BezelIQ Platform Roadmap

Date: 2026-04-23 Status: Active Supersedes: docs/roadmap.md

---

Build a platform to demonstrate, show its power, and generate quick cash. Two revenue tracks run in parallel:

• Product sale — we build your AI operations team, lump sum. CCOS2 is the proof.
• Tech license — here's the platform, deploy it yourself, revenue share.

The demo sequence proves both: small → medium → big.

1. Assistant — personal AI, daily brief, task management, scheduled routines 2. Marketing — content calendar, BDR outreach, watch it run in real time 3. Dev project — big project decomposes into phases → tasks → execution, watch a phase complete end-to-end

---

The platform must handle big projects today. A project decomposes into phases. Phases decompose into tasks. Tasks have dependencies. Claire orchestrates at every level.

What gets built:

• workflow_instances gains parent_id uuid references workflow_instances(id)
• Three templates: project, phase, dev_task
• workflow_steps gains depends_on uuid[] — task dependencies

• BA (Priya) does the intellectual decomposition — receives the project brief, produces the phase breakdown with specs, receives each phase and produces the task list. The thinking work.
• Claire orchestrates state — spawns instances, routes work to the right role, advances the workflow when steps complete, holds at exec gates. The plumbing work.

Workflow hierarchy: `` Project instance → Claire posts to roles.ba → BA (Priya) produces: phase list, each phase named + scoped + specced → Exec gate: Justin reviews phase breakdown before anything spawns → Claire spawns N Phase instances, posts each to roles.ba → BA (Priya) produces: task list for the phase, each task specced with dependencies → Exec gate (optional): Justin reviews task breakdown → Claire spawns M Task instances respecting dependency order → dev_task: spec → code → review (existing) → All tasks in phase complete → phase done → Claire reports up → All phases complete → project done ``

• Receives BA output at each level, materializes child instances
• Dependency gate: task not posted to queue until all depends_on tasks are complete
• Holds at exec gates — workflow does not advance until approval recorded in Exec Inbox

• Project phase breakdown complete → exec review before phases spawn
• Phase task breakdown complete → exec review before tasks spawn (optional per template)
• Gates surface in Exec Inbox (Priority 2)

• Any agent can post a message to any other agent via agents.* topics — informational only
• Any agent can post to roles.exec to reach Justin directly — this is messaging, not task routing
• Hard guardrail: only Claire posts to peer role topics (roles.ba, roles.developer, etc.) to route tasks
• Each agent reads its own agents.* inbox at session start — messages injected as context alongside memory
• agents.justin = lightweight channel for things worth knowing but not decisions needed
• Messages that require a decision surface in Exec Inbox instead

The Cockpit is the operational view for running projects. Not the full Boardroom yet — just the project and agent monitoring surface needed to run real work.

Four sections:

• Per-project Gantt view: phases as bars, tasks within phases
• Status color: pending / in-progress / blocked / complete
• Target dates on phases (set at project spawn, tracked against actual)
• Dependency visualization — blocked tasks show what they're waiting on

• Per-project, per-phase, per-task: token usage × model rate = dollar cost
• Running total as work progresses
• Cost pulled from task.complete payloads (model, token_count, duration_seconds)

• Per-project breakdown view: how the project was decomposed by BA
• Phase list with status, owner, target date
• Each phase links to its artifacts: spec doc, design doc, roadmap, HCS definition snapshot
• Task list within each phase — dependencies visualized, what's blocking what
• The full paper trail from brief → phases → tasks, all in one place

• Everything needing Justin's attention: gates, blocked workflows, flagged items, escalations from agents
• Priority-sorted. Oldest blocking item at top.
• Each item: context, what's blocked, what decision is needed

• All running agents: role, current task, last activity timestamp
• Idle agents shown — confirms platform is live
• Quick view of what each agent completed today

---

Everything beyond the immediate priorities. Ordered by logical sequence.

Time-triggered and event-triggered workflows. Needed before Assistant and Marketing demos.

• scheduled_triggers table: id, name, workflow_type, schedule (cron), payload, client_id, active
• pg_cron job fires due triggers → posts work items to queue
• Event triggers: event_triggers table, conditions evaluated on Supabase DB events or webhook inbound
• Calendar data model: every workflow instance gets scheduled_at, started_at, completed_at

Full company OS. Extends the Cockpit into a strategic operating surface.

• Now — live operations feed. BDR sent LinkedIn at 8:30, 8:47, 9:02. Who she sent to. What assistant is doing.
• Strategy — declared goals and priorities. Lives in BezelBrain. Atlas owns it. Updated by talking to Atlas.
• Measuring Strategy — KPIs tracking against strategic goals. Gap detection automated.
• Gaps — delta between strategy and reality. Surfaced automatically.
• Planned — calendar view of scheduled workflows. Content calendar. Upcoming milestones.
• Projects — the Cockpit project view, embedded.
• Finance — revenue, expenses, runway, invoices. Stripe + Mercury.
• Pipeline — deals in flight, stage, expected close. HubSpot.
• Team — agent roster. Who's running, idle, what they accomplished today.
• Exec Queue — promoted from Cockpit, cross-domain. All decisions needed.
• Intelligence Digest — what the system learned this week. BezelBrain contributions, Sentinel signals, patterns.

On naming: The Boardroom, not Cockpit. This is where the company runs from.

Agent definitions + workflow templates for each business function.

• Assistant — daily brief, task management, calendar awareness, morning routine
• Marketing / BDR — content creation, LinkedIn outreach, campaign workflows, content calendar
• Sales — HubSpot pipeline management, deal follow-ups, meeting prep
• Finance — Stripe/Mercury reporting, invoice management, weekly revenue summary
• Sentinel — competitive monitoring, signals posted to BezelBrain
• API Scout — discovers new APIs, populates API Library

Intent expression → workflow execution. Self-healing when APIs break.

Full spec: docs/specs/2026-04-23-semantic-intent-layer.md

• Intent Resolver agent (roles.intent)
• API Library (pgvector semantic search over available integrations)
• Self-healing loop: task.blocked → escalate to resolver → re-resolve → continue
• Every heal writes a learning to Agent Brain

RedKey as a licensable product.

• Onboarding flow: new deployment in < 30 minutes via scripts
• Usage billing: per-client token cost tracking → Stripe billing
• Agent marketplace: browse and deploy community agent definitions
• Customer portal: client-facing view of their own workflows

---

Small — Assistant: Open The Boardroom. Show Now section. Morning brief running. Daily tasks queued. "This is your AI chief of staff."

Medium — Marketing: Show calendar. BDR messages scheduled. Watch one fire at exactly 8:30 — LinkedIn outreach to a named prospect. Content post went live at 9am. "This is your AI marketing and sales team."

Big — Dev project: Submit a real project brief. Priya (BA) reads it, breaks it into 4 phases with specs. Exec Inbox fires — review the breakdown, approve. Claire spawns Phase 1. Priya specs the tasks, dependencies mapped. Agents claim and work. Open Structure view — see the full paper trail: brief → phases → tasks → specs linked. "This is your AI engineering team taking a project from brief to shipped."

---

The underlying infrastructure is not referenced in any UI, demo, or external document. All external language uses: "durable execution," "distributed agents," "guaranteed delivery," "permanent audit trail." The specific technology is a strategic advantage revealed at the right moment — not before.

---