This document draws on public reporting on the five case-study companies (Clay, Writer, Legora, Sierra, Hugging Face); two facilitated ON_Discourse group chats among AI-native operators (July 16 and July 23, 2026); and Domain’s own experiment building a collaborative AI product from zero.
Direct quotes and specific examples in the pattern and pricing sections are drawn from participants in two ON_Discourse operator sessions held under the Chatham House rule. Contributors are not identified individually. Companies and roles have been generalized.
Sources: clay.com/blog; writer.com/blog; First Round Review podcast with May Habib; arcticstartup.com and Business Insider coverage of Legora; startupriders.com Sierra Growth Playbook; CNBC, TechCrunch, and SiliconAngle coverage of Sierra’s May 2026 round; Sequoia Training Data podcast with Bret Taylor; Sacra and Bloomberg on Hugging Face; Acquired podcast with Clément Delangue.
Subscribe here to get the whole series.
Introduction: The playbook that isn’t a playbook
Every era of software has produced a canonical go-to-market motion. In the 2000s it was outbound enterprise sales. In the 2010s it was product-led growth stitched to a freemium funnel. Both worked because they were legible: a founder could copy the shape and know roughly what would happen next.
The AI-native era has broken that pattern. Five of the fastest-growing companies of the last three yearsHugging Face, Clay, Writer, Legora, and Sierra have almost nothing in common. One gave away its core technology. One sold to Fortune 500 CIOs from day one. One picked a single vertical and refused to leave it. One got Fortune 50 CEOs on the phone because of who its founder was. One never ran a sales team at all.
The temptation is to conclude there is no playbook that each of these is a founder-specific cheat code, unrepeatable by design. This document argues the opposite: that beneath five very different motions sits a smaller set of shared conditions and patterns that any AI-native founder can learn from, even when the specific unfair advantage isn’t theirs to copy.
It is written for the founder, operator, or investor who is building in the current market 2025 to 2027 and needs to make real decisions about category, pricing, distribution, and hiring before the ground moves under them again. It draws on desk research into the five companies, two facilitated group chats among AI-native operators, and Domain’s own live experiment building a collaborative AI product from zero. Domain is included at the end as a hypothetical sixth case, not a peer, a way of testing what happens when a founder tries to apply the emerging patterns while the ink is still wet.
The takeaway upfront: the playbook is not a set of moves. It is a set of decisions that AI-native companies are forced to make earlier and more explicitly than SaaS companies ever were. Name the category before the buyer can. Pick the buyer before the market picks one for you. Launch the business model with the product, not after it. Everything else is downstream of those three.
Part 1: Five case studies
Each case follows the same shape: one-line playbook, the numbers, the arc from zero to a hundred million, the unique moves that made it work, and what breaks when someone tries to copy it without the underlying conditions.
1. Hugging Face: Open-source community as distribution
The one-line playbook: give away the infrastructure everyone needs, become the neutral hub while every lab races to own proprietary models, then monetize the enterprise once they cannot operate without you.
The numbers: $70M ARR in 2023, up 367% year-over-year. Series D at $4.5B valuation, $396M total raised. Investors include Google, Nvidia, Salesforce, Sequoia, Coatue, Lux, IBM, Intel, AMD, and Qualcomm on the same cap table. 13M-plus users, 500K-plus organizations, more than 30% of the Fortune 500 with verified accounts.
Founded in 2016 as a chatbot for teenagers by Clément Delangue, Julien Chaumond, and Thomas Wolf. Consumer product failed. In 2018 they open-sourced the Transformers library, a Python wrapper around BERT, GPT-2, and the models that followed, which became the standard toolkit in every ML research lab within a year. That is the moment.
From 2019 to 2022 they extended a library into a platform: Model Hub, Datasets, Spaces. GitHub for AI models. When ChatGPT arrived in late 2022, every enterprise suddenly needed AI infrastructure. Hugging Face was already the neutral hub, not competing with OpenAI or Anthropic, hosting all of them plus the open-source alternatives. Enterprise revenue exploded. By 2023 the Series D put every major infrastructure incumbent on the same cap table.
The unique moves that mattered: giving away the load-bearing tool created a decade-long developer loyalty asset that no competitor could catch up to. Becoming Switzerland in a war refusing to pick a side between OpenAI, Anthropic, Google, and Meta, turned neutrality itself into the pitch. Enterprise wasn’t sold; it was inherited from the developers already using Transformers in production. Category ownership came through infrastructure ubiquity rather than marketing. And the Series D cap table was not fundraising but coalition-building.
What breaks the playbook: neutrality holds only while the labs let it. Free-then-monetize requires a decade of runway that most founders cannot access. And the specific open-source moment that created Transformers a new architecture family needing a standard wrapper, does not exist in every category.
2. Clay: Bottom-up screenshot virality
The one-line playbook: turn the product into content, let GTM operators sell it for you, then productize the community.
The numbers: $100M ARR in 2025, from $1M just two years earlier after six years of foundational work. Series B $40M at $1.25B (Jan 2025, following 6x growth in 2024). Later round: $100M at $3.1B. Enterprise net revenue retention over 200%. Never churned an enterprise customer.
Founded 2017 by Kareem Amin, Nicolae Rusan, and Varun Anand. For six years the product was a general-purpose spreadsheet-automation tool with roughly twenty customers paying two hundred dollars a month. Nearly died more than once. One operator in the July group chat, who knows Kareem from prior work, confirmed that “eight-year overnight success” undersells how close the company came to zero five near-death moments, not one.
The wedge arrived in 2023: GTM and sales teams using Clay for data enrichment and personalized outbound started posting screenshots on LinkedIn and Twitter. In 2024 revenue grew 6x. Screenshots became the marketing channel. Users posted absurd workflows (”I built a system that finds all Series A companies whose CTO went to my college and sends them a personalized email”) and every screenshot recruited more users. By 2025 Clay had crossed $100M ARR and formalized the community: a new job title”GTM Engineer”with a certification program, hundreds of agencies building on Clay, and self-organized user clubs in every major city.
Unique moves: naming a job category and becoming the credential. Thousands of people now list “GTM Engineer” on LinkedIna retention moat because you cannot fire someone whose entire job title is your product. Building an agency ecosystem so implementation happened without a Clay services team. Product as content, produced by users. Community loops built before sales loops.
What breaks the playbook: Clay’s motion depends on the buyer being technical enough to film a workflow. It doesn’t translate to categories where the buyer isn’t the operator, compliance, procurement, anywhere the person paying is not the person using.
3. Writer: Enterprise trust and governance
The one-line playbook: build the compliance and governance layer competitors ignored, sell to the office of the CIO, own the enterprise before the category exists.
The numbers: Series C $200M at $1.9B (Nov 2024). Series B $100M at ~$500M (early 2024). Founded 2020 by May Habib and Waseem Alshikh, who came out of Qordoba where they had been exposed to transformer models pre-ChatGPT.
From 2020 to 2022 Writer sold as a writing assistance and brand voice tool, finding customers among users of Grammarly Enterprise. Early anchor: Accenture at 1,500 users. When ChatGPT arrived, Writer repositioned overnight to sell to IT and “head of AI” buyers as the enterprise generative AI platform. From 2023 they locked in enterprise features from day oneSOC 2 Type 2, user tagging, permission inheritance, industry-specific expertise in financial services and healthcare, which became the moat. From 2024 onward they ran a classic land-and-expand from marketing use cases into customer experience, expert assist, and knowledge management.
Unique moves: enterprise-first from the start, when the rest of the AI cohort chased self-serve or SMB. Sold to the office of the CIO rather than the practitioner, when ChatGPT scared every enterprise IT team, Writer was the safe answer. Model quality wasn’t the pitch; risk posture was. Named the category “enterprise generative AI”before it existed, which won early Forrester and Gartner coverage because Writer was the only player big enough to name. And a disciplined refusal to sell to enterprises without a strong champion.
What breaks the playbook: as foundation models commoditize and enterprises get sophisticated buyers, “trust and governance” as a wedge shrinks. Writer’s next act depends on the platform actually differentiating.
4. Legora: Vertical reference-selling into law
The one-line playbook: pick one high-trust, relationship-driven vertical, land the most prestigious references first, let peer pressure do the selling.
The numbers: Founded 2023 in Stockholm by Max Junestrand, Anna Nordell-Westling, and Sigge Labor with $50K in angel funding. $100M ARR by fall 2025. Series C $150M at $1.8B (Oct 2025). Series D $550M at $5.55B (March 2026). 400-plus organizations across 40-plus markets.
Two Swedish founders in their twenties pitching AI to law firms, the world’s most reference-driven buyers. Managing partners call each other before signing anything. In 2024 they landed marquee references: Linklaters, Cleary Gottlieb, Goodwin, MinterEllison. Positioned as “the collaborative AI for law,” not another chatbot. Crossed $100M ARR in under two years.
Unique moves: treating customers as strategic collaborators rather than accounts”the lawyers built this with us” as both pitch and moat. Going for the top of the market first, against the standard vertical-SaaS instinct to start mid-market. Category framing”collaborative AI” vs. “legal AI”that positioned them against Harvey as the tool lawyers actually use together, not a black-box research assistant. And product depth over horizontal expansion: refusing to leave the legal vertical even at $100M ARR.
What breaks the playbook: it only works in industries where trust is peer-referred, buying is relational, and there’s an obvious top-of-market to anchor. It would not work for consumer, mid-market SaaS, or fragmented buyer sets.
5. Sierra: Founder-brand top-down enterprise access
The one-line playbook: leverage founder credibility to get Fortune 500 CEOs on the phone directly, build in stealth with four hand-picked design partners, publish an outcome-based pricing manifesto that reframes the category.
The numbers: Founded early 2023 by Bret Taylor and Clay Bavor. Stealth Series A $110M at just under $1B (Oct 2023). Series B at $4.5B with $20M ARR (Oct 2024). Series C $175M at $10B (fall 2025). Latest $950M at $15B (May 2026). ARR from $20M to $165M in just over a year. Serves 40% of the Fortune 50 two years after public launch.
Bret Taylor left Salesforce in late 2022, nerd-sniped by ChatGPT. He and Clay Bavor started Sierra in early 2023 and stayed in stealth for eleven months, recruiting four design partners chosen for the hardest constraints: WeightWatchers, SiriusXM, Sonos, OluKai. Public launch February 2024, immediately serving major enterprise brands. In December 2024 they published a pricing manifestoSierra only gets paid when the AI agent resolves the case, that reframed the entire category. Land-and-expand plays turned single customer-service deployments into $5M+ ACV.
Unique moves: founder brand as the GTM. Bret Taylor can call any Fortune 500 CEO. Deployment speed as a wedgeSierra ships in about five weeks against a 6-18 month industry norm. Outcome-based pricing as category redefinition: every competitor charges per seat or per query, Sierra charges per resolved case, which aligns incentives with the buyer and starves competitors of comparable metrics. Naming “agent operations” before Salesforce woke up. And selling to the CFO rather than the CX leader, using unit economics human call $10, Sierra $0.10 to $1as the frame.
What breaks the playbook: it requires a Bret Taylor or Bret Taylor-equivalent. Not a repeatable motion. And the outcome-pricing model depends on resolution rates staying high; if they drop, the whole model inverts.
Next Up: Part 2
Nine Patterns that Repeat
Don’t want to miss it? Subscribe today.