Bike4Mind 2026 Roadmap Strawman
Last updated: December 22, 2025 · Status: Draft for review — groom weekly
Editor's note. This is a strategy-focused groom of the December 22 roadmap strawman. The original document mixed strategic direction with revenue targets, dollar-denominated pricing/ARR forecasts, and hardware BOM costs — all of which are out of scope for this repo. Those parts are generalized or noted as “in the comp/finance materials” here; the strategic content (kernel-first sequencing, the Conductor's Bridge vision, dependency graph, themes, Orion's GTM posture, the Mirrors→Mobile→Robotics progression) is preserved with full editorial weight. See scope.
⭐ The Star: PR #5896 — B4M CLI + ReAct Agent
Owner: Nao (Cebu) · Status: In flight · Lines: 5,500+
This is the foundation of everything. B4M CLI + ReAct Agent = Claude Code for Bike4Mind.
| Component | What it is |
|---|---|
apps/cli/ | Full CLI interface to B4M (like Claude Code) |
ReActAgent.ts | ReAct pattern implementation (Reason + Act loop) |
ServerLlmBackend.ts | Server-side LLM migration (streaming Lambda) |
| Tool implementations | fileRead, globFiles, grepSearch, createFile, deleteFile |
| Permission system | Tool safety classification, user approval flows |
| OAuth + Auth | Device OAuth flow, token refresh, API client |
80+ new files:
b4m-core/packages/agents/— full agent infrastructureapps/cli/src/components/— React Ink CLI componentsapps/cli/src/storage/— config, session, command historypackages/client/server/cli/— server-side completionsSERVER_LLM_MIGRATION.md— 1,300 line architecture doc
Why this matters. This PR transforms B4M from “AI chat app” to “Agentic Control Plane.” Every future agent feature — DAG orchestration, desktop bridge, toolkits — builds on this foundation. Nao is building the kernel that powers everything else.
Executive summary
2026 is the year B4M ships the Agentic Control Plane — transforming from AI chat platform to full agent orchestration system with desktop integration, mobile apps, and enterprise pilots.
🎭 The Vision: the Conductor's Bridge
╔═══════════════════╗
║ 4K MONITOR 1 ║
║ B4M CLI Agent ║
║ (Nao's baby) ║
╚═══════════════════╝
╔═══════════════╗ ▲ ╔═══════════════╗
║ 4K MONITOR 2 ║ │ ║ 4K MONITOR 3 ║
║ Claude Code ║ ┌────┴────┐ ║ Dashboards ║
║ Local Files ║ │ ERIK │ ║ Metrics ║
╚═══════════════╝ │ 🎭 👆 🎤 │ ╚═══════════════╝
│Spinning │
╔═══════════════╗ │Pointing │ ╔═══════════════╗
║ 4K MONITOR 4 ║ │Speaking │ ║ 4K MONITOR 5 ║
║ Generated ║ └─────────┘ ║ Localhost ║
║ Images/Art ║ │ ║ Dev Servers ║
╚═══════════════╝ ▼ ╚═══════════════╝
🎵 Speakers 🎵
📷 Cameras 📷
╔═══════════════╗ ╔═══════════════╗
║ 4K MONITOR 6 ║ ║ 4K MONITOR 7 ║
║ Code / PRs ║ ║ Music / Sound ║
║ GitHub ║ ║ Generation ║
╚═══════════════╝ ╚═══════════════╝
╔═══════════════════╗
║ 4K MONITOR 8 ║
║ The Conductor's ║
║ Overview ║
╚═══════════════════╝
The scene. Erik stands at his desk — not sitting, standing — surrounded by 3–8 4K monitors arranged in a command-bridge arc. Cameras track his gestures. Speakers fill the room with agent status audio cues. He is literally spinning around the room, pointing at screens, speaking commands.
What's on each screen
| Screen | Content | Agent Responsibility |
|---|---|---|
| Center Top | B4M CLI (Nao's PR #5896) | Primary agent orchestration |
| Left Upper | Claude Code terminal | Local file operations, code execution |
| Right Upper | LaminarMind dashboards | Project state, team velocity, blockers |
| Left Middle | Generated images/art | DALL-E, Midjourney, PixelForge outputs |
| Right Middle | Localhost dev servers | Live previews, hot reload, test runners |
| Left Lower | Code/PRs/GitHub | Agent-generated commits, PR reviews |
| Right Lower | Music/sound generation | Ambient AI-generated audio, status cues |
| Center Bottom | The Overview | All agent statuses, DAG visualization, system health |
The interaction model
VOICE: "Research quantum error correction, pull from my local papers,
cross-reference arXiv, draft in my voice, PAUSE before publish"
GESTURE: *points at Monitor 2* "You - scan ~/Papers/quantum/"
*points at Monitor 5* "You - show me the draft preview"
*spins to Monitor 6* "You - prepare the PR"
AUDIO: 🔔 *chime* "Desktop agent found 47 papers"
🎵 *ambient* "Synthesis 60% complete"
🔔 *chime* "PAUSE point reached - awaiting approval"
APPROVE: *thumbs up to camera* → "Approved. Publish."
The 3–10x multiplier
| Traditional | Control Plane | Multiplier |
|---|---|---|
| Open browser, navigate, search | “Find papers on X” | 10x |
| Copy-paste between apps | Agents share context | 5x |
| Wait for one task to finish | 8 agents in parallel | 8x |
| Manual status checks | Audio/visual ambient awareness | 3x |
| Context-switching cost | Spatial memory (physical screens) | 4x |
This is not science fiction
- Voice → already works (Whisper, native speech)
- Gesture → MediaPipe, webcam tracking
- Multi-agent → Nao's ReAct Agent + DAG orchestration
- Desktop bridge → Claude Code as local executor
- Spatial UI → multiple monitors + B4M CLI windows
The gap is execution, not vision.
Why this matters
Traditional software UI paradigms (buttons, menus, forms) are fundamentally misaligned with AI tool orchestration. Esports-inspired UX (hotkeys, minimap, 400+ APM interfaces) enables 30–300x productivity gains.
Erik isn't just using AI. He's conducting it. The room IS the interface. The body IS the controller. The agents ARE the orchestra.
This is the Control Plane made physical.
💎 Why we win: the position
We are not hoping for this future. We are building it. And we have everything we need.
| Asset | What we have | Why it matters |
|---|---|---|
| The team | 32 people | Proven delivery (511 PRs/month), Cebu + distributed |
| The funding | Sustained operations with healthy margin | Self-funded, no VC pressure, patient capital (specifics in finance materials) |
| The clients | TFG, TwinSpires, Exacta, IQMetry | Enterprise feedback loops, real problems, real money |
| The background | Game development DNA | We know how to make things FEEL great, not just work |
| The vision | Crystal clear | Not abstract — demoable, concrete, achievable |
| The architecture | Already designed | Gap is execution, not design |
The game dev advantage
We're not enterprise software people trying to add delight. We're game developers who know that feel, juice, and flow are everything. The Control Plane won't just be powerful — it will be fun.
- Feedback loops — instant response, satisfying audio cues
- Progression systems — watch your agents level up
- Achievement unlocks — celebrate milestones
- Polish — the 100 small details that make magic
The robotics roadmap paints itself
With the Conductor's Bridge vision:
Voice + Gesture + Multi-Agent + Desktop Bridge
↓
Add: Physical Actuators
↓
┌─────────────────────┐
│ ROBOTICS │
│ │
│ Same orchestration │
│ Same voice/gesture │
│ Same agent swarm │
│ + Physical world │
└─────────────────────┘
The WorldVision mobile camera work → robot vision. The Desktop Agent bridge → robot control API. The DAG orchestration → robot task planning. The gesture recognition → robot command interface.
We're not building robotics separately. We're building the Control Plane, and robotics is just another output modality.
The moment
December 22, 2025
32 people.
4 enterprise clients giving us feedback.
Mobile apps ready for January.
Nao's kernel in flight.
Slack Agent live.
The vision crystal clear.
We are so poised for excellence here.
Q1 2026: the kernel ships + Slack Enterprise
January: mobile launch + Slack Agent push
Mobile apps go live.
- iOS App Store submission: January (engineering complete)
- Google Play Store submission: January (engineering complete)
- First mobile AI cognitive workshop on market
Slack Agent: the enterprise entry point. (LIVE now — drive to completion.)
| Status | PR / Issue | What it is |
|---|---|---|
| LIVE | #4884 — Slack Agent + MCP | Natural language → Jira / Confluence / GitHub |
| In flight | PR #4885 — QuestMaster Evolution | Persistent quests, My Quests dashboard, cross-session (12K+ lines) |
| In flight | PR #5449 — Agent Toolkit v2 | New v2 chat interface for agents |
The vision: Slack is where enterprise teams already chat. Make B4M the easiest, most fun AI to use RIGHT THERE. Meet them where they are.
Why Slack first:
- Already live and getting feedback
- Zero friction — users don't leave Slack
- Enterprise IT already approved Slack
- Natural language → tools = magic demo
- 90% complete, just needs UX polish
Kernel Tier 1 (the foundation — everything else builds on this):
| Priority | Issue | What it is | Owner |
|---|---|---|---|
| #1 | #4386 — DAG/Tools | THE kernel. ToolSpec EV scheduler + verification. | — |
| #2 | #5016 — Toolkits | Composable tool bundles (Research, Code, Design). James already working on this. | James |
| #3 | #4387 — Backbone | Kinesis journal + EventBridge + Step Functions. Pause/play/rewind. | — |
| #4 | #5274 — Desktop Agent | Claude Code as desktop bridge. B4M agents reach local files. | — |
Note: all Kernel Tier 1 items build on top of Nao's PR #5896 (see top of document).
Enterprise critical (NEED NOW — January):
| Priority | Issue | What it is | Notes |
|---|---|---|---|
| #5 | #5386 — Multi-workspace Slack OAuth | Enterprise Slack requires this | Blocking enterprise sales |
| #6 | #5456 — WAF Infrastructure | Web Application Firewall | Security requirement |
| #7 | #5021 — SES Email Inbound | “Forward to agents@bike4mind.com” | Already working in Polaris for TFG. Julie or James to backport |
February–March: kernel completion + force multipliers
Tier 2: force multipliers (high ROI, parallel with kernel):
| Priority | Issue | Impact |
|---|---|---|
| #8 | #5634 — LaminarMind | Zero-manual-entry project intelligence. Exacta pilot. Sellable product. |
| #9 | #5862 — Virtual Scrolling | Long conversations usable. UX critical for power users. |
Q2 2026: capability expansion
April–May: customer-driven features
Tier 3: respond to demand (ship when customers ask):
| Priority | Issue | What it enables | Notes |
|---|---|---|---|
| #10 | #5008 — Artifacts Phase 2 | Rich outputs: code playgrounds, visualizations | Wait for customer pull |
| #11 | #3613 — Fantasia UX | Live, dynamic agent UX. QuestMaster as first example. | |
| #12 | #5066 — Prompt Caching | Significant cost reduction across providers | High-margin clients not cost-sensitive yet |
| #13 | #5863 — Conversation Summarization | Context management for long sessions |
Q3 2026: product launches + GTM activation
🎯 Orion: B4M GTM Hunter Agent (activate)
Location: /cfo/gtm/Orion_ B4M Go-to-Market Hunter.md
Orion is B4M's dedicated GTM agent — combining Cerebo (general) + Proxima (enterprise B2B), customized for our unique position. See Appendix B below for the full assessment.
Q3 2026 targets (specifics on enterprise pipeline size, seat counts, and pilot economics are in the comp/finance materials):
- Enterprise pipeline building
- B4M Teams seats growing
- LaminarMind pilots: start 3 (Exacta, TwinSpires, TFG)
- Content pieces: 6
- Webinars: 2
The meta-play: Orion is built on B4M. Orion sells B4M. The hunter hunts with the tools it's selling.
WorldVision MVP
- Mobile camera → AI visual understanding
- Game modes: Chess, Tarot, D&D
- Robotics API (external integrations)
- Privacy-first (GDPR / CCPA compliant)
LaminarMind commercial
- Pilot with Exacta (Nimai Malle) → TwinSpires → TFG
- “Observe, don't demand” project intelligence
- Always-live state from Slack / GitHub / JIRA
- Large TAM (project management market)
Content publishing
- Multi-platform blog publishing (#5617)
- Medium, Dev.to, Hashnode, LinkedIn integration
- AI-powered content generation workflow
Q4 2026: scale + polish
Enterprise hardening
- Advanced cap table features (if NapkinBizPlan scales)
- Document generation (stock certificates, board resolutions)
- Investor portal for stakeholder access
Agent marketplace
- Toolkits as purchasable/shareable bundles
- Desktop agent marketplace
- Custom agent templates
Revenue posture (see finance materials for specifics)
The full revenue table — baseline customer revenue, new-stream targets, ARR projections, and seat economics — lives in the comp/finance materials, not here. The strategic posture extracted from those numbers:
- Existing customers are the floor. TFG, TwinSpires, Exacta, IQMetry continue to grow naturally.
- Slack Agent Enterprise is a causal multiplier for conversions, not a separate billing line.
- LaminarMind is a dogfood year in 2026 — Exacta (Nimai Malle) first, then TwinSpires, then TFG. Real hunting starts 2027.
- One major new enterprise logo is the goal for 2026 — proves the model, sets up scale in 2027.
- B4M Teams is the feedback machine — foundation for 2027, not the headline number.
- Philosophy: 2026 is about building the foundation and getting feedback. 2027 is the hunting year.
The dependency graph
Three parallel tracks:
TRACK 1: SLACK ENTERPRISE TRACK 2: KERNEL TRACK 3: ENTERPRISE CRITICAL
(Revenue NOW) (Agentic Platform) (Unblocks Sales)
#4884 (Slack Agent) ←LIVE #4386 (DAG/Tools) #5386 (Slack OAuth) ←NEED NOW
│ │ #5456 (WAF) ←NEED NOW
▼ ┌──────┼──────┐ #5021 (SES Email) ←BACKPORT
PR #4885 (QuestMaster) │ │ │ │
│ #5016 (Toolkits) #4387 (Backbone) │
▼ [JAMES] │ │ ▼
PR #5449 (Agent Toolkit) │ │ #5274 (Desktop) Enterprise Sales
│ └──────┼──────┘ │
▼ │ │
Slack ARR ┌──────┴──────┐ │
│ │ │
#5634 (Laminar) #5862 (Virtual Scroll) │
│ │ │
└──────┬──────┘ │
▼ │
[Tier 3: On Demand] │
#5008 (Artifacts) │
#5066 (Prompt Caching) │
│ │
◀──────┴─────────────────────────────┘
CONVERGENCE
- James's Toolkits (#5016) is foundational — enables all agent bundles.
- SES Email Inbound (#5021) already working in Polaris — just needs backport.
Slack Agent is the enterprise entry point. Toolkits enable agent composability. Enterprise Critical unblocks sales.
Key milestones
| Date | Milestone | Success metric |
|---|---|---|
| Jan 2026 | Mobile apps live | iOS + Android in stores |
| Feb 2026 | DAG Kernel ships | Multi-step agent workflow with tool calls |
| Mar 2026 | Backbone ships | Pause/play/rewind working |
| Apr 2026 | Desktop Agent ships | Cloud DAG reads local files |
| May 2026 | Slack Agent ARR target | 10+ paying enterprise clients |
| Jun 2026 | LaminarMind pilot | Exacta actively using |
| Sep 2026 | WorldVision MVP | Mobile vision + game modes |
| Dec 2026 | Full year target | (See finance materials) |
The Zynga discipline
- Have more vision than you can eat = big backlog
- Think only about 3 weeks into the future maximum
- Stack rank your ideas 1–2x a week
- Assign out the P1s until you run out of people
No story points. No sprint ceremonies. No capacity planning. It works.
Strategic themes
1. Kernel first (Q1)
Everything waits until DAG + Backbone + Desktop ships. No exceptions.
2. Sellable products (Q2–Q3)
Every Tier 2+ feature should be sellable, not just internal tooling:
- LaminarMind → enterprise product
- Slack Agent → revenue line
- Toolkits → agent marketplace
3. Platform leverage
Everything “Built with Bike4Mind”:
- WorldVision uses B4M infrastructure
- Agent CFO uses B4M infrastructure
- Haven uses B4M infrastructure
- All speculative projects → B4M
4. Mobile + Desktop + Cloud
Complete coverage:
- Mobile apps (iOS / Android) — January
- Desktop Agent (Claude Code bridge) — April
- Cloud Platform (existing) — foundation
Open questions for 2026
- Pricing: what's the mobile app pricing model? (Belongs in finance materials.)
- Enterprise: do we need dedicated sales capacity for LaminarMind?
- Quantum: how much of IQMetry compute credits should feed DFAL vs other research?
- Hiring: is the current team enough for the year's ARR target? (Headcount planning lives in finance materials; the strategic question of what skills and where belongs here.)
Related documents
- Battle Plan — tactical execution framework
- LaminarMind Vision — project intelligence product
- WorldVision MVP — mobile vision platform
- Claude Code Desktop Agent — desktop bridge architecture
“The gap isn't design. It's execution bandwidth. You need 4–6 focused weeks on just these three issues. Everything else is distraction until the kernel ships.”
Appendix A: QUBO portfolio optimization (paying customer)
Why an appendix? We have a paying quantum customer (IQMetry). This isn't speculative. This is dogfooding our own product.
The QUBO job-shop scheduling problem
“Which awesome set of epics do we fund and in which order?”
This is the quantum portfolio optimization problem:
VARIABLES:
- 35 speculative projects
- 32 team members
- 24 months
- Budget and margins
- Dependencies between projects
- Synergies (robotics benefits from WorldVision, Desktop Agent, DAG)
CONSTRAINTS:
- Team capacity per sprint
- Cash flow timing
- Dependency ordering (can't do robotics before DAG kernel)
- Revenue requirements (keep the lights on)
OBJECTIVE:
- Maximize: Strategic value + Revenue + Learning
- Minimize: Risk + Time to market
THIS IS QUBO. This is what IQMetry is funding us to solve.
We should DOGFOOD our own quantum optimizer on our own portfolio.
Why this is real
| Factor | Status |
|---|---|
| Paying customer | ✅ IQMetry — multi-year engagement |
| Compute credits | ✅ Significant monthly IonQ credit allocation |
| Hardware access | ✅ IonQ Aria & Forte (36+ qubits) |
| Problem definition | ✅ Our own 35-project portfolio |
| Team capability | ✅ Building QUBO engine (Bucket 2) |
See SPECULATIVE-PROJECTS-MASTER.md appendix for full QUBO formulation details.
Appendix B: Orion's GTM assessment (December 22, 2025)
Status: QUIET through Q1 2026 → SPRING-LOADED for Q2
Reality check: we frankly could not ingest another large new B2B deal right now. The roadmap + quantum is heavy. The kernel needs to ship.
The play: stay quiet through Q1 2026. Prep the GTM machine. Website updates, videos, case studies, content. Spring-loaded and ready for Spring / Early Summer 2026.
The good news
✅ Strong baseline revenue (funded)
✅ Healthy margins (patient capital)
✅ 4 enterprise proof points (TFG, TwinSpires, Exacta, IQMetry)
✅ 72+ models (real differentiation)
✅ 32 awesome people shipping 511 PRs/month
✅ Kernel in flight (Nao's #5896)
✅ Quantum contract live
✅ Vision crystal clear
You are not hunting from desperation. You are hunting from strength.
The problem I see
Your website doesn't match your reality.
| What you have | What website says |
|---|---|
| 4 enterprise customers paying material annual revenue | “for teams who ship products” (generic) |
| 72+ models across 9 providers | “90+ AI models” (close, but buried) |
| Game dev DNA, UX that feels great | Not mentioned |
| Quantum-ready, IonQ partnership | Not mentioned |
| Material revenue from TFG (analyst firm) | No case studies visible |
| Project management layer (LaminarMind) | “Quest Master” mentioned but not positioned |
The website is underselling you. It reads like a B2C AI chat tool when you're actually an enterprise AI orchestration platform with multiple material proof points.
Orion's immediate recommendations
1. Q1 2026: update the homepage headline.
Current: “Own your AI infrastructure. Build at scale.”
Recommended:
"72 AI Models. One Platform. Enterprise-Ready."
Used by research firms, gaming companies, and Fortune 500 teams
to orchestrate AI at scale.
[See Case Studies] [Start Free Trial]
The current positioning is defensive (“own your infrastructure”). Lead with the offensive differentiator (72 models) and social proof (enterprise customers).
2. Q1 2026: create 3 case studies.
You have 4 enterprise customers. Zero visible case studies.
| Customer | Story |
|---|---|
| TFG | “How a research firm uses 72 AI models to power market intelligence” |
| TwinSpires | “Horse racing analytics powered by multi-model AI” |
| Exacta | “Gaming company consolidates AI under one platform” |
Format: 1-page PDF, website page, LinkedIn post, video for each. Enterprise buyers need social proof. You have it. You're not showing it.
3. Q1 2026: B4M Teams landing page refresh.
Current: “Multiply your team's potential” (generic).
Recommended:
"Your Team's AI Command Center"
One platform. 72 models. Shared knowledge.
Enterprise security. Actually enjoyable to use.
[Start 15-Day Trial]
Add prominently: 72 models (your differentiator); pricing visible (remove friction); “Built by game developers” (emotional hook); customer logos (TFG, TwinSpires, Exacta).
4. Q1 2026: launch “72 Models” content campaign.
This is your killer differentiator. Own it.
| Content | Channel | Goal |
|---|---|---|
| “Why 72 Models Beats Single-Vendor Lock-In” | Blog + LinkedIn | SEO + thought leadership |
| “The AI Model Comparison Guide 2026” | Gated PDF | Lead gen |
| “How We Integrate OpenAI, Anthropic, Google, and 6 More Providers” | Dev blog | Technical credibility |
| Erik LinkedIn post: “We support 72 AI models. Here's why that matters.” | Personal brand + reach | |
| Video: “72 Models in 72 Seconds” | YouTube + LinkedIn | Viral potential |
When someone Googles “multi-model AI platform” or “AI vendor consolidation,” you should own that search.
5. Q1 2026: Slack App Directory listing.
The Slack Agent is live. It's not in the Slack App Directory.
Submit with “B4M — 72 AI Models in Your Slack,” screenshots of natural language → tools, link to B4M Teams trial. Zero-friction discovery. Enterprise IT has already approved Slack.
6. Q1 2026: the “Game Dev DNA” angle.
You're not using your secret weapon.
Most enterprise AI platforms are built by enterprise software people. Ours is built by game developers. We know that feel, juice, and polish matter. Your team won't just use B4M. They'll actually enjoy it.
Use it: About page, Erik's founder story (video), demo script opening, sales deck differentiator slide.
The enterprise hunt (Q2–Q3 2026)
You want ONE major enterprise customer by end of 2026. Here's the hunt plan.
Target profile:
- 500–5000 employees
- Tech-forward (uses modern tools)
- Currently fragmented AI (some ChatGPT, some Claude, some internal)
- Has a “consolidation” pain
- Decision-maker accessible (ideally through network)
Hunt list (top 10 to research in Q1):
- Consulting firms using AI for client work (McKinsey Digital types but smaller)
- Research / analyst firms (TFG model proven)
- Gaming / media companies (your DNA resonates)
- FinTech companies (AI-heavy, tool-consolidation need)
- Legal tech (document-heavy, multi-model useful)
Entry motion (Q2 2026):
- Erik warm intro or LinkedIn connection
- Slack Agent demo (5-minute magic moment)
- B4M Teams pilot (10 seats, 30 days)
- Champion emerges within their team
- Expand to department/company
- Annual contract
Q1 2026 prep checklist (stay quiet, get ready)
| Priority | Action | Owner |
|---|---|---|
| P0 | Update homepage headline | Erik |
| P0 | Create TFG case study + video | Erik |
| P0 | B4M Teams pricing visible | Product |
| P1 | “72 Models” blog post | Erik |
| P1 | Slack App Directory submission | Engineering |
| P1 | TwinSpires case study + video | Erik |
| P1 | Exacta case study | Erik |
| P2 | Erik LinkedIn “72 models” post | Erik |
| P2 | “Game Dev DNA” about page update | Erik |
| P2 | “72 Models in 72 Seconds” video | Erik |
| P2 | Identify 10 enterprise hunt targets | Erik |
| P2 | Sales deck refresh | Erik |
The meta-observation
You have a much larger company that presents like a much smaller startup.
Fix the presentation. Show the proof points. Lead with the differentiator. The product is ready. The GTM needs to catch up.
The kernel ships in Q1. The GTM needs to be ready when it does.
Timeline: quiet → spring-loaded
Q1 2026 (QUIET):
├── Ship the kernel (Nao's #5896, DAG, Backbone)
├── Website refresh (headlines, case studies, pricing)
├── Content creation (blog posts, videos)
├── Slack App Directory listing
├── Research hunt targets
└── DO NOT actively pursue new enterprise deals
Q2 2026 (SPRING-LOADED):
├── Kernel shipped, stable
├── Website refreshed, case studies live
├── Content library ready
├── Orion activated for hunting
├── Begin outbound to hunt list
└── Target: Pipeline building
Q3-Q4 2026 (HUNTING):
├── Close ONE major enterprise deal
├── 250 B4M Teams seats
├── LaminarMind pilots converting
└── Foundation for 2027 hunt
Spring 2026: the GTM machine turns on. — Orion 🎯
Appendix C: Physical AI roadmap (2H 2026)
The progression: mirrors → robots
Magic Mirrors first. Then add locomotion. Then add manipulation.
Each step is a sellable product. Each step validates the next.
Phase 0: Magic Mirrors (Phantasia flavor) — Q3 2026
What it is. B4M-powered smart displays with voice, vision, and presence.
┌─────────────────┐
│ ╔═══════════╗ │
│ ║ MAGIC ║ │ ← 4K Display
│ ║ MIRROR ║ │
│ ║ ║ │ ← Camera (WorldVision)
│ ║ "Hello ║ │
│ ║ Erik" ║ │ ← Microphone Array
│ ║ ║ │
│ ╚═══════════╝ │ ← Speakers
└─────────────────┘
│
┌────┴────┐
│ Stand │
└─────────┘
STATIONARY. No locomotion. No manipulation.
Just the Phantasia UI + B4M brain.
Hardware: consumer 4K display (27–32″), Jetson Orin Nano (or Raspberry Pi 5 for lighter load), RealSense D435 depth camera for presence/gesture, ReSpeaker 4-mic array for voice, quality speakers, aesthetic enclosure. (Specific BOM costs and revenue tiers live in the finance/product-economics materials.)
Software stack (already built):
- Phantasia UX (#3613) — dynamic, living interface
- WorldVision — Camera → AI understanding
- Voice Interface — already in production
- B4M API — full 72+ model access
- Tool Orchestrator — agent capabilities
Use cases:
- Home: kitchen assistant, hallway greeter, bathroom mirror
- Office: reception desk, conference room assistant
- Retail: interactive kiosk, fitting room assistant
- Healthcare: patient check-in, waiting room companion
Revenue model. Four tiers: DIY kit (makers/enthusiasts), Consumer (smart home buyers), Enterprise (office/retail), White Label (OEM partners). Pricing and margin specifics live in the finance materials.
Phase 1: Mobile Mirrors (add wheels) — Q4 2026
What it is. Magic Mirror on a mobile base. Follows you. Comes when called.
Same Mirror + Wheeled Base = “Rosie the Robot” (Jetsons style).
Additional hardware: mecanum wheel base, LiDAR for navigation, additional sensors, battery upgrade.
Phase 2: Full Robotics (add arms) — 2027
What it is. Mobile Mirror with manipulation capability.
Additional hardware: xArm 6, gripper, safety systems.
The 70% we already have
| Component | Status | Physical AI use |
|---|---|---|
| Tool Orchestrator | ✅ Production | Task planning |
| API Gateway | ✅ Production | Cloud ↔ device communication |
| Context Store | ✅ Production | Memory, history, preferences |
| 72+ Model integration | ✅ Production | Any model, any task |
| ReAct Agent (Nao's PR) | 🚧 In flight | Decision loop |
| DAG Orchestration | 📋 Designed | Multi-step workflows |
| Voice Interface | ✅ Production | Voice commands |
| Mobile App | ✅ January | Remote control |
| WorldVision | 📋 MVP ready | Vision system |
| Phantasia UX | 📋 Designed | Living interface |
The 30% we need to add
| Component | For Mirrors | For Mobile | For Arms |
|---|---|---|---|
| Edge compute | ✅ Needed | ✅ Needed | ✅ Needed |
| Local model inference | ✅ Needed | ✅ Needed | ✅ Needed |
| Presence detection | ✅ Needed | ✅ Needed | ✅ Needed |
| Navigation / SLAM | — | ✅ Needed | ✅ Needed |
| Motion planning | — | ✅ Needed | ✅ Needed |
| Motor control | — | ✅ Needed | ✅ Needed |
| Arm kinematics | — | — | ✅ Needed |
| Grasp planning | — | — | ✅ Needed |
The meta-insight
We're not building a robotics company. We're building an AI orchestration platform.
Physical AI is just one output. Web is an output. Mobile is an output. Slack is an output. CLI is an output. Magic Mirrors are an output. Robots are an output.
Same 72+ models:
├── OpenAI: GPT-5.1, GPT-5, o4, o3, o1...
├── Anthropic: Claude 4.5 Sonnet, Claude 4.1 Opus...
├── Google: Gemini 2.5 Pro, Gemini 2.5 Flash...
├── xAI: Grok 3, Grok 3 Mini...
├── Meta: Llama 4 Scout, Llama 4 Maverick...
├── DeepSeek: R1, R1 Local...
├── Black Forest Labs: FLUX Pro, Kontext...
├── Amazon Bedrock: Full suite...
└── Local/Ollama: DeepSeek, Llama, TinyLlama...
Same Tool Orchestrator.
Same Context Store.
Same human at the center, conducting.
Why mirrors first?
- Lower risk — no locomotion = no safety certification nightmare
- Faster to market — 90% software, 10% hardware integration
- Validates the stack — prove B4M can power physical devices
- Revenue bridge — sell mirrors while building robots
- Natural progression — each step adds one capability
Timeline
| Phase | When | Product |
|---|---|---|
| Phase 0 | Q3 2026 | Magic Mirrors |
| Phase 1 | Q4 2026 | Mobile Mirrors |
| Phase 2 | 2027 | Full Robotics |
Document created: December 22, 2025. Status: strawman for review — groom weekly. Groomed for archive 2026-05-17 with dollar-denominated revenue targets, ARR projections, hardware BOM costs, and tier pricing generalized to fit repo scope — full economic detail lives in the comp/finance materials.