggregating it.

  • What's the measurable outcome for the relationship/emotiona… ≈ relates to How should the relationship package help honestly — without…

    • strategy
    • design

    This metric (unblocking, not relationship survival) is what makes the "it can say leave, not just stay" honesty structurally safe.

  • What is brain's core value to an AI experience — the wedge… ↗ supports What's the single operating instinct underneath my judgment…

    • strategy
    • personal

    Grounding the AI in the user's essence so it can say no instead of a sycophantic yes is substance-over-surface as the product wedge.

  • Package the ethics stance, the instrumentation design, and… ≈ relates to How do I build checks-and-balances / instrumentation that c…

    • strategy
    • design

    Consolidates the instrumentation design into the unified honesty-as-substance finding (transparency as guardrail).

  • What tools already do the relationship / personal-memory th… ≈ relates to Is there a second use case for brain's schema-based memory…

    • strategy

    Market sizing and competitive landscape for the relationship/emotional-cognition second use case.

  • For the thought-externalization / reflection tier aimed at… ≈ relates to Is there a second use case for brain's schema-based memory…

    • design
    • strategy

    These help/hurt heuristics are the safety design layer for the second use case (emotional-cognition + relationship grounding).

  • What's my legacy, and how do I rank what I built vs. who I… ↗ supports What's the single operating instinct underneath my judgment…

    • personal

    Legacy through who I helped and what I stood for — values as the thing that endures — is the substance-over-surface self-model at the level of a whole life.

  • What's overhyped enough to bet against vs. quietly underrat… ↗ supports What's the single operating instinct underneath my judgment…

    • technology
    • personal

    Betting against surface tools and long on the data/backend substrate is the same instinct aimed at technology.

showing 12 strongest threads · 14 more edges on these records

06 / resonance

which observations have lodged in the reasoning?

  • 60imp 5observation

    OB1 vendoring (three phases, all merged today) closed the loop on a meta-question: how do you compose a personal system on top of an open-source framework without ending up with two fighting models o…

    • open-source
    • personal system
    • recommendation
  • 35imp 4observation

    Phase 1.5 enrichment is now inline — migration 0017 extends the metadata-sync trigger to handle importance, and the OpenRouter extractMetadata prompt (now in supabase/functions/_shared/openrouter.ts,…

    • metadata enrichment
    • tech debt
    • OpenRouter
  • 35imp 4observation

    Telegram video capture (Phase 3) shipped as a thumbnail+audio composition rather than multi-frame sampling. The plan assumed sampling N frames + describing each, which would need ffmpeg in the Deno e…

    • media processing
    • serverless
    • integration
  • 35imp 4observation

    Closed out the Notion agent sweep. Patina (mask spec agent, fabrication pipeline App 2) landed as work_pattern external_source notion-agent-spec, external_id patina-v1. SF UX Orchestrator case study…

    • Notion
    • agents
    • UX
07 / projects

what's been built, shipped, or maintained?

202420252026now
  • helical

    • agency
    • design
    • ongoing
    sep 2023ongoing

    ux design agency for ai-native software teams. partner-tier delivery, in-house token systems, agent-augmented workflow.

  • yukon camper conversion

    • vehicle
    • real-world
    • shipped
    apr 2024aug 2025

    ground-up overland conversion. solar, lithium, full insulated build-out. multi-year fabrication + electrical learning curve.

  • helical.design

    • site
    • design
    • shipped
    aug 2025jan 2026

    fully tokenized single-file html. wcag ratios documented inline. design language portable to any consumer — including this dashboard.

  • helical agent specs (patina, sf-ux-orchestrator, nova, forge)

    • engagement
    • ai
    • ongoing
    nov 2025ongoing

    agent-as-role spec system for helical's content + delivery pipelines. multi-agent roster with tool, capacity, and stage coverage defined per role.

  • brain — personal memory system

    • system
    • systems
    • ongoing
    feb 2026ongoing

    ob1 + supabase + mcp. four layers (capture, attention, identity, thinking) with privacy parity. live at brain.bfurg.com/mcp with 11 tier-aware tools.

  • brain-dashboard

    • site
    • code
    • ongoing
    apr 2026ongoing

    this. vaporwave-helical fusion portfolio at dashboard.bfurg.com. d3-direct visualizations, framer motion, radix primitives, plain css modules.

08 / meta

how was this made?

  1. 01

    capture

    OB1 + Telegram bot + ChatGPT export + YouTube history. Multi-surface, single sink.

    • OB1
    • Telegram Bot API
    • Supabase Edge Functions
  2. 02

    store

    Postgres + pgvector. Typed tables (work_patterns, lessons_learned) over snapshots when the shape fits.

    • Supabase Pro
    • Postgres
    • pgvector
  3. 03

    recall

    MCP tools, two-tier auth (PRIVATE / WORKSAFE). Thoughts callable from any LLM session.

    • MCP
    • brain.bfurg.com/mcp
  4. 04

    render

    This dashboard. D3 + Framer Motion + Radix on Next.js 16. Tokens ported from helical.design.

    • Next.js 16
    • React 19
    • D3
    • Framer Motion
    • Radix
09 / horizon

what does the system know it doesn't know?

twenty named blind spots — the questions this cognitive system can't answer about itself from inside the loop. solid border = a written position; dashed = open. the point isn't to answer them all, it's to keep them named so they don't silently drift past.

  • position taken
  • engaged · undefined
  • open question
epistemicis the system doing what I think it's doing, or just producing outputs I accept?
  • 01

    survivorship bias in lessons

    open · 1 capture

  • 02

    whose mental models are these?

    open · 1 capture

  • 03

    outsider read

    open

  • 04

    confirmation bias in retrieval

    open · 4 captures

  • 05

    frequency illusion

    open

  • 06

    recency dominance

    open

  • 07

    medium-shaped thought

    open · 1 capture

  • 08

    decisions vs. rationalizations

    open

identity / valueswhat is the system for, beyond utility?
  • 09

    disagreement

    position · 2026-05-07 · 1 capture

  • 10

    refusal

    engaged · needs definition

  • 11

    death and forgetting

    position · 2026-05-07

  • 12

    access

    position · 2026-05-07

  • 13

    five-year success metric

    position · 2026-05-07

  • 14

    proactive pushback

    position · 2026-05-07

second-orderwhat is changing about me because the system exists?
  • 15

    cognitive offloading

    open

  • 16

    performance vs. capture quality

    open · 1 capture

  • 17

    framing for retrieval

    open · 1 capture

  • 18

    the cost of capturable thinking

    open

  • 19

    closure-bias

    open

  • 20

    identity drift

    open