/ 01 — Right NowThe company you have not heard enough about
On any given Tuesday, an analyst at a Fortune 100 bank opens her laptop, types a question into something that looks vaguely like a chat window, and gets back a fully-cited research memo grounded in her firm's own private documents. She does not know the model's name. She does not need to. The thing works. The thing is Writer.
Writer is six years old, San Francisco-based, profitable in the ways that matter to enterprise software, and worth $1.9 billion at last count. It trains its own large language models - the Palmyra family - and stacks them with a knowledge graph, a guardrails layer, and a builder called AI Studio that lets a marketing ops lead ship an AI agent on a Wednesday afternoon. It is, in short, what the rest of the AI industry keeps promising to become.
And it got there without selling its soul to a single hyperscaler.
/ 02 — The ProblemWhat the chatbots could not fix
By 2022, every large company in America had the same headache. The board wanted "an AI strategy." The CIO wanted compliance. The CMO wanted output. The general counsel wanted nobody to feed customer data into a Discord channel. And the engineers, somewhere in the basement, wanted any of it to actually work.
The available answer was a public chatbot with a memory like a goldfish, a habit of inventing case law, and terms of service that lawyers read with the same enthusiasm reserved for tax audits. It was, generously, not enterprise-ready. It was, less generously, a liability with a friendly avatar.
The real enterprise AI gap
It was never about model quality. It was about whose data the model saw, whose policy it followed, and whether the legal team could sleep at night. Writer noticed this earlier than most.
— Filed under: things obvious in hindsightWriter's founders had been staring at this gap for years. May Habib, the CEO, had run a translation-tech company called Qordoba; she knew exactly how content moves through a global enterprise - which is to say, badly. Waseem AlShikh, the CTO, had built NLP systems back when "transformer" was still a Hasbro toy. They had a thesis: the enterprise did not need another model. It needed a stack.
/ 03 — The BetTwo founders, one unfashionable idea
The unfashionable idea was this: train your own models. Own the whole thing. Do not rent intelligence from a competitor. In 2020, when Writer started, this sounded like a startup trying to build its own chip foundry. Cute. Expensive. Probably doomed.
Habib and AlShikh did it anyway. They built Palmyra - named, fittingly, after an ancient Syrian trading city - from scratch. Then they did it again with Palmyra-Med for healthcare, Palmyra-Fin for financial services, Palmyra-X for general enterprise work. They tuned each one on the kind of dense, jargon-laced, regulation-heavy text that public models stumble over.
May Habib
Born in Lebanon, raised in Canada, formerly of Qordoba and the World Bank. Has been described, accurately, as under-the-radar.
Waseem AlShikh
NLP researcher. Wrote a lot of the first Palmyra model himself. Tends to use the word "graph" more than most.
/ 04 — The ProductA stack, not a chatbot
Writer's platform has four moving parts, none of which sound thrilling on a slide and all of which matter when you are running a global business.
Palmyra is the model layer - a family of LLMs that consistently rank near the top of enterprise benchmarks like Stanford HELM, while costing less to run than the household names. Knowledge Graph is Writer's answer to RAG: a graph-based retrieval system that grounds outputs in a company's actual data, not a hallucinated approximation of it. AI Guardrails is the layer your general counsel has been quietly praying for - policy, brand, and compliance checks baked into the model's outputs. AI Studio is the builder, the surface where non-engineers wire all of this into agents and workflows.
Together, they do something almost subversive: they make generative AI feel boring. Boring as in: reliable. Boring as in: deployable. Boring as in: the CFO stopped asking.
What you can actually do with it
Spin up an AI agent that drafts compliant marketing copy in twelve languages. Build a clinical-trial summarizer that lives inside the regulatory team's existing workflow. Stand up a finance research desk where every memo is grounded in your own filings. Ship a customer-support copilot that does not invent refund policies.
— Not a roadmap. A customer list./ 05 — The ProofThe customer list does the arguing
You can tell a lot about an AI company by who actually paid for it. The names in Writer's customer slide are the kind that make a startup founder spit out their coffee. Accenture - which also invested. Vanguard. L'Oreal. Uber. Intuit. Salesforce. Mars. Kenvue. Qualcomm. These are not "design partners." These are line items in a procurement system.
The Series C in November 2024 was co-led by Premji Invest, Radical Ventures, and ICONIQ Growth, with Salesforce Ventures, Adobe Ventures, B Capital, Citi Ventures, IBM Ventures, and Workday Ventures piling in. When the same enterprises that buy your product are also the ones writing checks, something has gone right.
/ 06 — The MissionOwning the stack on purpose
Writer's mission statement is unglamorous and, for that reason, credible. It is not curing cancer. It is not democratizing intelligence. It is bringing generative AI to the enterprise with full-stack control - models, data, and guardrails owned end to end.
That is the sentence. Read it twice. The whole company is built around the word owned. The model is Writer's. The data layer is Writer's. The compliance layer is Writer's. When a Fortune 100 deploys Writer, it is not stitching together five vendors and one prayer; it is buying one stack from one company that built all of it on purpose.
This is a boring position. It is also, increasingly, the winning one.
Culture, briefly
Writer is remote-first, technical, and noticeably allergic to AI theater. Its engineering team writes papers; its marketing team writes case studies; its CEO writes posts that are mostly about customers, occasionally about models, and almost never about Sam Altman.
— A radical departure/ 07 — TomorrowWhy this gets more interesting, not less
The next phase of enterprise AI will not be won by whoever has the biggest model. It will be won by whoever can hand a regulated industry a system that is grounded, governed, and accountable - and have the General Counsel sign off without three weeks of red-lining.
Writer has been building toward that, brick by unsexy brick, since 2020. The agent stuff is real. The model stuff is real. The customer stuff is real. The thing that is missing - the part where the rest of the AI industry catches up - is mostly a matter of time, and time is exactly what Writer has been buying.
Back to that analyst at the Fortune 100 bank. She closes her laptop. The memo is filed. Compliance signed off in the background. She does not know which model wrote it. She does not need to. The thing works. The thing is Writer. Filed: 2026