---
title: "Vanta CEO on AI Token Markups: Why Per-Token Pricing Loses to Claude"
date: "2026-10-25"
video_id: "7700704312195845406"
tiktok_url: "https://www.tiktok.com/@iltb_podcast/video/7700704312195845406"
creator: "@iltb_podcast"
category: "Business"
tags: ["ai pricing", "vanta", "christina cacioppo", "claude api pricing", "token markup"]
excerpt: "Vanta CEO Christina Cacioppo says marking up Claude tokens invites a comparison you lose. What she said, what 4x would cost and why list price hides real margins."
---
Vanta CEO Christina Cacioppo argues that AI applications should not bill customers per token, because customers will compare that price with Anthropic's list price and go try Claude directly. A TikTok from @iltb_podcast cuts the moment under the title "You Have Just Marked Up a Claude Token 4x." The 4x is a customer objection she acts out, not a measured price, and Vanta's own policy is to absorb token costs instead of reselling them.

## Where is the clip from?

The speaker is Christina Cacioppo, co-founder and CEO of Vanta, on Invest Like the Best with Patrick O'Shaughnessy. The episode is [EP.496](https://muckrack.com/podcast/invest-like-the-best-with-patrick-oshaughnessy/), "Christina Cacioppo - Building an Enduring Software Company in the Age of AI," published on 20 October 2026 and running about 57 minutes. Colossus, the show's publisher, lists it as ["Becoming a Category Winner."](https://colossus.com/episode/becoming-a-category-winner/) The TikTok was posted five days later.

Vanta was founded in 2018 by Cacioppo and Erik Goldman and is based in San Francisco. [Fortune](https://fortune.com/2026/04/29/exclusive-vanta-arr-300-million-sequoia-shadow-ai-claude-cursor/) reported in April 2026 that it had reached $300 million in annual recurring revenue with more than 16,000 customers. [Sacra](https://sacra.com/c/vanta/) puts that at 69 percent year-over-year growth. Its last disclosed valuation was $4.15 billion, after a $150 million Series D led by Wellington Management in July 2025.

## What did Cacioppo say about token markups?

She describes what Vanta hears from customers: "if you try to charge them based on tokens, they are very quick to say, oh, you've just marked up a Claude token 4x." She continues in the customer's voice: "Are you 4x as good as Claude? I don't know. Let me go try it out."

Her conclusion is that "if you're trying to mark up at all, users are very smart and you will lose on that comparison," and that "anyone who is marking up Claude tokens 4x" and passing that on to users is making a mistake.

The 4x is her example of a customer complaint. She names no company and quotes no price, and the clip shows no pricing data.

## How does Vanta price AI features?

The episode's show notes say Vanta "won't mark up a single token." [BigGo's coverage](https://finance.biggo.com/news/6364e4c9897ec8fe) quotes her: "I think charging tokens themselves really puts you up against Claude." [Scuttlebutt's summary](https://getscuttlebutt.substack.com/p/youre-really-only-as-good-as-the) says Vanta "eats the AI token cost rather than disguise a markup as 'credits'." Vanta charges a software fee and absorbs the inference cost.

## What would a 4x markup on Claude actually cost?

Anthropic's [pricing page](https://platform.claude.com/docs/en/about-claude/pricing), checked on 26 October 2026, lists these prices per million tokens (input / output):

- Claude Haiku 5.5: $0.10 / $0.50 for prompts up to 100,000 tokens
- Claude Haiku 4.5: $1 / $5
- Claude Sonnet 5 and Sonnet 5.5: $2 / $10
- Claude Sonnet 4.6 and 4.5: $3 / $15
- Claude Opus 5.5: $4 / $20
- Claude Opus 5: $5 / $25
- Claude Fable 5.1: $10 / $50

An app reselling Claude Sonnet 5 at four times list would charge $8 / $40. That is more than Anthropic charges for Opus 5.5 or Opus 5. A customer with a calculator would ask why they are paying more than Opus prices to reach Sonnet.

## Why list price hides real AI margins

Anthropic's pricing includes three things that make the markup hard to read from either side.

- **Prompt caching.** Cache hits cost 0.05 times the base input price on Opus 5.5 and Sonnet 5.5, and 0.1 times on most other models. On Sonnet 5.5 that is $0.10 per million cached input tokens against a $2 list price.
- **Batch API.** Batched requests get 50 percent off both input and output.
- **Tokenizer changes.** Claude 4.7 and later models use a tokenizer that produces about 30 percent more tokens for the same text than Sonnet 4.6 and earlier, so token counts are not comparable across model generations.

An app could charge exactly Anthropic's list price, show customers a 1.0x multiple and still make a wide margin after caching and batching. An honest pass-through on a long-context workload could also look like a markup. Customers compare against list price, not the vendor's real cost. That supports Cacioppo's practical point: per-token billing invites a comparison the vendor cannot control.

## What are agent harnesses, and why do they matter here?

In the full audio, Cacioppo briefly says "we're going to talk about harnesses" before returning to pricing. Anthropic's engineering post on [Managed Agents](https://www.anthropic.com/engineering/managed-agents) defines a harness as "the loop that calls Claude and routes Claude's tool calls to the relevant infrastructure." How many tokens an application burns depends on its harness, so pricing and harness design are linked.

## A note on the "Forex" transcription

Automatic transcripts render "4x" as "Forex." [Podcast Alpha's summary](https://podcastalpha.substack.com/p/christina-cacioppo-how-ai-is-repricing-7da) of the episode published the quote as "You've just marked up a Claude token Forex." The video's own captions read "4x," which is correct.

## Bottom line

Cacioppo's argument is that reselling tokens puts an AI application in a price comparison against the model it is built on, and the application loses. The 4x is a hypothetical complaint, not a market figure. At Anthropic's current prices, a 4x markup on Sonnet 5 would cost more than Opus list, while caching and batching mean list price says little about a vendor's real cost.

## Sources

- [Anthropic pricing](https://platform.claude.com/docs/en/about-claude/pricing) - list prices, caching, batch discount and tokenizer note
- [Invest Like the Best episode listing, Muck Rack](https://muckrack.com/podcast/invest-like-the-best-with-patrick-oshaughnessy/) - episode number and date
- [Invest Like the Best, Apple Podcasts](https://podcasts.apple.com/us/podcast/invest-like-the-best-with-patrick-oshaughnessy/id1154105909) - feed listing and runtime
- [Becoming a Category Winner, Colossus](https://colossus.com/episode/becoming-a-category-winner/) - publisher's episode page
- [Fortune, 29 April 2026](https://fortune.com/2026/04/29/exclusive-vanta-arr-300-million-sequoia-shadow-ai-claude-cursor/) - Vanta's ARR and customers
- [Vanta on Sacra](https://sacra.com/c/vanta/) - growth rate and funding history
- [Vanta (company), Wikipedia](https://en.wikipedia.org/wiki/Vanta_(company)) - founding and founders
- [BigGo Finance, 20 October 2026](https://finance.biggo.com/news/6364e4c9897ec8fe) - Cacioppo's quote on token pricing
- [Podcast Alpha, 23 October 2026](https://podcastalpha.substack.com/p/christina-cacioppo-how-ai-is-repricing-7da) - episode summary with the "Forex" error
- [Scuttlebutt, 20 October 2026](https://getscuttlebutt.substack.com/p/youre-really-only-as-good-as-the) - episode summary
- [Scaling Managed Agents with stable interfaces, Anthropic](https://www.anthropic.com/engineering/managed-agents) - definition of a harness


## Video description and verification notes

**A 57.70 second MP4 (`mov,mp4,m4a,3gp,3g2,mj2` container, 1,686,965 bytes, overall bitrate 233,890 bps) carrying one HEVC Main video stream at 720x1280, 29.97 fps (`r_frame_rate=2997/100`), 1,729 frames, 162,891 bps, duration 57.691024 s, plus one AAC HE-AACv2 audio stream at 44,100 Hz stereo, 64,202 bps, duration 57.700998 s; TikTok format id `bytevc1_720p_233886-1`; posted by handle `@iltb_podcast` (uploader id `7698308816836232206`) under the account nickname "Invest Like The Best" at epoch `1792959958`, which is 2026-10-25 20:25:58 UTC, a Sunday; audio credited in metadata as `track: "original sound"` by `artist: "Invest Like The Best"`; the description is 77 characters long and consists only of the title plus five hashtags; at my capture time (metadata.json mtime 2026-10-26 00:01:29 local) the clip showed 502 views, 7 likes, 0 comments and 1 repost, so this is a near-zero-engagement upload read roughly three and a half hours after posting. I read 17 of the 29 extracted frames (001, 003, 004, 005, 006, 009, 011, 015, 016, 017, 018, 020, 022, 024, 026, 028, 029) at 2 s intervals, and the Whisper transcript is 10 segments and 180 words. Every frame is the same vertical split screen: the top half holds a man in a grey quarter-zip over a white collar, seated at a podcast boom mic in front of a blown-out window with a glass of amber liquid beside him, and he never speaks; the bottom half holds a woman in a dark green long-sleeve top at a second boom mic in a dimmer room with a framed landscape painting of two figures on a lawn behind her left shoulder and a white cabinet behind her right. A static two-line title card sits over the top half in a white-filled, black-outlined serif for the entire 57 seconds: "You Have Just Marked / Up a Claude Token 4x". Rolling burned-in captions sit across the seam between the two halves and read, in order, "We have talked to our customers a lot" (frame 001, 0 s), "And I think what we've generally heard is like," (frame 003, 4 s), "people are real smart, but actually," (frame 005, 8 s), "and so if you try to charge them based on tokens," (frame 006, 10 s), "oh, you've just marked up a Claude token 4x." (frame 009, 16 s), "I don't know, let me go try it out." (frame 011, 20 s), "against the models." (frame 015, 28 s), "Didn't probably no application really wants to do it." (frame 016, 30 s), "but like, you don't want to get there." (frame 017, 32 s), "And so what we've seen on token pricing is like," (frame 018, 34 s), "if you're trying to mark up at all," (frame 020, 38 s), "users are very smart and you will lose on that comparison," (frame 022, 42 s), "and this gets to the margin question, right?" (frame 024, 46 s), "But anyone who is marking up Claude tokens 4x" (frame 026, 50 s) and "I think it's just like that is not a user model." (frames 028 and 029, 54 s and 56 s). No chart, no pricing table, no dollar figure, no product UI, no terminal, no company logo and no lower-third name card appears in any frame I read; the only numeral on screen anywhere in the clip is "4x", and neither speaker is identified on screen at any point. The spoken argument is a single compressed claim: an application that bills its users per token invites them to divide that price by the model vendor's list price, discover a multiple, and go test the raw model instead, which drags the application into a head-to-head comparison against the model it is built on and which it will lose. The transcript's last segment ends at 56.74 s against a 57.70 s audio stream, so nothing is timestamped past the end of the audio and there is no end-of-file hallucination in this transcript.**

## The clip names nobody, and both speakers are identifiable: the episode is Invest Like the Best EP.496, published 20 October 2026

Nothing on screen says who is talking. The account nickname in the metadata is "Invest Like The Best", which is the podcast hosted by Patrick O'Shaughnessy, and the speaker in the lower half is Christina Cacioppo, co-founder and CEO of Vanta.

Muck Rack's episode listing for the show gives the most recent five episodes as "Christina Cacioppo - Building an Enduring Software Company in the Age of AI [EP.496]" on 20 October 2026, "David Vélez - Building Nubank and the Future of Global Banking [EP.495]" on 13 October 2026, "Andrew Huberman - The Frontier of Neurotechnology [EP.494]" on 6 October 2026, "Noah Shinn - Building Instinct: The Personal Agent [EP.493]" on 28 September 2026 and "Gabe Stengel - Building Investing Superintelligence [EP.492]" on 22 September 2026. The Apple Podcasts listing for the show gives the same Cacioppo title with a 57 minute runtime. Colossus, which publishes the show, carries the same episode under episode number 496 but displays it with the shorter title "Becoming a Category Winner", so the feed title and the publisher's display title differ. I could not read the Colossus transcript, which is behind a login, and a fetch of Patrick O'Shaughnessy's post about the episode on X returned HTTP 402, so I am not citing either.

On spelling: it is Christina Cacioppo, two c's and two p's, and Patrick O'Shaughnessy with the apostrophe. Cacioppo co-founded Vanta in 2018 with Erik Goldman, both formerly of Dropbox, and the company is headquartered in San Francisco. Fortune reported on 29 April 2026 that Vanta had reached $300 million in annual recurring revenue and served more than 16,000 customers including Snowflake, Atlassian, Duolingo, Ramp, Cursor and Harvey. Sacra puts that $300 million at 69 percent year-over-year growth, up from $250 million at the end of 2025, with roughly $19,000 ARR per customer. The last disclosed valuation is $4.15 billion after a $150 million Series D led by Wellington Management in July 2025, against $2.45 billion at the Series C led by Sequoia in July 2024, for about $504 million raised in total.

The timing matters for how to read this upload. The episode went out on 20 October 2026 and this clip was posted on 25 October 2026 at 20:25:58 UTC, five days later. It is a promotional cutdown from a back catalogue of one long conversation, not a reaction to anything that happened that week.

## "4x" is a customer's hypothetical objection, not a price anyone in the clip measured

The title card asserts a fact: "You Have Just Marked Up a Claude Token 4x". The clip does not support it as one.

At 9.72 s the speaker says "if you try to charge them based on tokens, they are very quick to say, oh, you've just marked up a Claude token [4x]", and at 17.78 s she continues "[Are you 4x] as good as Claude? I don't know. Let me go try it out." Both lines are her acting out what a customer says to a vendor. She is narrating an objection, not reporting a measurement. No company is named. No price is quoted. No pricing page is shown or cited anywhere in the 57 seconds.

At 50.36 s she generalises it: "anyone who is marking up Claude tokens [4x] and trying to, like, pass that off to their users, I think it's just, like, that is not a [user model]". That is a conditional about a hypothetical vendor. It still does not assert that any particular vendor charges four times Anthropic's list price.

As a market figure, the 4x is unverified. I found no primary source in which Cacioppo or Vanta publishes a measured markup multiple for any named application, and the clip supplies none. What is corroborated is the policy she is contrasting it against. The episode's own show notes state that Vanta "won't mark up a single token", and secondary write-ups of the episode agree on the mechanism: Vanta absorbs the inference cost and charges a software fee rather than reselling tokens or dressing a markup up as credits. BigGo's coverage of the episode quotes her as saying "I think charging tokens themselves really puts you up against Claude." That is the same argument the clip preserves, with the 4x stripped of its conditional framing and promoted to a headline.

## Whisper heard "Forex", the burned-in captions heard "4x", and at least one published summary shipped the error

The local Whisper transcript renders the number as "Forex" in all three places it occurs: at 9.72 to 16.78 s ("you've just marked up a Claude token Forex"), at 17.78 to 19.46 s ("Are you Forex as good as Claude?") and at 50.36 to 56.74 s ("anyone who is marking up Claude tokens Forex"). The spoken phrase is "four ex", and an automatic transcriber with no domain prior collapses that to the currency-trading term.

The burned-in captions in the video get it right. Frame 009 at 16 s reads "oh, you've just marked up a Claude token 4x." and frame 026 at 50 s reads "But anyone who is marking up Claude tokens 4x". The title card reads "4x". So on this specific word the video's own captions are the better source and the local transcript is wrong, which is the reverse of the usual direction.

This is not a local quirk. Podcast Alpha's published write-up of the episode, dated 23 October 2026, quotes the line as "You've just marked up a Claude token Forex. Are you Forex as good as Claude?" An ASR artifact made it into a human-facing summary of the episode unaltered. Anyone searching for the quote in good faith can land on a version of it that reads as a foreign-exchange reference.

The captions are themselves machine-generated and carry their own errors in the other direction, which the next section covers. Neither transcription is authoritative on its own. Where they disagree, the two have to be read against each other.

## A literal 4x markup, priced against Anthropic's live list, lands above the frontier tier

The argument in the clip is that a user who sees a per-token price will divide it by the model vendor's published price and act on the quotient. Checking that against Anthropic's current price list, fetched from the Claude platform pricing documentation on 26 October 2026, shows why the comparison is unflattering for the application.

The list prices per million tokens (input / output) are: Claude Haiku 5.5 at $0.10 / $0.50 for prompts up to 100,000 tokens and $0.50 / $2.50 for prompts over that threshold; Claude Haiku 4.5 at $1 / $5; Claude Sonnet 5 and Claude Sonnet 5.5 at $2 / $10; Claude Sonnet 4.6 and 4.5 at $3 / $15; Claude Opus 5.5 at $4 / $20; Claude Opus 5, 4.8, 4.7, 4.6 and 4.5 at $5 / $25; and Claude Fable 5.1 and Claude Fable 5 at $10 / $50. The documentation also notes that the $2 / $10 pricing for Claude Sonnet 5, originally announced as introductory pricing through 31 August 2026, is now the standard price and that the scheduled increase to $3 / $15 on 1 September 2026 will not happen.

Run the multiple. An application reselling Claude Sonnet 5 at four times list charges $8 per million input tokens and $40 per million output tokens. Claude Opus 5.5 costs $4 / $20 at list, and Claude Opus 5 costs $5 / $25. So a 4x markup on a mid-tier model prices the application above Anthropic's own top-end Opus tier and most of the way to Claude Fable 5.1 at $10 / $50. Her rhetorical question at 17.78 s, "Are you [4x] as good as Claude?", understates the problem. The real question a user with a calculator asks is why they are paying more than Opus list to reach Sonnet.

## Cache reads, batch pricing and a changed tokenizer make the markup harder to see than the line implies

The clip treats the markup as something a user can simply read off. Anthropic's pricing page lists three mechanisms that break that arithmetic in both directions.

First, prompt caching. Cache hits and refreshes are priced at 0.05 times the base input price on Claude Opus 5.5 and Claude Sonnet 5.5, at 0.025 times base on Claude Fable 5.1 and Claude Mythos 5.1, and at the standard 0.1 times base on everything else. On Claude Sonnet 5.5 that is $0.10 per million cached input tokens against a $2 list. An application with a long stable system prompt and tool definitions pays a fraction of list on most of its input volume.

Second, the Batch API applies a 50 percent discount to both input and output tokens, taking Claude Sonnet 5 to $1 / $5 and Claude Opus 5.5 to $2 / $10.

Third, the tokenizer changed. The pricing documentation states that Claude 4.7 and later models and Claude Mythos Preview use a newer tokenizer that produces approximately 30 percent more tokens for the same text, with the exact increase depending on content and workload, while Claude Sonnet 4.6 and earlier use the previous tokenizer. Token counts are therefore not comparable across model generations for the same input.

Put together, an application can resell tokens at exactly Anthropic's list price, show a user a 1.0x multiple, and still run a wide gross margin because its blended cost after caching and batching sits well below list. The reverse also holds: an honest pass-through of real cost on a long-context workload can look like a markup to a user comparing against the headline number. The comparison a user actually runs is against list price, not against the application's cost, and the clip's framing does not distinguish the two. That strengthens the practical version of her conclusion, which is that per-token billing invites a comparison the vendor does not control, rather than the literal version, which is that the multiple is visible and fair.

## The burned-in captions dropped a clause, and the last sentence is unresolved between two transcriptions

The two machine transcriptions diverge in a way that matters for anyone quoting this clip.

Between 28.68 s and 34.28 s Whisper gives "I think probably no application really wants to, like, we're going to talk about harnesses, but, like, you don't want to get there." The burned-in caption covering the same stretch reads "Didn't probably no application really wants to do it." at frame 016 (30 s), then "but like, you don't want to get there." at frame 017 (32 s). The caption garbles "I think" into "Didn't" and drops the self-interrupting clause "we're going to talk about harnesses" entirely. That clause is the only signal in the clip that the full conversation goes on to discuss agent harnesses, which is the architectural layer the pricing argument sits on top of. Anthropic's engineering post "Scaling Managed Agents with stable interfaces", published 8 April 2026, defines a harness as "the loop that calls Claude and routes Claude's tool calls to the relevant infrastructure". An application's token consumption is a property of its harness, so the aside is substantive rather than filler.

The last sentence does not resolve. Whisper ends at 56.74 s with "that is not a user-on-all", which is not English. The burned-in caption at frames 028 and 029 reads "that is not a user model." Both are machine output and neither is corroborated by a primary transcript I could reach, so I am not quoting the ending as settled. What survives the disagreement is the clause before it, where both agree: she is rejecting the practice, not describing it neutrally.

One further point of context sits outside the clip. Secondary summaries of EP.496 report Cacioppo citing roughly 80 percent growth at roughly 80 percent gross margins for Vanta, and 25 percent of customers currently in an agent-led flow against a 75 percent goal. I could not verify any of those four figures against a primary transcript or a Vanta disclosure, and the growth number is in tension with Sacra's 69 percent year-over-year ARR figure for April 2026. Treat all four as unverified.

## Key Takeaways

- **Correction:** The title card states "You Have Just Marked Up a Claude Token 4x" as a fact. In the clip the line is Christina Cacioppo acting out a customer objection at 9.72 s and generalising it conditionally at 50.36 s. No company is named, no price is quoted and no pricing page appears in any of the 29 frames.
- **Correction:** The local Whisper transcript renders the number as "Forex" at 9.72 s, 17.78 s and 50.36 s. The spoken word is "4x", confirmed by the video's own burned-in captions at frames 009 and 026 and by the title card.
- **Correction:** Podcast Alpha's 23 October 2026 write-up of the episode reproduces the same ASR error, publishing the quote as "You've just marked up a Claude token Forex. Are you Forex as good as Claude?"
- **Correction:** The burned-in captions drop the clause "we're going to talk about harnesses" from the 28.68 to 34.28 s stretch and render "I think" as "Didn't" at frame 016. The captions are machine-generated and are not more reliable than Whisper outside the "4x" token.
- **Verified:** The source is Invest Like the Best episode 496, "Christina Cacioppo - Building an Enduring Software Company in the Age of AI", published 20 October 2026, hosted by Patrick O'Shaughnessy, running about 57 minutes. The TikTok was posted five days later at epoch 1792959958, which is 2026-10-25 20:25:58 UTC.
- **Verified:** Anthropic's current list prices per million tokens are $2 / $10 for Claude Sonnet 5 and Claude Sonnet 5.5, $4 / $20 for Claude Opus 5.5, $5 / $25 for Claude Opus 5, $0.10 / $0.50 for Claude Haiku 5.5 under a 100,000 token prompt, and $10 / $50 for Claude Fable 5.1. A literal 4x markup on Sonnet 5 is $8 / $40, which is above Opus 5.5 and Opus 5 list.
- **Clarification:** Anthropic's price list includes a 50 percent Batch API discount and cache hit pricing at 0.05 times base input on Opus 5.5 and Sonnet 5.5, so an application's blended cost per token can sit far below list. A visible 1.0x multiple against list does not mean the application is passing through its own cost.
- **Clarification:** The pricing documentation states that Claude 4.7 and later models and Claude Mythos Preview use a newer tokenizer producing approximately 30 percent more tokens for the same text than Sonnet 4.6 and earlier. Token counts are not comparable across model generations, which complicates any user-side markup calculation.
- **Context:** Vanta reached $300 million ARR and more than 16,000 customers by April 2026 per Fortune, at 69 percent year-over-year growth per Sacra, with a last disclosed valuation of $4.15 billion after a $150 million Series D led by Wellington Management in July 2025. The company was founded in 2018 by Christina Cacioppo and Erik Goldman.
- **Unverified:** The 4x multiple as a description of any real vendor's pricing. Also unverified: the roughly 80 percent growth, roughly 80 percent margin, 25 percent agent-led and 75 percent target figures attributed to the episode by secondary summaries, none of which I could trace to a primary transcript or a Vanta disclosure.
- **Unverified:** The clip's final words. Whisper gives "that is not a user-on-all" and the burned-in caption gives "that is not a user model". Both are machine output and the two disagree.
- **Context:** The Colossus transcript for episode 496 is behind a login and a fetch of Patrick O'Shaughnessy's X post about the episode returned HTTP 402, so neither is cited here.

## Resources

- [Pricing, Anthropic / Claude platform documentation, fetched 2026-10-26](https://platform.claude.com/docs/en/about-claude/pricing) establishes every per-million-token list price, the 50 percent Batch API discount, the cache hit multipliers of 0.05x and 0.025x and 0.1x, the Claude Sonnet 5 introductory-to-standard pricing note, and the statement that Claude 4.7 and later use a tokenizer producing approximately 30 percent more tokens for the same text.
- [Invest Like the Best with Patrick O'Shaughnessy episode listing, Muck Rack](https://muckrack.com/podcast/invest-like-the-best-with-patrick-oshaughnessy/) establishes that the Cacioppo episode is EP.496, titled "Building an Enduring Software Company in the Age of AI", published 20 October 2026, and gives the surrounding episode numbers and dates I use to fix the release order.
- [Invest Like the Best with Patrick O'Shaughnessy, Apple Podcasts](https://podcasts.apple.com/us/podcast/invest-like-the-best-with-patrick-oshaughnessy/id1154105909) establishes the 57 minute runtime and confirms the episode title in the podcast feed.
- [Becoming a Category Winner, Colossus](https://colossus.com/episode/becoming-a-category-winner/) establishes the publisher's own episode page at number 496 and the shorter display title that differs from the feed title. Its transcript requires a login and is not cited.
- [Exclusive: Vanta hits $300 million ARR as 'shadow AI' explodes across corporate America, Fortune, 29 April 2026](https://fortune.com/2026/04/29/exclusive-vanta-arr-300-million-sequoia-shadow-ai-claude-cursor/) establishes the $300 million ARR figure, the 16,000-plus customer count, the named customers and the $4.15 billion July 2025 valuation.
- [Vanta revenue, valuation and funding, Sacra](https://sacra.com/c/vanta/) establishes the 69 percent year-over-year ARR growth rate for April 2026, the $250 million end-of-2025 ARR, the roughly $19,000 ARR per customer, and the full funding history including the $504 million total raised.
- [Vanta (company), Wikipedia](https://en.wikipedia.org/wiki/Vanta_(company)) establishes the 2018 founding, the Christina Cacioppo and Erik Goldman co-founder names, the San Francisco headquarters and the dated round-by-round valuations I cite.
- [Vanta's Cacioppo says SaaS multiples are wrong, token markups are a losing fight, and top engineers now cost 10x, BigGo Finance, 20 October 2026](https://finance.biggo.com/news/6364e4c9897ec8fe) establishes the direct quote "I think charging tokens themselves really puts you up against Claude" and the description of Vanta absorbing inference cost rather than reselling tokens.
- [Christina Cacioppo: How AI Is Repricing Every Software Company, Podcast Alpha, 23 October 2026](https://podcastalpha.substack.com/p/christina-cacioppo-how-ai-is-repricing-7da) establishes that the "Forex" transcription error propagated into a published summary of the episode, and is the source of the unverified 25 percent / 75 percent agent-led flow figures and the 80 percent growth and margin figures I flag.
- [Christina Cacioppo on Invest Like the Best: Vanta CEO, Scuttlebutt, 20 October 2026](https://getscuttlebutt.substack.com/p/youre-really-only-as-good-as-the) establishes the 20 October 2026 date, the 57 minute duration, Patrick O'Shaughnessy as host, and the summary line that Vanta "eats the AI token cost rather than disguise a markup as 'credits'". Its full text is paywalled.
- [Scaling Managed Agents with stable interfaces, Anthropic, 8 April 2026](https://www.anthropic.com/engineering/managed-agents) establishes Anthropic's own definition of a harness as "the loop that calls Claude and routes Claude's tool calls to the relevant infrastructure", which I use to show why the clause the captions dropped was substantive.
