The technology media tends to focus on the latest foundation models for AI, products like ChatGPT, Gemini, Claude, DeepSeek etc. These are general-purpose models that can be adapted to specialist uses. What gets less attention in the press is that a whole industry of such specialist tools has sprung up to take advantage of the capabilities of these core foundation models. Typically, they are tailored to a particular industry or a specific use case. Let’s have a look at what tools are emerging and where they are making an impact.
One of the most common uses for large language models (LLMs) has been to produce marketing copy. LLMs are fluent writers and can produce product descriptions and brochures in moments, something that was previously the domain of human copywriters. An example of a successful add-on tool in this area is Jasper from venture-backed start-up Jasper.ai in Texas, which fine-tunes foundation models in a way that is tailored specifically to marketing. Jasper can learn a company’s brand values and style, and has prebuilt tools for campaign planning, SEO and templates for blogs and adverts. Though a private company with no public financial figures, sources like TechCrunch estimate 2023 revenues for Jasper.ai at around $120 million, rising to $145 million in 2024.
Elicit is a research tool produced by Ought, a non-profit organisation. It uses underlying LLMs to scan large research databases. It can summarise papers, extract key data and generate literature review summaries, comparing evidence across sources. By providing and cross-checking multiple resources, it endeavours to overcome the thorny problem of LLM hallucination, which plagues LLM foundation models. It provides predefined steps for search, screening, extraction, and synthesis, rather than a researcher having to reinvent the workflow in prompts every time. It has half a million registered users and 60,000 monthly active (paying) users. Again, Ought is a private organisation, but Elicit probably generates a few million dollars in annual revenue.
Grammarly has about 30 million daily users and generates over $250 million in annual revenues for its owners, based in San Francisco. Grammarly is a writing assistant that helps draft, improve and check your writing. As well as spell checking, it can suggest improvements to sentence tone and clarity, and integrates with tools like Word, Slack and Gmail. It keeps track of your past writing and learns your personal style of writing, which stand-alone LLMs do not (every fresh session with a standard LLM is a brand-new experience). Grammarly can also enforce style guidelines across a company. With its huge user base, it is one of the more successful AI add-on products.
Another tool in the writing space is Notion AI. This is aimed at organising project notes, brainstorming ideas and document summaries. It is more in the project management and knowledge work area than a tool to check grammar. Notion has about 20 million active users, and its owner, Notion Labs Inc did around $400 million in revenue in 2024, up from $250 million in 2023. It is a venture-backed start-up, headquartered in San Francisco.
The education field is another with a range of AI tool add-ons. Quizlet produces flash cards, practice quizzes and study games, with around 50 million users. Duolingo teaches languages and is a product with ten million paid subscribers and $748 million in revenue in 2024. It is a publicly traded company listed on NASDAQ. Khanmigo from Khan Academy is a structured teaching tool, a sort of interactive tutor, with 1.4 million users. Tools in this sector have to overcome the issue of notoriously tight education budgets in most countries, and pitch to a mix of institutional buyers and consumers.
The legal field is another that has eagerly adopted AI, often with troublesome results. Specialist tools here include LexisNexis, a legal research tool making $1.7 billion in revenue that has added AI features. It competes with Westlaw from Thomson Reuters, another large legal research tool. Pure AI start-up Harvey goes beyond research to drafting and reviewing legal documents, the product earning over $50 million in annual revenue. There are many smaller start-ups in the field, though the legal field is one where precision and accuracy are crucial, and there have already been hundreds of examples of where lawyers’ misplaced trust in the accuracy of AI tools has led to faulty, hallucination-ridden court submissions, sometimes leading to sanctions for the lawyers involved.
The software industry is one of the poster child use cases for AI, with vendors claiming to be able to write code as effectively as software engineers. “Vibe coding”, where non-developers invoke an LLM to produce a complete software application, has become a trend in itself. While some of the claims may be overblown, there is no doubt that the ability of LLMs to code has greatly improved, especially towards the end of 2025, with the well-received release of Claude Opus 4.5, compared to earlier versions. This trend has seen a rich vein of AI add-ons like Cursor, Copilot and more. GitHub Copilot from Microsoft is the most widely adopted, a coding assistant that suggests code completions, generates functions and answers developer questions. It has around 15 million paid seats, with estimated annual revenues of about $2 billion. Cursor from Anysphere had around $1 billion in annual revenue in 2025. Replit AI, a web-based integrated development environment, generated annual revenue of around $125 million in 2025. There are yet more such products, such as Amazon CodeWhisperer, Tabnine and Codeium.
These examples merely scratch the surface of the AI add-on tool market, which is reaching into fields as diverse as medicine, life sciences, finance and advertising. How large is this market overall, and how does it fit into the AI ecosystem?
The AI tool add-on market is already huge, estimated to be generating about $25 billion in revenue in 2024, and growing fast. This is actually about twice the size of the foundation tool market, which was around $14 billion in size in 2024. The latter is quite hard to accurately estimate, since it contains specialists like Anthropic (with Claude) and OpenAI (with ChatGPT), but also broader companies like Google (with Gemini) and Meta (with Llama). There is already a thriving tools market for AI add-ons, piggybacking on the core capabilities of foundation model LLMs. There is also a thriving market of consultants implementing these tools (with maybe $8 billion in 2024), and chipmakers such as Nvidia, whose GPU processors are the devices where most AI queries actually run today.
While there is investment and growth in revenues in the AI tool market, what is less clear is how profitable all of this actually is. Certainly, the foundation model vendors are not making a profit, and may not do so for many years. Anthropic projects profits in 2028, OpenAI in maybe 2029/2030, and this all assumes continued rapid and uninterrupted growth. The AI tools vendors, who have lower research and development costs, are generally healthier, with some already profitable and many at least with profitability in sight. Consultants, of course, are generally profitable, as they can quickly adjust manpower in the light of the level of business coming in. However, all these rosy projections assume a steady state of growth, and the global economy could easily provide some nasty surprises that may shred some of these assumptions. Certainly, there has been a major stock market sell-off of AI-related technology companies in February 2026, as investors have begun to question the wisdom of the vast AI-related capital expenditure programs of companies like Google, Oracle and Amazon. An economic crash, which some predict, would put considerable strain on the economics of the burgeoning AI industry. The World Economic Forum Survey published on January 16th 2026, found that 52% of those surveyed expected US stocks to fall in value in 2026. If that majority is right, the impact on the AI industry could be unpleasant. Ironically, it may be that enduring profits may accrue not to the LLMs that dominate attention and capital spending, but to the specialised, vertical tools built on top of them.
