# Judgment, accelerated — enhanced demo brief

## The point

This is not a demonstration of AI replacing product judgment. It shows how a leader can use AI at three decision moments:

1. **Ideate:** challenge an assumption before committing.
2. **Align:** make a strategy tangible before asking a room to agree.
3. **Deliver:** reconcile team experience with the work record before intervening.

AI handles high-volume interrogation and synthesis. The leader owns the evidence, the decision, and the consequences.

## The story

All data and organizations in the demonstration are invented.

FieldFlow distributes embedded payments through vertical software. A broad integration strategy produced 11% payment attach against a 30% planning assumption. The leader must decide whether the gap is a discovery problem or a workflow-fit problem.

Napkin → Narrative stress-tests that conviction. The resulting six-week decision gate focuses the next learning cycle on field services. The Playable Roadmap turns that choice into a staged product experience for Northside HVAC. Team Pulse then finds that an unresolved integration-ownership boundary is repeatedly slowing the same roadmap.

The three questions are therefore connected:

- Are we making the right bet?
- Can everyone see the same future?
- Can the organization deliver it?

## Workflow 01 · Napkin → Narrative

**Input:** a rough conviction, context, evidence, and the decision at stake.  
**AI’s unique job:** ask the questions that expose the weak assumption, then retrieve a relevant defender, critic, and decision condition from a structured expert corpus.  
**Human decision:** what changes and what test will settle the argument.

The original conviction—“go broad now, fix attach later”—becomes a falsifiable strategy: isolate the cause of weak attach for six weeks; go deep if workflow gaps dominate; stay broad only if discovery is the primary issue and shared integration materially fixes it.

### The Lenny’s analysis engine

The engine is a purpose-built reasoning system, not a simulated panel prompt.

- A first AI pass extracted frameworks, claims, contexts, caveats, and contrarian positions from a frozen subscriber-archive snapshot.
- A second pass created cross-corpus synthesis and 4,367 explicit pairs of opposing views.
- At query time, the tool parses the draft’s claim, retrieves by meaning, prefers useful disagreement, and returns a defender, critic, condition, source, and date.
- Named viewpoints appear only when retrieved evidence supports them. Generated dialogue is labeled AI synthesis, never a quote.
- Each claim is triaged as **verified**, **directional**, or **needs human review**.
- If retrieval is unavailable, the workflow uses generic unlabeled expert lenses. It never fabricates named authority.

Frozen working snapshot: 624 distinct posts and episodes, 15,669 typed evidence fragments, 4,367 opposing-view pairs, and 345 guest profiles.

This prototype publishes derived analysis only. It does not redistribute the raw Lenny’s Newsletter / Podcast archive. Source credit: Lenny Rachitsky. Not affiliated with or endorsed by Lenny’s Newsletter.

## Workflow 02 · The Playable Roadmap

**Input:** the chosen strategy, north star, stage model, and constraints.  
**AI’s unique job:** translate every stage across four synchronized planes—customer experience, platform capabilities, business value, and controls.  
**Human decision:** what is coherent, feasible, and worth funding.

The example moves from:

- **Crawl:** turn an accepted estimate into a deposit request.
- **Walk:** make milestone payment a repeatable part of the job.
- **Run:** provide an explainable view of job cash inside the field-service product.

The artifact is one self-contained HTML file. It has no backend or external requests and works offline.

## Workflow 03 · Team Pulse

**Input:** an anonymous aggregate pulse, a read-only scoped work-item export, and the leadership decision context.  
**AI’s unique job:** group ticket prose into recurring system themes, link every theme to its evidence, and compare the record with what the team raised.  
**Human decision:** which hypothesis to investigate and what system to change.

In the synthetic example, the team ranks integration ownership fourth. The work record finds it in 22 tickets, the second-most common pattern, with a directional median 6.2 days of added wait. The conclusion is not that the team was wrong. The hypothesis is that the team normalized a chronic boundary.

Team Pulse is team- and system-level only. It must never name, rank, or score individuals. AI proposes themes; a human reviews every linked claim before leadership sees it.

## Recommended demonstration

Use the pre-made artifacts. Do not generate the narrative, authenticate, or connect live data during the meeting.

The default presentation is ten minutes, including two minutes on the analysis engine. A five-minute fallback preserves the engine but removes supporting detail. See `PRESENTER-GUIDE.md` for the exact click path and talk track.

## Responsible next step

Run one bounded pilot in an approved environment with:

- sanitized or synthetic inputs;
- read-only, minimum-necessary access;
- one named decision owner and one human reviewer;
- source traces for important claims; and
- success, stop, and escalation criteria agreed in advance.

Napkin → Narrative is the lowest-data starting point. It can prove whether the interrogation and evidence-retrieval pattern improves decision quality before any operational connection is considered.