Cycle on May 24, 2026

I am going to check the open threads for this topic to understand what specific areas or questions need to be pursued in this cycle.

I will now search the web to investigate the pricing structures of hypergrowth AI agent startups like Harvey, Mercor, Sierra, and others to see if they are utilizing outcome-based, consumption-based, or hybrid subscription models.

I will search specifically for "Sierra AI" pricing model, "Mercor" pricing model, and other hypergrowth agent companies like Decagon or Cognition. I will first fetch the LinkedIn post, the Reddit thread, and the Metronome page for Harvey, and then run searches for Sierra and Mercor. Let's do a search on Sierra AI pricing first.

This is highly relevant! Sierra has pioneered an outcome-based pricing model, paying only for successfully resolved customer service queries. Let's do a search on Mercor AI's pricing model next, and then we will fetch some of these pages to get precise quotes and details.

Let's search for "Decagon AI" pricing model, "Cognition" / "Devin" pricing model, and check other hypergrowth AI startups. Let's run a search for Decagon AI pricing model.

This is fantastic! Let's fetch some of these pages to gather precise details. I will fetch: 1. `https://decagon.ai/blog/pricing-ai-agents` 2. `https://decagon.ai/glossary/what-is-resolution-based-pricing` 3. `https://cheekypint.substack.com/p/bret-taylor-of-sierra-on-ai-agents` 4. `https://lennysvault.com/insights/growth-scaling-tactics/e0d5de29-37ce-4302-84e5-cd2b7f2a25fc` Let's fetch the first two first.

This is a very insightful article. It highlights a key tension in AI agent pricing: while "outcome-based" (per-resolution) pricing sounds ideal in theory, in practice, the majority of customers choose "per-conversation" pricing because of predictability, transparency, and simpler alignment of incentives (avoiding disputes over what counts as a "resolution"). Let's fetch the other Decagon glossary page on what is resolution-based pricing.

This is highly informative. Decagon offers both "per-conversation" and "per-resolution" pricing, but notes that most clients end up choosing per-conversation because of predictability, simpler tracking, and avoiding arguments over the definition of a "resolution". Now let's fetch the Sierra / Bret Taylor articles to see how Sierra approaches this. I will fetch `https://cheekypint.substack.com/p/bret-taylor-of-sierra-on-ai-agents` first. Let's do that.

Wow, this is an incredible source! This is an interview of Bret Taylor (co-founder of Sierra and Chairman of OpenAI) by John Collison (Stripe co-founder) from March 10, 2026. Let's analyze some key details: 1. **Sierra's metrics**: - Reached $100M ARR in 7 quarters. - Reached $150M ARR in 8 quarters. - Currently around $165M ARR (as of March 2026), one month into their next quarter. - Note: The previous finding says "AI application layer companies hit $100M ARR in 8-10 months". Sierra hit $100M in 7 quarters (21 months), which is still incredibly fast, and $150M in 8 quarters. Let's make sure we quote this accurately. 2. **Sierra's pricing model**: - Outcome-based pricing. - For customer service: "if the AI agent resolves the case, no human intervention, there's a pre-negotiated rate for that. If we do have to escalate to a person, that's free." - For sales: "sales commission." - Bret Taylor says: "We do outcomes-based pricing... I think the analogy of going from impression-based ads to CPC ads is apt." - Comparing outcome-based to usage-based: "I would actually argue outcomes-based is pretty different than usage-based... Outcomes-based is what business outcome is this agent designed to produce and did it produce it effectively?" 3. **Decagon's metrics and pricing**: - Decagon offers both per-conversation and per-resolution pricing. - Surprisingly, Decagon's Bihan Jiang writes that "the vast majority of our customers choose per-conversation pricing" because it is "predictable and transparent," has "aligned incentives" (avoiding arguing over the definition of "resolution"), and provides a "foundation of trust." - Let's check Decagon's metrics. Let's run a search for Decagon's ARR or funding to see if we can find hypergrowth stats. Let's do a web search.

Let's fetch one of the Forbes or TechCrunch articles or the Decagon blog about the $4.5B valuation. Let's do `https://decagon.ai/blog/decagon-closes-employee-tender-at-4-5-billion-valuation` or search for its details. Let's run a search for the Forbes article about Decagon's valuation and ARR. Let's fetch the Decagon blog about the tender offer first. Wait, let's search for "Decagon ARR Forbes 2026" or "Decagon revenue" to see what their actual revenue/ARR is. Let's do a search.

Let's fetch `https://sacra.com/c/decagon/` to understand Decagon's revenue, valuation & funding, or `https://techcrunch.com/2026/03/04/decagon-completes-first-tender-offer-at-4-5b-valuation/`. Let's fetch the TechCrunch article first.

This is incredibly interesting. Let's see: 1. **Decagon's valuation and growth**: - $4.5B valuation in March 2026 (tripled from $1.5B in June 2025). - Completed its first employee tender offer in March 2026. - Raised a $250 million Series D less than two months prior (Jan/Feb 2026). - "While Decagon has not disclosed its revenue figures since late 2024 — when its annual recurring revenue (ARR) surpassed eight figures — its rapidly climbing valuation suggests the company's growth remains on a steep upward trajectory." - Forbes estimated its 2024 revenue at $3 million and end-of-2025 revenue at $12 million. 2. **Pricing model dynamics of Decagon**: - Offers per-conversation and per-resolution pricing. - Per-conversation pricing has been chosen by the vast majority of customers because it's predictable, transparent, and avoids complex negotiations over what qualifies as a "resolution". - Per-resolution pricing charges a higher fixed rate, but only when the AI fully handles and resolves a customer issue. If the case is escalated to a human, there is no charge. Let's look at Sierra's metrics and pricing in more depth. Sierra: - Founded in 2023, launched in Feb 2024. - Reached $100M ARR in 7 quarters, $150M ARR in 8 quarters. - Currently around $165M ARR (as of March 2026). - This is a staggering growth curve, rivaling the fastest GTM software companies in history. - Sierra's pricing model: Outcome-based (per-resolution) pricing. - "Sierra earns a set fee for each AI-resolved call, directly tying revenue to the $10-$20 cost a customer avoids per deflection." (Lenny's Vault quote or other source). - "We do outcomes-based pricing. For a customer service context, that means if the AI agent resolves the case, no human intervention, there's a pre-negotiated rate for that. If we do have to escalate to a person, that's free. For sales, it would be a sales commission. And wherever possible, there's a way to align our interests with our clients', we choose it." (Bret Taylor on Cheeky Pint, March 2026). Let's search for Mercor's pricing dynamics to see what they are doing. Let's do a web search.

Amazing! Let's fetch `https://www.eesel.ai/blog/mercor-ai-pricing` and `https://techcrunch.com/2025/10/27/mercor-quintuples-valuation-to-10b-with-350m-series-c/` to understand Mercor's business, pricing model, and revenue. Let's do that.

This is awesome! This is a detailed look at Mercor, which explains its transition from an AI hiring platform to an "Expert-as-a-Service" (EaaS) platform that provides specialized human training data (doctors, lawyers, PhDs doing RLHF data labeling) to frontier AI labs like OpenAI and Anthropic. Let's analyze some key details: 1. **Mercor's business**: - Originally a talent marketplace for connecting engineers. - Pivoted to connecting AI labs (OpenAI, Anthropic, etc.) with specialized professionals (doctors, lawyers, PhDs) who train and improve AI models via RLHF and data labeling. - Known as "Expert-as-a-Service" (EaaS). - Serves over 30,000 contract consultants. - Reached a $10B valuation in late 2025 (Series C round of $350 million in October 2025 led by Felicis, Benchmark, General Catalyst, etc.). - Growth: Quintupled valuation to $10B from $2B in February 2025. - Payouts: Reportedly paying out over $1.5 million daily to human experts training AI (over $500M annually). 2. **Mercor's pricing model**: - For its EaaS model: "cost-plus" hourly billing. Clients pay an hourly rate for the experts' time, and Mercor takes a service cut. - For its recruiting platform model: Contingent fee model, around 30% of a hired candidate's salary. - The pricing is highly opaque and variable, typical for enterprise-level services. Let's do a quick search on "Cognition AI" / "Devin" pricing model to see if we can find any details about how they price. I'll search on "Cognition AI" OR "Devin" pricing model.

This is incredibly interesting! Devin by Cognition originally launched at $500/month per instance (or for teams), then introduced a pay-as-you-go model, and then in 2026 retired the old Core and Team plans to introduce new self-serve plans. Let's see: - TechCrunch (April 3, 2025): "Devin, the viral coding AI agent, gets a new pay-as-you-go plan." It says Devin became generally available for teams at $500 per month. - Let's search for the latest self-serve plans of Devin in 2026. Let's do a web search.

Let's fetch `https://cognition.ai/blog/new-self-serve-plans-for-devin` to understand Devin's 2026 pricing changes. This is extremely fresh and relevant! Let's do that.

This is spectacular! This is from April 14, 2026. Let's analyze Devin's new pricing model: - **Shift from high flat-rate to multi-tiered hybrid seat/usage model**: - Cognition is retiring the old **Core** and **Team** plans (which had a high entry point of $500/month for teams). - Introducing a new lineup: **Free, Pro ($20/month), Max ($200/month), Teams ($80/month minimum, usage-based), and Enterprise**. - They are moving from a pure flat fee or complex ACU (Agent Compute Unit) billing for self-serve to a hybrid model where a base subscription includes a quota, and additional usage is billed directly in dollars based on real compute/model cost. - They are also starting to charge for advanced features/products that were previously free, such as **Ask Devin** (Deep Mode), **Devin Review** (PR review agent), and **DeepWiki** (knowledge base generation). These will move to usage-based billing after trials or for advanced modes. - This addresses a key customer friction point: the high entry barrier of $500/month, replacing it with a $20/month Pro plan and an $80/month minimum Teams plan, while ensuring they can offset the high compute costs of "deep thinking" and agentic runs by charging in dollars for extra usage. Let's do a search on other major AI agent startups to see if there are any other interesting pricing or GTM updates, such as **Harvey** (legal tech), **Mercor** (recruiting), **Sierra** (customer support), **Decagon**, or **Crescendo**. Wait, what about **LlamaIndex**, **LangChain**, or others? No, the brief specifically mentions "go-to-market strategies AI-native startups are using to displace incumbents: product-led growth tactics, pricing model experiments, open-source plays, community-building approaches, partnership announcements, and launch strategies that are actually working." Let's see if we can find any other major pricing model experiments or GTM playbook updates for AI-native startups in 2026. Let's do a web search.

Oh, wow! Let's fetch `https://www.chargebee.com/blog/pricing-ai-agents-playbook/` to see if there's a 2026 playbook for pricing AI agents. Let's do that!

This is an absolute goldmine of information! This is a comprehensive 2026 playbook for pricing AI agents published by Chargebee, detailing several key lessons, metrics, and models. Let's analyze the core components of this playbook: 1. **The Three-Body Problem of Agentic AI Monetization**: - Pricing responds to rapid changes in: **Product capabilities**, **User consumption patterns**, and **Underlying compute/inference costs**. 2. **Pricing Models & Shift from Per-Seat to Outcome/Hybrid**: - **Outcome-based pricing (Pay for Results)**: - Tying revenue directly to business metrics the customer tracks (leads, resolved tickets, meetings booked). - Example: **Intercom Fin AI agent** charges **$0.99 per resolution** (when it fully resolves a customer issue without human intervention). - Example: **Sierra** charges a set fee for each resolved customer service call (deflection), directly tying revenue to the $10-$20 cost a customer avoids per deflection. - *Friction/Trade-offs*: Defining what a "resolution" is can be extremely tricky (e.g., a customer leaving chat mid-way, or a semi-delivered answer). It also leaves AI compute costs for "unsuccessful" attempts under-monetized. This is why Decagon notes that while they support "per-resolution" pricing, the vast majority of their customer base actually chooses "per-conversation" pricing because of predictability and simpler tracking. - **Action-based or Workflow-based Pricing (Pay for Work Done)**: - Pay for every instance of work done, which abstracts raw token/compute costs into user-understandable units (e.g. workflows run, API calls, research credits). - Example: **n8n** charges for workflows run (e.g., "10k workflows"). - Example: **Clay** uses a **credit-based pricing system** with a "burn table" where different actions (e.g., enriching a lead, finding an email, running an AI research step) consume different amounts of credits from a prepaid bucket. - **Hybrid Pricing (Predictable Base + Variable Usage Tail)**: - Solves the problem of high request variability and unpredictable compute costs by pairing a flat platform fee or per-seat base (to secure a price floor and predictable subscription revenue) with usage-based overages (credits, tokens, or action-based fees) for heavy usage. - Example: **Relevance AI** charges a standard recurring team fee with included seats and credit thresholds, charging for extra seats or overage credits as add-ons. - Example: **Lovable** charges a recurring fixed fee per user with included credits, allowing users to upgrade tiers or buy more credits on a usage basis. - Example: **Devin (by Cognition)**: In April 2026, retired its old $500/month Team plan for a new tiered hybrid model: **Free, Pro ($20/month), Max ($200/month), Teams ($80/month minimum, usage-based), and Enterprise**. Included usage counts against plan quotas, and overages are billed directly in dollars based on underlying model/compute costs. They also started charging for compute-heavy features (Ask Devin Deep Mode, Devin Review, DeepWiki) on a usage basis to reflect real compute costs. Let's do a search on "Mercor ARR" or "Mercor valuation" to see if there are any other details on Mercor's business. Wait, let's fetch `https://sacra.com/c/mercor/` or search for its details to confirm Mercor's ARR and revenue. Let's do a web search first.

Ah, the web search budget for this run is exhausted! That's fine. We have a massive amount of concrete, fresh, 2026-grounded evidence from our fetches and previous searches. Let's review the findings we can write. We have a major thread to resolve: `Outcome-Based and Credit Pricing Structures Across Hypergrowth AI Agent Startups` Let's structure the finding to tell the story through concrete evidence, naming and linking the actual sources, using verbatim quotes, and interpreting what it means. Let's look at the prior findings that were auto-retrieved: - The Death of Flat-Seat SaaS: The Shift to Outcome-Based and Pooled Consumption Pricing (slug: `ai-pricing-models-outcome-consumption-2026`, score 0.63) - DevTools Growth Playbook: GitHub-Led, Community-First Acquisition Strategies for AI-Native Startups (slug: `devtools-growth-playbook-github-community`, score 0.67) - AI Application Layer Companies Hit $100M ARR in 8–10 Months — Compressing GTM Velocity (slug: `ai-app-layer-growth-velocity`, score 0.66) Wait, the finding about "The Death of Flat-Seat SaaS" is highly relevant to what we just uncovered. We can update/refine that finding using its slug `ai-pricing-models-outcome-consumption-2026`! Wait, let's also write a fresh finding about the hypergrowth metrics and GTM/pricing trajectories of Sierra, Decagon, Mercor, and Devin as of early 2026. Or we can update `ai-pricing-models-outcome-consumption-2026` to be a comprehensive, state-of-the-art overview of how the pricing landscape has solidified in early 2026. Let's see if there are other slugs. Wait, let's look at `ai-pricing-models-outcome-consumption-2026` and see what's already there or write a fresh one or update it. Let's write a detailed note updating `ai-pricing-models-outcome-consumption-2026`. Let's read the prompt guidelines carefully: "For each finding worth persisting, call write_note. A finding TELLS THE STORY THROUGH CONCRETE EVIDENCE — it does not just describe the vibe. Every substantive finding must: - name and link the actual thing (the story, article, paper, or page that was posted/published) — not just "the discussion", - carry 1-2 REPRESENTATIVE VERBATIM QUOTES, copied exactly, each kept with the link it came from (an HN comment permalink, the article URL, etc.), - then say what it means." Let's look at the sources we can link and quote: 1. **Cheeky Pint Interview (Bret Taylor & John Collison, March 10, 2026)**: - URL: `https://cheekypint.substack.com/p/bret-taylor-of-sierra-on-ai-agents` - Quotes: - Bret Taylor: `"We do outcomes-based pricing. For a customer service context, that means if the AI agent resolves the case, no human intervention, there's a pre-negotiated rate for that. If we do have to escalate to a person, that's free. For sales, it would be a sales commission. And wherever possible, there's a way to align our interests with our clients', we choose it. I'm a huge believer in this. I think the analogy of going from impression-based ads to CPC ads is apt."` - Bret Taylor on Sierra's metrics: `"We reached $100 million in ARR in seven quarters, $150 in eight quarters. We're, I think, around $165 now, one month into our next quarter."` - Bret Taylor on outcome vs usage: `"I would actually argue outcomes-based is pretty different than usage-based... Outcomes-based is what business outcome is this agent designed to produce and did it produce it effectively? That is really aligning because it creates this whole vertical alignment. As a company, reducing your token utilization for the same outcomes is your problem, not your customer's."` 2. **Decagon Blog / Glossary Page (Decagon, December 11, 2024 / 2026 updates)**: - URL: `https://decagon.ai/blog/pricing-ai-agents` and `https://decagon.ai/glossary/what-is-resolution-based-pricing` - Quotes: - Bihan Jiang (Decagon): `"Per-conversation pricing: A fixed rate for every incoming conversation, with flexible pricing for higher volumes... Per-resolution pricing: A higher fixed rate for each fully resolved conversation, with no charge for escalations... we've seen the majority of our customers gravitate towards per-conversation pricing."` - Reasons for choosing per-conversation: `"1. Predictable and Transparent... Customers avoid unpredictable invoices and the constant renegotiations often required with outcome-based pricing. 2. Aligned Incentives. By focusing on conversation volume rather than overly parsing the definition of 'outcome,' the incentives are clean."` - Decagon completed its first employee tender offer in March 2026 at a $4.5B valuation, tripling from $1.5B in June 2025. - URL: `https://techcrunch.com/2026/03/04/decagon-completes-first-tender-offer-at-4-5b-valuation/` 3. **Cognition AI Blog (April 14, 2026)**: - URL: `https://cognition.ai/blog/new-self-serve-plans-for-devin` - Quotes: - `"We're retiring the old Core and Team plans and introducing a new lineup: Free, Pro, Max, Teams, and Enterprise. We're also beginning to charge for products that have been free until now, including Ask Devin, DeepWiki, and Devin Review."` - `"Pro — $20/month, with included quota... Teams — Usage based, with a minimum spend of $80/month... usage that is included in your plan will count against quota. If you go beyond your included quota, additional usage will be priced and billed in dollars rather than ACUs."` - `"These products do substantial work on your behalf, and in each case their cost depends on how much model usage they consume. We want pricing to reflect that more clearly."` 4. **Chargebee Blog (2026 playbook for pricing AI agents)**: - URL: `https://www.chargebee.com/blog/pricing-ai-agents-playbook/` - Quotes: - `"Agentic AI monetization is a three-body problem. Your pricing responds to the rapid changes in: your product, how individual users consume/interact with it, and the underlying costs incurred by your system to service your customers."` - `"Credits operate as an abstraction layer for when usage is spread across multiple actions and workloads... The variable amount at which every action consumes credit is controlled by you through a 'burn table'."` 5. **eesel AI Blog on Mercor AI pricing (November 6, 2025)**: - URL: `https://www.eesel.ai/blog/mercor-ai-pricing` - Quotes: - `"The 'Expert-as-a-Service' (EaaS) model means Mercor AI pricing for clients is based on an hourly 'cost-plus' rate for the specialized human experts."` - `"Mercor, an AI recruiting startup founded by a few 22-year-olds, just hit a wild $10 billion valuation... The company manages a network of over 30,000 contract consultants... paying out $1.5 million+ daily to human experts training AI."` This is incredibly rich, highly specific, and perfectly addresses the open thread! Let's write the note using the slug `ai-pricing-models-outcome-consumption-2026` to update/refine the existing findings about AI pricing models. Wait, is there any other slug we should update? Let's see: `ai-app-layer-growth-velocity` is also highly relevant because Sierra's metrics ($100M ARR in 7 quarters, $150M ARR in 8 quarters) and Decagon's metrics (tripled valuation to $4.5B in early 2026) directly speak to application layer growth velocity! Let's write a note for `ai-pricing-models-outcome-consumption-2026` first, and then we can write or update another note if needed, or put it all into one powerful, cohesive update. Let's update `ai-pricing-models-outcome-consumption-2026` with the pricing details, and we can also update `ai-app-layer-growth-velocity` with the incredible ARR and valuation metrics we just uncovered. Let's first write the updated note for `ai-pricing-models-outcome-consumption-2026`. Let's draft it carefully.

This is excellent! The note has been written and the slug `ai-pricing-models-outcome-consumption-2026` has been successfully updated. Let's now look at `ai-app-layer-growth-velocity`. The original note mentions: "AI Application Layer Companies Hit $100M ARR in 8–10 Months — Compressing GTM Velocity" In our research, we found that Sierra reached $100M ARR in 7 quarters (21 months) and $150M ARR in 8 quarters (24 months), reaching $165M ARR in early 2026. This is still a historic hypergrowth trajectory, and is more grounded and precise than the "8-10 months" general claim (which might apply to a few outlier consumer apps like ChatGPT/Character.ai, but for enterprise application layers like Sierra, 7 quarters is the absolute gold standard). Let's also look at Decagon, which tripled its valuation to $4.5B in early 2026 (completed its first employee tender offer in March 2026), having been founded just in 2023. This is an incredible GTM velocity. Let's refine `ai-app-layer-growth-velocity` to reflect these exact, confirmed metrics for Sierra and Decagon. Let's do a write_note on `ai-app-layer-growth-velocity`. Let's draft it.

This is excellent! The note has been written and the slug `ai-app-layer-growth-velocity` has been successfully updated. Let's see if there are any other open threads or if we should open a new one. The only open thread listed was: `Outcome-Based and Credit Pricing Structures Across Hypergrowth AI Agent Startups` (id: `019e53de-a34b-7595-8ee9-f42a0afb941b`) We have fully addressed and investigated this thread! We've researched the pricing models (outcome-based, credits, tiered hybrid) and customer friction points across Sierra, Decagon, Mercor, and Devin. Let's resolve this thread. I will call `resolve_thread`.

Now let's check if there are any other open threads or questions we should open for a future cycle. A good follow-up question for a future cycle would be: "How are enterprise buyers measuring and auditing the ROI of AI agents over time? Are we seeing a shift from initial pilot adoption to long-term contract renewals, and what customer churn patterns or contract renegotiation dynamics are emerging as these agents scale?" Let's record a new thread for this. This is a very natural and highly valuable question for GTM playbooks as these startups hit their 2-year and 3-year marks.

Now that we have written the findings and resolved/recorded the threads, we can call `finish_cycle`. Let's do that.

Investigated the pricing structures and growth metrics of hypergrowth AI agent startups in early 2026, centering on Sierra, Decagon, Mercor, and Devin (Cognition). Resolved the open question on outcome-based and credit pricing models, updated the corresponding growth velocity and pricing notes, and logged a new thread on enterprise ROI auditing and churn patterns for future research.