Focus

It breaks by addition. Nobody decides to lose focus. They keep everything, one reasonable yes at a time, and the portfolio quietly becomes the strategy. The anchor case is Jobs in 1997: Apple was 90 days from insolvency by his own account, carrying roughly 350 SKUs, and his documented tell from the product review meetings was a question nobody in the room could answer: *which machine do I tell my friends to buy?* The fix was a 2x2 grid on a whiteboard: consumer and pro, desktop and portable, one great product per quadrant. He killed roughly 70% of the product line, revenue attached and all. Profitable within a year. iMac within two. The metric that survives from that story: focus is measured in what you stopped, never in what you kept.

The second break is timing, and Alex Hormozi named it: the red dress. The seductive new opportunity doesn't appear when you're failing; it appears right after traction, when you finally have something worth diluting. New opportunities are drawn to fresh momentum, yours. Hormozi ran the rule on himself, selling Gym Launch to concentrate on one firm, which is what gives the rule its receipts.

1. Name the phase, because the phase sets the rule. The apparent contradiction in the operator record resolves here. Pieter Levels ran 70+ launches and killed nearly all of them fast, then concentrated hard when PhotoAI worked. Both halves were correct. Search phase: many small bets, killed quickly and cheaply. Winner phase: concentration is the alpha. Diversify searches, concentrate winners. State your phase in writing before arguing about any single project, because most focus fights are two people in different phases using the same words.

2. Run the constraint test on anything new. Before a new bet gets oxygen: what is the current constraint of the core business, and does this new thing solve it or dodge it? If the shiny project and the scary constraint arrived in the same quarter, assume dodge until proven otherwise, the same way you'd check a pivot against the morale graph.

3. Apply the punchcard. Buffett's frame, converted to attention math: imagine a card with 20 punches for your whole investing life, and watch how quality goes up when shots are scarce. Your version is smaller: two or three real strategic bets a year at full power. Every keep spends a punch. Every punch spent on a sub-scale side thing is unavailable to the winner, and the winner is where returns compound. One boundary so this doesn't get misapplied: the punchcard governs strategic bets and founder attention. Iteration inside the winner wants high shot volume; don't punchcard your A/B tests.

4. Run the kill test on everything already running. Jobs's question, generalized: for each live project, would you recommend it out loud to someone you respect, and would you start it today knowing what you know? Two nos is a kill. Including, especially, when revenue is attached. Sub-scale working things are the hardest kills in the record, defended by real customers and sunk pride, and "working a little" is the most expensive state a project can occupy: too alive to kill, too small to matter, billing founder attention monthly.

5. Execute the kill like a call, because it is one. Write the memo: what dies, why, and, critically, where the freed capacity goes. A kill without a named beneficiary is just loss. Price the political cost honestly: whose project it was, who's embarrassed, which customers feel it, and be generous on the way out, because survivors and market are both watching. Then premortem the kill itself: it's 18 months from now and killing this was the mistake. What's the mechanism? If the story names real option value with a date attached, set a tripwire instead of a funeral. If it's generic mourning, proceed.

• Diversifying the same quarter the core constraint got scary. Timing is the tell.
• The red dress after traction.
• Keeping a sub-scale thing because it works a little.
• Portfolio as identity. "We're a lab" is sometimes true and sometimes cover for never concentrating after signal. The phase test tells you which.
• The half-kill. Maintenance mode is a kill you're still paying for.
• Killing by silence. Unannounced deaths breed zombie work: someone's still patching the thing you decided was dead.

Founder, five years in: core SaaS at $60k MRR growing 4% monthly, a $9k MRR legacy WordPress plugin, a paid community at $3k, and a new AI tool idea "to diversify revenue." Phase: the SaaS is a winner, so the rule is concentrate. And the AI idea arrived two weeks after the SaaS's key channel got 30% more expensive, which answers the constraint test before it's even run: the real bottleneck is a new acquisition channel, and the AI tool dodges it exactly. Punchcard: two punches this year, and the channel problem deserves both. Kill test: he wouldn't start the plugin or the community today, and he's stopped recommending the plugin out loud. Both die, despite $12k of combined MRR, the hardest yes of the session. Mechanics: a migration path and partner referral for plugin customers, the community handed to its two most active members, memos written, political cost priced at two awkward emails and a forum thread. The premortem surfaces one real risk (an acquirer who'd once wanted the plugin bundled), so a tripwire holds it with a date instead of reopening the kill.

Freed capacity: 15 founder hours a week, pointed at the channel. Two quarters later the SaaS is at $78k MRR on a second working channel, and the friend-test answer takes one word.