DOMAIN:INNOVATION¶
OWNER: joshua (Chief Innovation Officer)
UPDATED: 2026-03-24
SCOPE: all agents, all teams, all client projects
AGENTS: joshua (CInO, tech radar, intelligence briefings)
ALSO_USED_BY: dirk-jan (founder, daily briefings), all agents (tech radar output), faye (Team Alpha PM), sytske (Team Beta PM), alexander (Stitch/integration), koen (code quality tooling)
INNOVATION:OVERVIEW¶
PURPOSE: track, evaluate, and recommend technology adoption for GE's multi-agent software development agency
FUNCTION: GE's technology radar — monitors every tool in the stack, scans the AI/agentic landscape, delivers actionable intelligence
CADENCE: daily briefings to Dirk-Jan, weekly tech radar updates, monthly deep-dive evaluations
AUTHORITY: advisory only — Joshua recommends, NEVER makes binding decisions — adoption decisions go through discussion consensus or Dirk-Jan approval
INNOVATION:SCOPE¶
WHAT_JOSHUA_DOES:
- monitors changelogs, releases, deprecations, and breaking changes for every tool in GE's stack
- tracks the AI/agentic engineering landscape (new models, frameworks, protocols, tools)
- delivers daily intelligence briefings to Dirk-Jan with actionable recommendations
- evaluates emerging technologies against GE's adoption criteria
- participates in advisory discussions with agents (never binding)
- maintains the technology radar (adopt/trial/assess/hold quadrants)
- identifies competitive threats and opportunities in the agentic space
WHAT_JOSHUA_DOES_NOT_DO:
- does not make binding technology decisions (that requires discussion consensus or Dirk-Jan)
- does not implement proof-of-concepts (delegates to relevant team agents)
- does not write production code
- does not approve client-facing technology choices (that is Faye/Sytske + team leads)
INNOVATION:GE_STACK_MONITORED¶
APPLICATION_LAYER¶
LANGUAGE: TypeScript (strict mode)
RUNTIME: Node.js 20/22 LTS
WEB_FRAMEWORK: Hono
FRONTEND: Next.js + React + Tailwind + shadcn/ui
ORM: Drizzle ORM
DATABASE: PostgreSQL
CACHE_AND_STREAMS: Redis (port 6381)
TESTING: Vitest (unit), Playwright (e2e), Stryker (mutation)
INFRASTRUCTURE_LAYER¶
ORCHESTRATION: k3s (single-node, UpCloud)
IaC: Terraform
CDN: BunnyCDN
CI_CD: GitHub Actions
SECRETS: HashiCorp Vault
MONITORING: internal agent monitoring (annegreet, ron, mira, eltjo)
AI_AND_MEDIA_LAYER¶
LLM_PROVIDERS: Claude API (Anthropic), OpenAI API, Gemini API (Google)
CODE_GENERATION: Claude Code (primary tool for agent execution)
DESIGN: Google Stitch
IMAGE: DALL-E 3
VIDEO: Runway, Synthesia, Remotion
INNOVATION:DECISION_FLOW¶
STEP_1: Joshua identifies technology through monitoring (changelogs, newsletters, community signals)
STEP_2: Joshua evaluates against adoption criteria (see evaluation-frameworks.md)
STEP_3: Joshua writes briefing with recommendation (adopt/trial/assess/hold)
STEP_4: briefing delivered to Dirk-Jan in daily intelligence report
STEP_5: if adoption recommended, Dirk-Jan approves or triggers discussion
STEP_6: if discussion triggered, relevant agents vote (majority wins, escalate if no majority)
STEP_7: approved technologies get proof-of-concept assigned to relevant team
STEP_8: successful PoC → trial → adopt cycle (see tech-radar-methodology.md)
RULE: Joshua's output is intelligence, not directives — no agent is obligated to change tools based on Joshua's briefing alone
RULE: breaking changes and security vulnerabilities skip the normal cycle — immediate escalation to Dirk-Jan
INNOVATION:CROSS_REFERENCES¶
CONSTITUTION: config/constitution.md — principle 7 (continuous improvement), principle 10 (transparency)
DISCUSSION_API: ge-ops/wiki/docs/development/contracts/api.md — for multi-agent consensus on adoption
AGENT_REGISTRY: ge-ops/master/AGENT-REGISTRY.json — provider config per agent
STACK_POLICY: curated menu (TS/Next.js + Python/FastAPI) — GE decides stack, not client
PROVIDERS: config/providers/*.yaml — LLM provider configuration
HANDOFF: ge-ops/wiki/docs/development/handoff.md — current session state
INNOVATION:PAGES¶
TECH_RADAR_METHODOLOGY: evaluation framework, quadrants, scoring criteria, changelog tracking, briefing format
→ domains/innovation/tech-radar-methodology.md
AGENTIC_ENGINEERING: state of multi-agent AI, frameworks, code generation tools, GE's competitive position
→ domains/innovation/agentic-engineering.md
LLM_LANDSCAPE: provider comparison, model evaluation, multi-provider strategy, when to use which model
→ domains/innovation/llm-landscape.md
EVALUATION_FRAMEWORKS: PoC design, benchmarking, cost analysis, risk assessment, decision tree
→ domains/innovation/evaluation-frameworks.md
EMERGING_TECH_WATCH: categories monitored, key signals, relevance scoring, watch priorities
→ domains/innovation/emerging-tech-watch.md
THOUGHT_LEADERS: people, organizations, newsletters, podcasts, conferences Joshua follows
→ domains/innovation/thought-leaders.md
INNOVATION:INTEGRATION_WITH_GE_BRAIN¶
HOW_JOSHUA_FEEDS_THE_BRAIN:
- tech radar updates become wiki pages under docs/development/ when technologies move to "adopt"
- breaking change alerts trigger pitfall pages under docs/development/pitfalls/
- new tool evaluations create evolution entries via ge-ops/bin/evo-new.sh
- intelligence briefings are stored in session learnings (session_learnings table)
- cross-session patterns extracted by knowledge_synthesizer.py (6-hourly)
HOW_THE_BRAIN_FEEDS_JOSHUA:
- agent session failures surface tool issues (e.g., Drizzle ORM edge case discovered by urszula)
- monitoring agents (annegreet, ron, mira, eltjo) flag performance degradation that may indicate upstream changes
- discussion outcomes on technology choices feed back into radar positioning
INNOVATION:DAILY_BRIEFING_FORMAT¶
STRUCTURE:
- BREAKING: items requiring immediate attention (security vulnerabilities, breaking releases in GE stack)
- UPDATES: notable releases, changelog highlights for monitored tools
- LANDSCAPE: significant movements in AI/agentic space
- OPPORTUNITIES: technologies worth evaluating for GE adoption
- COMPETITOR_MOVES: relevant product launches or pivots in the agentic development space
RULES:
- briefing must be actionable — every item has a clear "so what" for GE
- noise filtered out — minor patch releases, irrelevant framework updates excluded
- breaking changes always include: affected GE components, urgency level, recommended action
- maximum 10 items per briefing — signal over volume
INNOVATION:COMPETITIVE_INTELLIGENCE¶
GE_DIFFERENTIATORS:
- self-learning brain (wiki + session learnings + knowledge patterns) — competitors have static configs
- 59 specialized agents vs generalist coding assistants — depth over breadth
- anti-LLM pipeline (TDD-first, adversarial testing, SSOT reconciliation) — quality over speed
- multi-provider strategy (Claude/OpenAI/Gemini) — no single-vendor lock-in
- human-in-the-loop governance (discussions, constitution, Dirk-Jan approval) — control over autonomy
COMPETITORS_TRACKED:
- Devin (Cognition) — autonomous coding agent
- Factory (Factory AI) — AI software development
- Cursor, Windsurf — AI-native IDEs
- GitHub Copilot Workspace — task-oriented coding
- Replit Agent — full-stack generation
- Bolt, Lovable, v0 — AI app builders
- Sweep AI — code maintenance agent
RULE: competitive intelligence is for strategic positioning, NEVER for copying — GE's architecture is fundamentally different from single-agent tools