Project Doc

Learnings

What worked, what didn't, why

Updated: December 2025 Source: LEARNINGS.md

Learnings

What worked. What didn't. Why.

This is the most important document. Update it as you go. This becomes the source material for the eventual doctrine.

Format

## [Date] - [Topic]

**Context:** What were you trying to do?

**What happened:** What actually occurred?

**Learning:** What does this mean for Forge?

**Action:** What changes as a result?

December 2025 - Research Phase

Context: Explored whether to build a new "AI-native stack" with Go/Elixir, AG-UI/A2UI, EdgeDB/Turso, etc.

What happened:

  • Every alternative has critical gaps (see agentic-web-standard research)
  • The thesis (React is wrong for agentic coding) appears valid
  • But alternatives aren't mature enough
  • Realized the opportunity isn't a framework, it's a workflow

Learning: Don't build novel infrastructure. Build opinions on top of proven tools. The value is in the workflow, not the runtime.

Action: Using existing tools (Next.js, Postgres, Vercel) with Forge conventions, not building new frameworks.

December 2025 - Specification vs Implementation

Context: Started writing detailed specs, JSON schemas, HTML documentation for "The Agentic Web Standard."

What happened:

  • Created extensive documentation
  • Realized Rails, Spring, Django all extracted patterns from working apps
  • Nobody started with a specification

Learning: Specs come after, not before. Build the app, extract the patterns, document what worked.

Action: Stopped spec writing. Starting rally-hq build. Will document learnings as we go, extract spec later.

December 2025 - Scope Clarity

Context: Oscillated between "build rally-hq" and "build a standard others adopt."

What happened:

  • Challenged sycophantic framing
  • Honest assessment: building a standard others adopt is unlikely for an individual
  • But: building a good app with clear patterns is achievable
  • The goal is C (intellectual itch about how web dev should work) leading to A (useful app)

Learning: Be honest about goals. Don't dress up personal learning as industry disruption. If the patterns are good, adoption might follow. If not, you still learned something.

Action: Focus on building rally-hq well. Extract tooling if patterns prove out. Don't prematurely optimize for adoption.

Template for Future Entries

Copy this when adding new learnings:

## [Date] - [Topic]

**Context:**

**What happened:**

**Learning:**

**Action:**