Meeting Whiteboard / 2026-04-20 / built from the Otter transcript

LLM Wiki For The Team

The monologue keeps circling one center of gravity: every meeting should update a persistent team memory that tracks tasks, ideas, gripes, vision, and reusable patterns. Everything else looks like a surface area, wedge, or spinout around that engine.

Committed Spine Parallel Tracks Open Questions Scope Cuts Compounding Overlap

core thesis

Every meeting becomes structured company memory

Action items should live both inside meeting notes and in the durable team task layer.
Ideas should land both in context and in a broader company idea base.
Disagreements and recurring frictions become explicit, queryable institutional memory.
Vision docs should update continuously instead of becoming stale static docs.

Build now Explore / sequence later Open question / risk Scope decision Shared infrastructure

Track Map

Product branches orbiting the same engine

Build now Compounds

1. Slack Bot For The Team

This reads as the first obvious surface area: it exposes the wiki to the place teams already live, turns memory into retrieval, and makes the meeting system feel useful between meetings.

Judgment: first shipping surface, not a side quest.

Explore next Channel fork

2. Group Context Bot: iMessage / WhatsApp / SMS

Very compelling because it carries shared memory into messy real-world coordination. Also where privacy, terms of service, and channel fragmentation get hardest fastest.

Judgment: high upside, but only after the core memory loop feels real.

Build now Workflow hinge

3. Meeting Ingest + Real-Time Outliner

This is the actual choke point. If ingestion from Otter / Gemini / Granola / Zoom is annoying, the whole vision becomes ceremony. The product lives or dies on low-friction update flow.

Judgment: integrate first, custom note app second.

Exploring Great demo

4. Ambient Meeting Display

Live fact checks, relevant context, branch dashboards, build-this-now prompts. Excellent demo surface and a natural “AI sees the meeting” manifestation of the system.

Judgment: feature or showcase first, standalone product only if usage proves it.

Dogfood Agent cockpit

5. Note Stream Mobile App

“Slack for myself” plus thread-level agents. Strong as an internal command center and note substrate. Less convincing as the first external wedge than the team-memory loop.

Judgment: valuable dogfood and content generator, not the clearest initial market story.

Roadmap risk Infra reuse

6. Better GPT

Coding-agent-backed GPT on private infrastructure is absolutely valuable. It is also broad enough to become its own company and steal attention from the more opinionated meeting-memory thesis.

Judgment: use internally, maybe spin out later, but do not let it blur the main story.

Center Of Gravity

Why the wiki matters

The transcript is not pitching “meeting notes.” It is pitching a living team memory system.
The important difference is schema: tasks, ideas, gripes, patterns, and vision docs all update distinct layers.
That means the real product is persistent structured memory plus retrieval plus permissions.
Once that exists, different interfaces can be added without rebuilding the substrate every time.

If something does not deepen the team-memory loop, it should justify itself as a distribution channel or a demo, not as a parallel worldview.

Most Important Fork

Group assistant: two product shapes

The messaging branch is not one thing. It splits into two very different products with different risk profiles.

A. Shared bot inside the group

A real participant in iMessage / WhatsApp / SMS / email threads. More magical, more obvious value in scheduling and memory, but higher friction for approvals, terms of service, adoption, and trust.

B. Personal-side agent only on your end

Watches your own message history and acts like an augmented assistant without joining the thread. Lower distribution magic, but much cleaner technically and safer to ship first.

My bias: if messaging becomes the next wedge, start with the personal-side agent, then graduate to a shared bot once privacy scopes and sharing approvals actually work.

Scope Cuts

Decisions that should not stay fuzzy

Market

B2B vs B2C

The monologue contains both. The cleanest story right now is B2B team memory first. Consumer messaging agents can follow once the memory substrate is battle tested.

Surface

Slack first, iMessage second

Slack is the easiest way to surface durable team memory. iMessage is strategically important in the US, but it is not the easiest first proving ground.

Ingestion

Integrate first, own the full recorder later

Existing note tools already exist. The risk is spending months rebuilding note capture before proving the higher-order memory system.

Open

Realtime assist vs post-meeting commitments

The live ambient display is exciting, but the enduring value may come from what the system remembers and updates after the meeting ends.

Detail

Wiki hub vs graph

Important architecturally, but the transcript itself labels this as a lower-order detail. Good reason not to let it block the product wedge.

Privacy

Approval gating is not optional

Scheduling, availability, and selective sharing are part of the core value proposition. Cross-context privacy scopes are a product feature, not just plumbing.

Synergy Map

Why these branches feel so adjacent

They reuse the same stack. That is good news and bad news: good because infrastructure compounds, bad because everything feels temptingly “almost the same” and can blur into roadmap sprawl.

1. Ingest

Meetings, chats, notes, docs, and written history all need to land in a consistent stream.

2. Structure

Extract tasks, ideas, people, conflicts, plans, decisions, and recurring patterns.

3. Memory

Persist canonical pages or graph entities that update over time instead of duplicating raw notes.

4. Permissioning

Scope what can be shown, shared, approved, or kept private across people and contexts.

5. Surfaces

Slack, ambient dashboards, messaging bots, mobile note tools, and coding-agent interfaces.

Roadmap Posture

Compounders vs sidecars vs distractions

Compounds the main thesis

Meeting ingest from existing tools
Structured wiki / team memory updates
Slack bot as the everyday retrieval surface
Selective privacy approval flows

Sidecar or showcase layer

Ambient dashboard for live branches and prompts
Note Stream as a dogfood capture surface
Messaging agent after the memory spine works

Likely distraction if pursued too early

Rebuilding the whole note-taking stack before proving memory value
Launching on every messaging channel at once
Letting Better GPT become the main product story

Impact × Effort Read

Priority matrix

Quick wins

Big bets

Useful but secondary

Tempting detours

High impact

Lower impact

Lower effort

Higher effort

Slack bot

Best proof that the wiki is useful outside the meeting. Should ship early.

Meeting ingest + structure

The least flashy item and the most existential one. This is the engine.

Group context bot

Very large upside, but distribution, privacy, and channel complexity all pile up fast.

Note Stream

Useful dogfood and agent cockpit. Good support system, not the clearest flagship.

Ambient display

Great demo layer and possibly a product, but strongest once it rides on proven memory.

Better GPT

Powerful, synergistic, and probably off-axis from the meeting-memory thesis if externalized too soon.

Opinionated Build Order

How I would sequence the possibility space

Make meeting outputs reliable

Ingest Otter and adjacent note sources cleanly, extract structured artifacts, and prove that a meeting really updates team memory with minimal human cleanup.

Surface the memory in Slack

Ship the bot that answers questions, recalls prior decisions, and shows that the memory graph actually helps between meetings instead of just after them.

Add privacy-scoped actions

Approvals, scoped sharing, and selective access become the bridge from passive memory to active coordination.

Expand into messaging

Start with the simpler personal-side agent if needed, then move toward shared-thread assistants when trust and infrastructure catch up.

Productize sidecars only after they earn it

Ambient dashboards, Note Stream, and Better GPT should either amplify the core product or be clearly spun out, not remain half-integrated ambition sinks.