But there is still some sort of chat component in writer apps.

When you interact with a traditional chatbot, you first have to “train” it by telling it what you want. Writer’s apps take that step away from you. So, for example, if you want to generate headlines for blog posts, you can tell a Writer app to think like an editor. You can then show it examples of headlines you like and instruct it to create headlines that are succinct and grab attention. Once you’ve established the ground rules with the app, you can just enter details about your article and it will generate headlines without you having to chat any further afterward.

“It’s becoming more and more prescriptive about what exactly an end user needs to give the generative AI tool and what results will come out of it,” Habib said.

Ask Writer and AI Studio

Clients of San Francisco-based Writer – including consulting firm Accenture, financial software company Intuit, cosmetics brand L’Oreal, digital music service Spotify and ride-sharing platform Uber – can access apps in two ways.

You can use Ask Writer, an out-of-the-box chatbot that the spokesperson described as “like ChatGPT, but designed specifically for businesses.”

It also includes around 25 ready-made apps that cover the entire spectrum from generating transcript summaries to creating sales emails and error messages to conducting image analysis.

Alternatively, customers can use AI Studio, which allows them to build their own apps using either a no-code development tool or through APIs that allow developers to create new applications using Writer technology, as well as an open-source app development framework, which is a library of tools and code that developers can also use to build apps.

According to Writer, this allows employees to collaborate on custom apps regardless of their technical knowledge level.

“At Writer, we say, ‘If you can write it, if you can describe it, you can build it,’” Habib said.

LLMs and RAG

These apps rely on Writer’s own large language models, AI models that are trained on huge data sets—though the speaker didn’t reveal what exactly they’re trained on—to understand and generate content. LLMs form the basis of popular chatbots like OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude.

Estimates from Gen-AI startup Vectara show that mainstream models such as GPT-4o, Llama 3.1 405B and Gemini 1.5 Flash hallucinate or generate false or misleading content between 1.3 and 7.4% of the time.

The author believes that this is not good enough for companies that require consistently accurate results.

The spokesperson said Writer’s LLM family will be evaluated by Stanford’s Holistic Evaluation of Language Models, which judges AI models based on criteria such as accuracy, general information and bias.

In July, Writer added LLMs specifically for healthcare and finance.

The healthcare model was trained on patient medical records so it could analyze them to support clinical decision making. The financial model was trained on how to rebalance a portfolio so it could do just that for its clients.

“If we can work backwards from the use cases we know customers need, we can train the models really effectively and efficiently,” Habib said.

Writer also has a general-purpose model and a model for visual analysis. The visual model can analyze images such as charts and graphs, as well as sketches or handwritten notes.

In addition to these LLMs, Writer offers RAG or Retrieval Augmented Generation, a type of natural language processing for questions.

“When a question is asked, it retrieves contextually relevant data points from a large corpus of uploaded data and passes them to LLMs to generate an accurate answer,” Writer said in a blog post.

The spokesperson said this will allow Writer to take all of a company’s data and link it to the LLM. The startup allows you to download files of your own data of up to 10 million words (about 20,000 pages) to ask questions, conduct research or generate results.

“The value is in your own company’s information and data,” she said. “And much of the industry has really struggled to develop RAG solutions that are accurate enough to be used by an enterprise.”

The spokesperson did not elaborate on whether Writer’s RAG offer is more accurate.

Writer also offers AI guardrails to help ensure the app does not violate any legal or ethical rules.

From Qordoba to Writer

A basic plan starts at $18 per user per month for up to five users. Writer offers custom pricing for enterprise users.

Habib and Writer co-founder and CTO Waseem Alshikh began working together in 2013 on natural language processing, the branch of AI that enables machines to understand human language, and machine translation, which uses AI to translate text from one language to another. In 2015, they founded AI writing assistant Qordoba.

In 2017, the research paper “Attention is All You Need,” published by researchers from Google and the University of Toronto, introduced the idea of ​​transformers in machine translation. Transformers are a type of neural network or machine learning model that works like the human brain, transforming an input into an output by learning the context and tracking the relationships between words.

Habib and Alshikh decided to move from pure language translation to translating business content into more usable content using this transformer-based approach. In 2020, they renamed Qordoba to Writer and started working on LLMs.

Writer has raised $126 million to date, including $100 million in a Series B funding round in September.

Investors include venture capital firms Iconiq Growth, Balderton Capital, Insight Partners and Aspect Ventures, investment firm WndrCo, and Writer clients Accenture and Vanguard.

“Always plan for more,” Habib said of future financing.