Whentor Product Brief

One-Line Pitch

Whentor is the follow-through layer for AI conversations.

It turns important AI conversations into a correctable receipt: what was decided, why, what action was committed, what would falsify it, where it went, and what happened after.

Core Thesis

AI advice is abundant. Follow-through is scarce.

The winning layer is not another chatbot, mentor persona, or AI CEO. It is the layer that sits between:

1. AI advice
2. human decision
3. committed next action
4. external handoff
5. observed outcome

Whentor should own that loop.

What Whentor Is

Whentor projects the relevant evidence from a bounded AI conversation into a draft receipt. The user corrects or confirms it, then Whentor records the handoff and reported outcome with provenance intact.

The product answers:

What did you decide, what will you do next, and what happened after?

What Whentor Is Not

Whentor is not an AI therapist, generic mentor chatbot, Character.ai for productivity, Asana replacement, AI CEO, model provider, or generic RAG app.

Personas are packaging. They are not the moat.

Product Claim

Advice dies when it stays in chat.

Whentor turns hard conversations into committed action and outcome memory.

Why In-Chat Apps Matter

Whentor should not start as another standalone SaaS dashboard.

The strategic surface is the in-chat app layer: MCP Apps or equivalent host-native app extensions that render interactive UI directly inside AI sessions.

Source: https://modelcontextprotocol.io/extensions/apps/overview

MCP Apps matter because they provide:

1. Context preservation: the follow-through UI lives inside the conversation where the decision, reasoning, and user intent already exist.
2. Bidirectional data flow: the app can call MCP server tools, while the host can push fresh results back into the app.
3. Host capability routing: the app can delegate actions to the host, which can invoke tools the user already connected, subject to consent.
4. Sandboxed trust boundary: apps run inside isolated iframes and communicate through postMessage / JSON-RPC.
5. Multi-step workflow support: the app can show corrections, commitments, handoffs, and outcomes without another tab.

Surface vs Product

Codex, ChatGPT, Claude, and MCP Apps are first capture surfaces, not the product boundary.

Whentor starts where AI advice happens, then hands work to where execution already lives:

AI chat
-> Whentor Follow-Through Receipt
-> Asana / Linear / Calendar / Slack / email
-> outcome trace back to Whentor

Whentor does not compete with task managers, calendars, or meeting assistants. Those systems own execution. Whentor owns the follow-through record: what the human chose, what commitment followed, where it was handed off, and what happened after.

Strategic Flow

AI session contains context
-> Whentor MCP App renders follow-through UI
-> RLM opens exact evidence and drafts the receipt
-> user confirms the commitment
-> host execution tools own the downstream work
-> user reports what happened
-> Whentor records the outcome trace

Moat

MCP itself is not the moat. Anyone can build an MCP App.

The moat is the action-adjacent decision telemetry captured because Whentor is embedded at the follow-through moment:

user intent
-> AI conversation
-> decision captured
-> action committed
-> handoff reference
-> reported outcome

This is more valuable than generic chat memory because it captures the moment where advice becomes action.

Over time, Whentor preserves:

• what evidence supported a decision
• what action the user confirmed
• how the user corrected the projection
• where the commitment was handed off
• what the user reported afterward

Initial Wedge

Start with founder-critical decisions where advice often dies after the AI session:

• investor updates
• cofounder conversations
• hiring/firing decisions
• customer escalations
• advisor asks
• pricing changes
• killing or committing to a feature

These are costly moments where preserving judgment and follow-through matters.

First Product

Decision Follow-Through.

Input:

• current question
• bounded conversation trace selected for that question

Output:

• draft decision
• evidenced reason the direction changed
• accepted commitment
• falsifier
• exact cited trace moments
• unresolved material uncertainty

Example:

Conversation:
"I think I need to confront my cofounder about missed deadlines, but I keep avoiding it."

Whentor:
- Decision: talk to cofounder directly
- Commitment: schedule a direct conversation by Friday
- Would change this: evidence that a written escalation is required first
- Uncertainty: the trace does not establish whether the cofounder agreed to meet

Decision Blueprints

Decision blueprints are implementation, not the headline.

They can help structure the follow-through loop, but users buy fewer abandoned decisions, not frameworks.

Personality = voice. Blueprint = repeatable decision procedure. Product = follow-through.

Examples:

• Manager Feedback Blueprint: facts, impact, expectation, support, next step
• Negotiation Blueprint: interests, BATNA, tradeoffs, next ask
• CBT Blueprint: distortion, evidence, reframe, behavior
• Stoic Blueprint: controllable/uncontrollable, virtue, action
• Risk Blueprint: downside, reversibility, second-order effects
• Founder Blueprint: user pain, urgency, retention, willingness to pay

Personas can remain as UX flavor, but the product should not depend on impersonation.

Stanford AI Index 2026 Provenance

Source: Stanford AI Index Report 2026, local file:

/Users/prateek/Downloads/Stanford AI index Report 2026 - Jul. 2026.pdf

Relevant proof points:

1. AI capability is accelerating, but reliability remains jagged. Models can hit advanced benchmarks while still failing simple real-world tasks. This argues against blind AI execution.
2. Responsible AI is lagging capability. The report says AI incidents rose from 233 in 2024 to 362 in 2025, while responsible AI benchmark reporting remains spotty.
3. Enterprise adoption is high, but agent deployment is still early. Organizational AI adoption reached 88%, but AI agent deployment remains in single digits across nearly all business functions.
4. Productivity gains are strongest in structured, monitorable work, and weaker in judgment-heavy tasks. Whentor keeps humans responsible while making follow-through observable.
5. Public trust is weak around relationships, emotion, and decision-making. This argues against pitching “AI understands emotions.”
6. AI companions are real but contentious. Personas can be interface, not core moat.

External Market Thesis

The X article preview “AI’s Biggest Winners Have the Lowest Margins” argues that low-margin businesses may benefit most from AI because small efficiency gains flow directly into profit.

Source: https://x.com/dkfromdk/status/2075696599242821979?s=20

Implication for Whentor: start where follow-through failure is expensive, not where advice is merely entertaining.

Do not start with “AI CEO.”

Why This Can Work

Levers:

1. Clear gap: people already ask AI for advice, then abandon the next step.
2. In-session surface: MCP Apps or host-native integrations let Whentor sit inside the current AI session.
3. Concrete loop: an explicit capture can become a draft, confirmation, handoff, and outcome.
4. Human agency: Whentor does not make the decision; the user corrects and confirms the receipt.
5. Compounding evidence: the product preserves the trace between judgment, action, and reported outcome.

Validation Questions

Ask users:

1. Show me the last important decision you talked through with AI.
2. What did you do after the chat ended?
3. Did the advice turn into a specific next action?
4. Did anything follow up with you later?
5. What decision do you keep avoiding?
6. What would have made you act sooner?
7. Would you trust a tool to remind you and record the outcome?
8. What outcome would prove this worked?

MVP Scope

Build only this:

Paste conversation or situation
-> project cited evidence into a draft
-> correct or confirm the receipt
-> record the external handoff
-> record the reported outcome

No AI CEO. No broad mentor library. No life graph. No enterprise platform. No custom model. No heavy retrieval unless needed.

Success Metric

Primary:

• percentage of captured receipts that are confirmed and later receive a logged outcome

Secondary:

• decision capture rate
• action commitment rate
• handoff and outcome logging rate
• repeated weekly use
• user-reported avoided delay
• user-reported decision quality
• repeat usage per week

Falsification

Kill or narrow the wedge if:

• users do not bring real decisions
• users like advice but do not want follow-up
• fewer than 30% of captured decisions become explicit next actions
• outcomes cannot be captured
• reminders feel annoying instead of useful
• users can get equivalent value by asking Claude to impersonate an advisor
• users do not return after the first few decision loops

Final Positioning

Whentor turns AI advice into decision follow-through.

It sits inside AI sessions, projects the relevant evidence into a correctable receipt, and preserves the confirmed handoff and reported outcome.

The product is not AI that understands emotions or impersonates a CEO.

The product is follow-through infrastructure for the moment after advice, not a system that claims to make decisions for the user.