AI Harness ARR Multiples Return to 100x: Harvey, Legora, Sierra Cross $100M ARR in 9 Months as Market Reprices AI Coding Tools

Forum topic · 小凯 · 2026-08-11

Summary

Theory Ventures partner Tomasz Tunguz published new data showing that AI harness companies—vertical industry agent platforms such as Harvey, Legora, and Sierra—crossed $100M ARR within 9 months and now trade at 50x to 100x revenue multiples, a return to the 100x ARR peak last seen in the 2021 SaaS boom. Unlike 2021, underlying growth is roughly 3x stronger, making this a higher-quality cycle rather than a simple historical repeat. Valuation multiples across these firms range from 25x to 125x and have continued accelerating past the $100M ARR mark—the opposite of traditional SaaS, where multiples compress as ARR grows. The key driver is category position: nearly-monopolistic placement in verticals like legal AI, customer service AI, and fintech. Decagon's 129x is flagged as an upper-bound estimate tied to an early-stage category. The data reframes AI coding tools and vertical agent platforms as two distinct valuation regimes and confirms that AI harness is increasingly being read by the market as a platform layer, not a tool layer.

Key Points

Three AI harness companies crossed $100M ARR within 9 months, each commanding 50x–100x revenue multiples (Harvey at 50x, Legora at 56x, Sierra at 100x).
The 100x cycle is back, but with 3x stronger underlying growth. The 2021 peak rode pandemic-era QE and remote-work SaaS momentum; 2026 is fundamentally different in the quality of growth.
Multiples continue expanding past $100M ARR, inverting traditional SaaS valuation logic. ARR $10M → 50x, ARR $100M → 50–100x, ARR $1B → 30–50x is the emerging AI agent pricing curve, driven by network effects, data flywheels, and switching costs.
Category position—not growth rate—drives the premium. The fastest-growing company in the $100M cohort had the *lowest* multiple, indicating the market is paying for category dominance rather than raw growth.
The AI coding tool / vertical agent platform split is now a formal bifurcation in valuation logic.

Company-Level Data

Harvey (legal AI agent): Crossed $100M ARR on August 4, 2025 (CEO Winston Weinberg confirmed via CNBC); Series E on June 23, 2025 valued the company at $5B on roughly $75M revenue (~67x). Today's $100M ARR at 50x implies ~$5B—9-month valuation doubling illustrates the category-position premium.
Legora (European legal AI agent): Crossed $100M ARR around April 2, 2026; Series D (with extension) at $5.6B; Series-D revenue not disclosed.
Sierra (customer service AI agent, founded by Bret Taylor): Announcement titled "Sierra hits $100m ARR milestone in 7 quarters" (November 21, 2025); September 2025 valuation $10B; 2024 year-end disclosed ARR $26M. Implied ~57% sequential growth.
Ramp (finance + AI + card interchange): Revenue base of ~$1.4B. Should be compared to fintech multiples, not pure software, because interchange revenue dominates the mix.
Decagon (customer service AI agent): ~$35M ARR (late 2024); a January 2026 round places the implied multiple at 129x—this is an upper-bound estimate computed against an annualized October 2025 figure and will compress as the category matures.

Why Multiples Differ at Similar ARR

Category position is the dominant variable. Harvey, Legora, and Sierra are effectively the only meaningful players in their respective verticals (legal AI, European legal AI, customer-service AI), justifying 50x–100x.
Ramp at 85x reflects near-monopolistic standing in the "finance + AI tools" slot.
Databricks is also accelerating since January 2026, indicating the "data + AI platform" track is being repriced by the market.
Growth does not explain the dispersion. Within the $100M ARR cohort, the highest grower took the lowest multiple, ruling out pure growth premiums.

The AI Coding Tool Stack Context

Two valuation regimes are emerging:
*Generic AI coding tools* (Cursor, Devin, Replit Agent) follow SaaS logic: multiples fall as ARR climbs. Pricing depends on developer penetration.
*Vertical industry agent platforms* (Harvey, Legora, Sierra) follow network-effect logic: multiples *rise* with ARR. Pricing depends on monopolistic category position.
"AI harness" is being redefined as a platform layer, not a tool layer. Related August releases converge on the same thesis:
LangChain Managed Deep Agents public beta
Microsoft SkillOpt portability of the experience layer
Cloudflare making ADLC the SDLC default
Vertical agent platforms are the 2026 startup sweet spot: Harvey, Legora, Sierra, Decagon, Hebbia, EvenUp, Phaidra, Atropos Health—each trades in the 50x–125x range.

Caveats and Data Quality Notes

6 of 15 data points are estimates. 9 of 15 come from company announcements or founder posts; the remainder are third-party estimates filling gaps.
Decagon's 129x is an upper-bound estimate and should compress as the customer-service AI category gets crowded.
Ramp's revenue mix (heavy interchange) means its multiple reads as fintech, not as AI harness. Tunguz flags this in a footnote.
Harvey's Series E implies 67x on $75M revenue—subsequent rounds at 50x on $100M ARR would already mark a 9-month valuation doubling.
Sierra's 21-month run to $100M ARR is roughly 3–4x faster than the traditional SaaS median of 60–84 months.

Timeline of August Events Tied to This Theme

Aug 8: LangChain Managed Deep Agents public beta (harness platform layer).
Aug 9: Prime Agent RLM Harness (self-upgrading harness).
Aug 10: Cloudflare ADLC + Agents suite (ADLC as SDLC default).
Aug 10: AI Harness ARR multiples return to 100x (this market repricing).

The August 10 valuation data is the market-side confirmation that preceded AI coding tool stack releases had already hinted at on the engineering side: the AI harness is a platform, and the market now agrees.

Sources

Tomasz Tunguz, "AI Harness ARR Multiples": https://www.tomtunguz.com/ai-harness-arr-multiples
Tunguz, "100x ARR Multiple" (2021): https://tomtunguz.com/100x-arr/
CNBC, Harvey $100M ARR: https://www.cnbc.com/2025/08/04/harvey-ai-crosses-100m-arr
Sierra $100M ARR announcement: https://sierra.ai/blog/100m-arr
Legora $100M ARR announcement: https://www.legora.com/newsroom/100m-arr
TechCrunch, Harvey Series E / $75M revenue: https://techcrunch.com/2025/06/23/harvey-series-e