The Second Station

Full text is archived in wiki/attachments/x-captures-2026-11-09/capture.json (key /saranormous/status/2119863974271545449).

Key claims

• The factory parable: a robot takes the slowest station from ~100 to 500 parts an hour, but the next station tops out at 125. The 25% gain gets you in the door; to be paid on the line you may have to help the plant buy the next machine or redesign the flow.
• Pharma has the longest line: discovery may be faster, but site activation, enrollment and regulatory review are the later stations, and “approved drugs in patient hands” needs all of them moving.
• SaaS sold seats into existing processes. 2010s point solutions made one person better at their piece and left “the rest is your problem”; systems of record were sticky enough.
• The diffusion gap is about returns. The trillion-dollar buildout assumes buyers pay for capability, but they pay at scale only for returns. Watney’s datacenter customers watch days-until-serving-tokens, so speeding one task makes them want the next one faster too.
• AI is metered and visible. A seat could be bought on faith; an agent bill is big enough for the CFO to review. “The mania of tokenmaxxing is already over.”
• Organizations have to agree to change. Work waits on other work, and the faster team creates unwanted work for the next one. Her Greylock homepage-review story: partners discovered shared preferences only through iteration and negotiation.
• Someone has to run that negotiation: of the five jobs now called “forward-deployed engineer,” the key one may be an outcomes owner who coaxes the P&L owner and security owner along. Agents will eventually need an “OQ” (organizational quotient) to win over “Sally from finance.”
• Pricing will trend to outcomes, despite complexity (baselines, attribution, true-ups, caps) that gives investors an allergic reaction. Sierra charges for business outcomes and passed $400M ARR in October 2026; Cognition refunds the gap, up to $10M, if Devin’s measured output is worth less than the price.
• Consolidation and insourcing: you can’t reorganize a business with a hundred strategic partners. Citi, with 14,000 engineers on Devin, cut tech vendor spend 30% in a year. The extreme form is buying the business: Sequence, Ithaca, Thrive Holdings, Long Lake. Sequence cut BankSouth’s loan origination from nine stages to three and roughly halved application-to-close.
• Close: if your promised return depends on the second station, make the second station your business too.

Connections

• Outcome-Based Pricing · Forward Deployed Engineering · AI Rollups · Services as Software
• Same-day companion: Own Your Means of Production (David Hsu), which argues the same endpoint from the labor-share side.