AI Harness ARR Multiples Return to 100x: Harvey, Legora, Sierra Cross $100M in 9 Months

On August 9, Theory Ventures partner Tomasz Tunguz published a short post with hard data that significantly impacts AI coding toolchain valuations: Harvey, Legora, and Sierra—three AI harness (vertical industry agent platform) companies—each crossed $100M ARR within 9 months, with market valuation multiples of 50x, 56x, and 100x. The multiples themselves aren't new—100x ARR multiples existed during the 2021 SaaS bubble—but Tunguz's key finding is that the underlying fundamentals this time are 3x stronger than 2021. In other words, the 100x multiple is back, but the underlying growth runs at 3x speed.

The numbers in detail:

• Harvey (legal AI agent) — announced ARR crossing $100M on August 4, 2025 (CEO Winston Weinberg confirmed on CNBC); Series E valuation of $5B on June 23, 2025; revenue of approximately $75M at Series E pricing (TechCrunch data).
• Legora (legal AI agent, Europe) — announced ARR crossing $100M on April 2, 2026; Series D (including extension) valuation of $5.6B in April 2026; revenue at Series D pricing not disclosed.
• Sierra (customer service AI agent, founded by Bret Taylor) — published "Sierra hits $100m ARR milestone in 7 quarters" on November 21, 2025; valued at $10B in September 2025; revenue at round pricing not disclosed, but public figure of $26M at end of 2024.
• Ramp (fintech + AI tools + card interchange revenue) — valuation based on $1.4B revenue; its ARR multiple should be compared against fintech benchmarks.
• Decagon (customer service AI agent) — valuation based on $35M revenue (end of 2024 data); a January 2026 funding round implied a 129x upper bound (calculated on estimated October 2025 revenue—should be read as an "upper-bound estimate," not an exact multiple).

Per Tunguz, these companies' ARR multiples fall in the 25x to 125x range and have been accelerating in recent months—Legora, Sierra, and Ramp's multiples have all been rising since January 2026. This inverts the typical SaaS pattern of "larger ARR, lower multiple"—these companies' multiples kept climbing even after crossing $100M ARR.

Tunguz himself flags several essential caveats:

• Growth rates don't explain the multiple gaps—among the three companies that crossed $100M ARR, the fastest-growing one was priced at the lowest multiple. The market isn't paying a "growth premium"; it's paying a "category position premium."
• Category position is the key variable—the 50-100x multiples for Harvey / Legora / Sierra reflect that they are nearly the de facto sole players in verticals like "legal AI" and "customer service AI." Ramp's 85x also comes from its near-monopoly position in "fintech + AI tools." Decagon's 129x is an optimistic estimate for a still-early category—multiples will compress as the category gets crowded.
• The same pattern at Databricks—Tunguz also notes Databricks' ARR multiple has been accelerating upward since January 2026, indicating the "data + AI platform" category is being re-rated.
• The 2021 100x cycle was a product of "liquidity flood + low rates"—Tunguz wrote his own 100x ARR analysis five years ago (November 4, 2021), when the story was "pandemic QE + remote-work SaaS boom." But the return to 100x five years later comes with 3x faster underlying growth—a "higher quality" 100x.

Framed within August's AI coding toolchain storyline, this reveals new pricing rules for the "AI coding tools / vertical agent platforms" category:

• Cloud SaaS era — 20x at $10M ARR, 10x at $100M ARR, 5x at $1B ARR. Multiples decline monotonically with ARR.
• AI agent era — 50x at $10M ARR, 50-100x at $100M ARR, 30-50x at $1B ARR. Multiples rise rather than fall after crossing $100M—agent platforms' network effects + data flywheels + customer lock-in are an order of magnitude stronger than traditional SaaS.

For AI coding toolchains specifically:

• Valuation logic is diverging between "AI coding tools" and "vertical agent platforms" — general AI coding tools like Cursor / Devin / Replit Agent follow SaaS logic (bigger ARR, lower multiples); vertical agent platforms like Harvey / Legora / Sierra follow network-effects logic (bigger ARR, higher multiples). The former needs developer penetration; the latter needs industry monopoly position.
• The boundary of "AI harness" is being redefined — LangChain's Managed Deep Agents (Aug 8, topicId 178630987), Microsoft SkillOpt's "portable experience layer" (Aug 9, topicId 178603080), and Cloudflare making ADLC an SDLC default (Aug 10, topicId 178630990) all point to the same thing: AI harness is becoming a platform layer, not a tool layer. Tunguz's data validates this from the valuation side.
• Vertical agent platforms are the sweet spot for 2026 AI startups — Harvey (legal), Legora (European legal), Sierra (customer service), Decagon (customer service), Hebbia (finance), EvenUp (legal), Phaidra (facilities), Atropos Health (healthcare)... every company on this list carries a 50-125x multiple.

A few things worth calling out:

• Six of Tunguz's data points are estimates — 9/15 data points come from company announcements / founder posts; the remaining 6/15 are third-party estimates. Decagon's 129x in particular is an upper-bound estimate that will compress as the category crowds.
• Most of Ramp's $1.4B revenue is interchange income — Ramp isn't pure subscription SaaS, so its multiple should be benchmarked against fintech, not AI harness. Tunguz notes this in his own footnotes.
• Harvey's $5B Series E was priced at $75M revenue — implying 67x; at $100M+ ARR now, a follow-on round at 50x would mean a $5B valuation, doubling in 9 months. That's the "category position premium" in action.
• The return of the "100x cycle" isn't history repeating — per Tunguz, "back to 100x five years later, but at 3x underlying growth." This is a re-rating of AI agent platforms' network effects + data flywheels, not a simple cycle repeat.
• Sierra crossed $100M ARR in 7 quarters — averaging ~57% quarterly growth (geometric). What Sierra did in 21 months takes traditional SaaS companies 60-84 months on average—a 3-4x speed advantage for AI agent platforms.

Within August's AI coding toolchain storyline, this connects into a clear line with other events:

• Aug 8: LangChain Managed Deep Agents public beta (topicId 178630987) — AI harness platform layer.
• Aug 9: Prime Agent RLM Harness (topicId 178603082) — self-upgrading harness.
• Aug 10: Cloudflare ADLC + Agents trio (topicId 178630990) — SDLC default for AI coding toolchains.
• Aug 10: AI Harness ARR multiples back to 100x (this article) — market re-pricing of the "AI harness platform layer."

August 10 is the key moment when the market confirmed the "AI harness platform" category — every prior August AI coding toolchain release argued "harness is a platform"; Tunguz's data proves from the valuation side that "the market also believes harness is a platform."

Sources:

• Tomasz Tunguz original: https://www.tomtunguz.com/ai-harness-arr-multiples
• Tunguz 2021 100x ARR Multiple: https://tomtunguz.com/100x-arr/
• Harvey CNBC coverage: 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 $75M revenue: https://techcrunch.com/2025/06/23/harvey-series-e