Make it feel simple.
Your job is to show three moments where AI improves judgment, not to teach every technical detail. The Lenny’s engine is the one deliberate exception: it proves this is a purpose-built tool, not a clever prompt.
Five-minute preflight.
If any item fails, use the static screenshots or speak to the artifact. Never debug in front of the room.
#start and refresh.roadmap.html in a backup tab.team-pulse.html in a backup tab.Ten minutes by default. Five if compressed.
The right-arrow key advances the story. Within the roadmap, click Walk. No other exploration is required.
10-minute path
Enough space for the engine and one interaction per workflow.
One story, three decision moments, invented data.
One conviction, two questions, reveal the revised decision.
Build once, retrieve tension, trace one objection.
Click Walk; show four planes moving together.
Open ownership evidence; state human review.
Restate ownership; propose one bounded pilot.
Take one question or land the final sentence again.
5-minute path
Keep the engine; remove supporting detail.
Arc + invented-data guardrail.
Read the original; reveal the conditional strategy.
Archive → evidence → disagreement → citation.
Click Walk; name the four synchronized planes.
Open the ownership evidence; one sentence on governance.
Human ownership + one bounded pilot.
Say this, then move.
The scripts are short on purpose. Let the artifacts prove the point.
Ideate · from conviction to falsifiable choice
The engine · a new kind of tool
Align · from memo to shared future
Deliver · reconcile two signals
Close · one responsible next step
Credible, precise, bank-safe.
These swaps preserve the ambition without implying unapproved production use or autonomous decision-making.
Short answers, no defensiveness.
Answer the question asked. Return to the human boundary if the conversation becomes abstract.
Is this already in production?
No. These are tested personal prototypes using invented data. They demonstrate the workflow. Production use requires approved tools, approved sources, security review, and the relevant control owner.
Why not just use a general chatbot?
A general chatbot can draft. The custom engine prepares a domain-specific evidence system, retrieves opposing support, returns source traces, and has an explicit weak-evidence fallback. The quality comes from the context architecture.
Are those experts really saying that?
The named critique is AI synthesis grounded in a dated documented viewpoint—not a direct quote or simulated participation. The source trail is visible, and a name is withheld if retrieval cannot support it.
Could this work with internal knowledge?
Yes, if the source is approved and the use is appropriate. The same pattern can organize internal research, decisions, policies, or operating knowledge without moving it outside the approved environment.
Does Team Pulse evaluate employees?
No. It is team- and system-level only. If it is used to score people, responses become less honest and the instrument stops being useful.
What happens when the AI is wrong?
Important claims must link to evidence and carry a status: verified, directional, or needs review. A human can reject or relabel them. Thin evidence produces an insufficient-data result—not confident filler.
Why a clickable mock instead of a deck?
Because the room can see customer experience, platform work, value, and controls change together. It turns an abstract roadmap into one shared object people can test.
What would you pilot first?
One bounded workflow with read-only inputs, a decision owner, a human reviewer, and explicit success and stop criteria. Napkin → Narrative is the lowest-data starting point.