Anthropic pasó de $87M a $30.000M de run rate en 27 meses: la infraestructura de agentes con precio por resultado es la escalada empresarial más rápida registrada.
Salesforce shipped three Agentforce price models in 18 months and now runs all three at once: nobody, even the incumbent, knows what the right unit is.
Sierra cobra solo si su agente resuelve el caso y aún así superó $100M de ARR en el trimestre siete: el precio por resultado funciona cuando el agente funciona.
Three high-profile crashes show the same pattern: when the agent cannot truly deliver the result, churn, restatements, and rehires arrive within months.
Cada sucursal necesitaba menos tellers, así que los bancos abrieron 43% más sucursales y el empleo siguió creciendo 2% al año: mano de obra barata expandió el mercado.
Decades after spreadsheets, judgment-heavy accountants still grow 5% while clerical bookkeepers fall 6%: agents will upgrade the senior role and hollow the junior one.
Décadas tras la hoja de cálculo, los contadores de criterio crecen 5% y los empleados contables caen 6%: los agentes elevarán al rol senior y vaciarán al junior.
Outcome pricing is rewriting startup playbooks: Anthropic, Cursor, and Sierra are scaling at unprecedented speed, but thinner margins, pricing chaos, and high-profile collapses prove the model punishes anyone who cannot actually deliver the work.
Key findings
Salesforce shipped three Agentforce pricing models in 18 months: $2/conversation (Oct 2024), $0.10/action via Flex Credits (May 2025), then $125/user/month AELA seats (late 2025) and now runs all three at once. Salesforce
AI SDR 11x raised $74M from a16z and Benchmark but TechCrunch found 70-80% churn, ~$3M real ARR vs ~$14M reported, customer logos used without permission (ZoomInfo, Airtable) and product hallucinations. TechCrunch
AI-native products project ~52% gross margin in 2026 vs 75-85% for traditional SaaS, and Bessemer's fastest 'Supernovas' hitting $100M ARR in ~18 months average just ~25% gross margin. ICONIQ Growth
Anthropic's annualized run rate went from $87M (Jan-24) to $30B (Apr-26); Claude Code alone hit $1B in six months and $2.5B by Feb-26 with >1,000 enterprises spending $1M+/year on Claude. VentureBeat
ATMs were supposed to kill bank tellers but FTE teller employment grew ~2.0% per year after 2000 because each branch needed fewer tellers (20→13) so banks opened 43% more urban branches. IMF Finance & Development
Builder.ai, peak $1.5B valuation and Microsoft-backed, collapsed in May 2025 after audited 2024 revenue was restated from $220M to $55M (75% overstatement) while burning $40M/quarter. Pragmatic Engineer
Cursor (Anysphere) is the fastest-scaling B2B SaaS ever: $1M ARR Dec-2023, $100M Jan-2025, $500M Jun-2025, $1B+ Nov-2025; sold mostly as per-seat-plus-usage, not pure outcome pricing. SaaStr
Decagon (AI customer-service agents) scaled from ~$10M ARR end-2024 to ~$35M by Oct 2025, with valuation jumping from $1.5B (June 2025) to $4.5B (Jan 2026 Series D led by Coatue and Index). Sacra
Intercom Fin charges $0.99 per resolved conversation with a 50-resolution monthly minimum, generated tens of millions in revenue under one year, and resolves over one million customer conversations per week. Stripe
Klarna's AI bot claimed work of 700 FTE and ~$40M profit lift, cutting resolution time from 11 to under 2 minutes; by May 2025 the CEO admitted quality suffered and Klarna started rehiring human agents. Klarna
ICONIQ's Jan-2026 survey of ~300 AI software execs shows outcome-based pricing jumped from 2% in Q2 2025 to 18%, usage-based to 35%, while subscription still holds 58% and 37% plan to change pricing within 12 months. ICONIQ Growth
Sierra (Bret Taylor) reached $100M ARR in 7 quarters and ~$165M by quarter 9 selling outcome-priced customer-service agents, paid only when the AI resolves the case; SoFi case study lifted NPS by 33 points. Cheeky Pint (Stripe)
The argument
Anthropic went from $87M to $30B run rate in 27 months, with Claude Code alone hitting $1B in six months.
Cursor reached $1B ARR in 24 months, the fastest B2B SaaS ramp ever recorded.
Outcome-based pricing jumped from 2% to 18% of AI companies in three quarters, and 37% plan to change pricing again within a year.
Sierra charges only when its agent resolves the case and still hit $100M ARR in seven quarters.
Intercom Fin sells resolutions at $0.99 each and now closes over one million conversations every week.
AI-native gross margins sit near 52%, versus 75-85% for traditional SaaS, and Bessemer's fastest scalers run at just 25%.
11x, Builder.ai, and Klarna show the failure mode: inflated ARR, restated revenue, and rehired humans when the agent cannot deliver.
History rhymes: ATMs grew teller jobs 2% per year and spreadsheets still leave accountants up 5% through 2034.
This research is published in English and Spanish. Ver en español
La investigación detrás de esta presentación
El precio por resultado reescribe el manual de startups: Anthropic, Cursor y Sierra escalan a velocidad inédita, pero márgenes más finos, caos de precios y colapsos sonados muestran que el modelo castiga a quien no entrega el trabajo.
Hallazgos clave
Salesforce lanzó tres modelos de precios para Agentforce en 18 meses: $2/conversación (oct 2024), $0,10/acción con Flex Credits (may 2025) y $125/usuario/mes AELA por asiento (fin 2025), y los corre los tres a la vez. Salesforce
La SDR IA 11x levantó $74M de a16z y Benchmark, pero TechCrunch reportó churn 70-80%, ~$3M de ARR real vs ~$14M reportado, logos usados sin permiso (ZoomInfo, Airtable) y alucinaciones del producto. TechCrunch
Los productos AI-nativos proyectan margen bruto ~52% en 2026 vs 75-85% del SaaS tradicional, y los 'Supernovas' más rápidos de Bessemer que llegan a $100M en ~18 meses promedian ~25%. ICONIQ Growth
El run rate anualizado de Anthropic pasó de $87M (ene-24) a $30.000M (abr-26); Claude Code solo alcanzó $1.000M en seis meses y $2.500M en feb-26 con más de 1.000 empresas gastando $1M+/año en Claude. VentureBeat
Los cajeros automáticos iban a eliminar a los cajeros bancarios pero el empleo FTE creció ~2,0% anual tras 2000 porque cada sucursal necesitaba menos cajeros (20→13) y los bancos abrieron 43% más sucursales urbanas. IMF Finance & Development
Builder.ai, con valoración pico de $1.500M y respaldo de Microsoft, colapsó en mayo 2025 tras una reformulación auditada de ingresos 2024 de $220M a $55M (sobreestimación del 75%) quemando $40M/trimestre. Pragmatic Engineer
Cursor (Anysphere) es el SaaS B2B de escalado más rápido: $1M de ARR en dic-2023, $100M en ene-2025, $500M en jun-2025, $1.000M+ en nov-2025; vendido sobre todo por asiento+uso, no por resultado puro. SaaStr
Decagon (agentes IA de atención) escaló de ~$10M de ARR a fines 2024 a ~$35M en oct 2025, con valoración saltando de $1.500M (jun 2025) a $4.500M (Serie D ene 2026, liderada por Coatue e Index). Sacra
Intercom Fin cobra $0,99 por conversación resuelta con mínimo de 50/mes, generó decenas de millones en menos de un año y resuelve más de un millón de conversaciones semanales. Stripe
El bot IA de Klarna afirmaba el trabajo de 700 FTE y ~$40M de mejora de utilidad, bajando la resolución de 11 a menos de 2 minutos; en mayo 2025 el CEO admitió daño a la calidad y volvió a contratar humanos. Klarna
La encuesta ICONIQ de enero 2026 a ~300 ejecutivos de software con IA muestra el precio por resultado saltando de 2% en Q2 2025 a 18%, el de uso a 35%, suscripción aún 58% y 37% planea cambiar precios en 12 meses. ICONIQ Growth
Sierra (Bret Taylor) alcanzó $100M de ARR en 7 trimestres y ~$165M en el trimestre 9 vendiendo agentes con precio por resultado, cobrando solo si la IA resuelve el caso; el caso SoFi subió NPS en 33 puntos. Cheeky Pint (Stripe)
El argumento
Anthropic went from $87M to $30B run rate in 27 months, with Claude Code alone hitting $1B in six months.
Cursor reached $1B ARR in 24 months, the fastest B2B SaaS ramp ever recorded.
Outcome-based pricing jumped from 2% to 18% of AI companies in three quarters, and 37% plan to change pricing again within a year.
Sierra charges only when its agent resolves the case and still hit $100M ARR in seven quarters.
Intercom Fin sells resolutions at $0.99 each and now closes over one million conversations every week.
AI-native gross margins sit near 52%, versus 75-85% for traditional SaaS, and Bessemer's fastest scalers run at just 25%.
11x, Builder.ai, and Klarna show the failure mode: inflated ARR, restated revenue, and rehired humans when the agent cannot deliver.
History rhymes: ATMs grew teller jobs 2% per year and spreadsheets still leave accountants up 5% through 2034.
Esta investigación se publica en inglés y español. Read in English
