Every failed first version in this course dies the same death: it tries to be the platform on day one. The cure is the wedge — the narrow, unmistakably useful first job that gets one person into the product and gives you real use to learn from.

What a wedge is

A wedge is the smallest job that is still completely useful: one person, one starting input, one finished output, one reason to return. Small enough to build, test, and support in weeks — valuable enough that someone changes behavior for it.

Compare:

Vague: "Build a social platform for investors."
Wedge: "Let one investor create a source-linked research brief for one public company and save it privately."

The vague version needs accounts, feeds, follows, moderation, notifications, and a network that does not exist. The wedge needs search, a brief template with source links, and a save button. One is a pitch deck. The other ships.

Run our idea pool through the same narrowing:

• House-price brief: not "a real-estate intelligence platform" — wedge is "paste one address, get five sourced comps in one brief."
• Used-car shortlist: not "an auto marketplace" — wedge is "paste ten listing links, get a ranked five with reject reasons."
• Trip filter: not "a travel planner" — wedge is "enter dates + budget + two constraints, get three viable options."
• Support triage: not "an AI helpdesk suite" — wedge is "route Monday's 100 tickets to the right agent with a draft reply each."
• Salary bands: not "a compensation platform" — wedge is "enter one role + city, get a sourced band table."
• Filing brief: not "legal ops" — wedge is "drop in one docket, get deadlines + owners."

Each wedge fits in one sentence because it fits in one sitting of real work.

The seven-test wedge test

A candidate wedge must pass all seven. Grade honestly — most ideas fail two:

1. Specific user. Can you name a real person, not a segment? "Priya, renting in Austin, offer due Friday" passes. "Homebuyers" fails. 2. Existing painful behavior. Are they already doing it manually — screenshots, spreadsheets, late-night scrolling, Monday triage hell? If there is no workaround, there is no pain. 3. Clear starting input. What do they hand you to begin? A ticker, an address, a ticket export, a role + city. "Tell us your dreams" is not an input. 4. Visible finished output. What do they hold when it works? A brief, a shortlist, a routed queue, a band table. If you cannot screenshot the finish, neither can they. 5. A reason to return. New listings, new tickers, next Monday's queue, next hire. One-shot wonders do not sustain products. 6. Small enough to build, test, and support. Could a controlled Part X slice deliver it with one flow across UI, rules, data, and evidence? If it needs three teams, it is not a wedge. 7. No dependency on an unbuilt network effect. It must work for the first user alone. A used-car shortlist works with zero dealers. A marketplace does not.

The trip filter wedge passes: Priya exists, she already has twelve tabs open, the input is dates + budget, the output is three options, she plans another trip next quarter, one flow builds it, and no network is required. "A social platform for travelers" fails tests 1, 4, 6, and 7 simultaneously.

The Part X slice, in product language

You already learned the vertical slice in Part X: build one thin journey across all layers rather than one fat layer with no journey. In product language that means: complete one user journey across interface, rules, data, and evidence — not a beautiful empty frontend, not an enormous backend waiting for users.

For the salary-bands wedge, the slice is: enter role + city → retrieve approved salary sources → render band table with citations → save the lookup. Interface (the form and table), rules (which sources count, how bands compute), data (the source set), evidence (links on every row). Skip team workspaces, alerts, and API access. Those are horizontal expansions for later slices, once this one proves someone trusts the table.

Four traps that look like wedges

• Copying a large category. "A better CRM" or "a ZillowEsperanto clone" is not a wedge; it is a ten-year company described in three words. Wedges pick the one sub-job the giant does clumsily — comparables for one street, triage for one queue.
• Building every user type. Buyers plus sellers plus agents plus lenders on day one means four onboards, four permission models, four support burdens. Pick the buyer. Let the others wait in the non-goals list.
• Feed before objects. A feed of "insights," "deals," or "events" with no useful underlying objects is an empty river. The used-car feed means nothing until individual verdicts are trustworthy. Objects first, feed later.
• "AI-powered" as the wedge. "AI-powered home search" describes the engine, not the job. Nobody returns for the engine. They return for the five sourced comps that survived the offer negotiation.

Exercise: three wedges ranked → WEDGE.md

Write three candidate wedges for one idea, then score each 1–3 on four axes: pain intensity, speed to proof, data/permission risk (lower is better), and ability to reach a first user. Lowest total wins — then sanity-check against the seven tests.

Example for the filing brief:

Then write WEDGE.md:

# Wedge

- Chosen job: One paralegal turns one docket export into a deadline + owner table.
- First user: Dana at our firm, who does this manually every Friday.
- Success: Dana completes Friday's docket in under 15 minutes with zero missed deadlines for 4 weeks.
- Exclusions: No client portal, no alerts engine, no multi-matter search, no billing integration.
- Input/output: CSV docket in → deadline table with source page refs out.
- Return loop: Every Friday filing batch.

Finish line: a WEDGE.md with one chosen job, a measurable success condition, explicit exclusions, and the named first user you will show it to.

Verify fast: ask, "can I demo this to Dana with her real Friday file within three weeks?" If not, narrow further. Common failure: picking the wedge you most want to build rather than the one Dana most needs done.

Check your understanding

1. Define a wedge in one sentence. 2. Which of the seven tests does "a social platform for investors" fail? 3. What does a vertical slice mean in product language? 4. Why is "AI-powered" not a wedge?

Next

The wedge tells you where to enter. Next, Lesson 54.5 turns the wedge into a first version that is also a learning instrument — one flow, honest states, trust boundaries, and a V1-SCOPE.md ready for roadmaps and task cards.