The most innovative companies in applied AI for 2025

It’s been gradual, but generative AI models and the apps they power have begun to measurably deliver returns for businesses. Organizations across many industries believe their employees are more productive and efficient with AI tools such as chatbots and coding assistants at their side. 

Numerous AI startups found traction offering such solutions during 2024. Glean, for example, puts cutting-edge AI search capabilities in the hands of employees so that they can tap into various apps and platforms to find documents and corporate intelligence. Contextual AI lets organizations put a company’s proprietary intelligence into a secure data store, then lets them build AI apps that can call on that data. Enterprises are also using AI apps to protect softer corporate assets, such as reputation. Blackbird.AI offers a web app that enterprises use to monitor how their brand name is portrayed in social media posts, videos, links, and memes. 

Other standout AI apps focuse on specific industries. Google DeepMind put drug discovery ahead by years when it improved on its AlphaFold model, which now can model and predict the behaviors of proteins and other actors within the cell. Harvey has found its legs within the legal industry by offering an AI legal assistant that can write briefs, summarize and compare cases, and more. 

Coding assistants grew considerably–both in capability and usage–during 2024. Anysphere’s Cursor tool, for example, helped advance the genre from simply completing lines or sections of code to building whole software functions based on the plain language input of a human developer.

1. Glean

For arming employees with the tools to get their jobs done

Companies contain a lot of information that’s crucial for employees to know, but it’s spread out across an array of workplace apps: Slack, Microsoft 365, Google Workspace, Salesforce, and more. Five-year-old Glean offers a user-friendly AI-powered search tool that allows employees to find information and generate answers across more than 100 data sources.

In June 2024, the company transformed its existing enterprise AI assistant and search engine into a platform called Work AI platform. It allows employees of all technical backgrounds to quickly generate personalized, accurate answers—and even create their own no-code tools to make the agents work better for the specifics of their jobs and businesses. The Work AI suite also includes a Glean Actions tool, which enables the AI assistant to directly take action on an employee’s behalf within a company’s connected applications. Actions could involve reading data from an application and executing a specific task, creating Jira tickets, publishing new content, or searching for code.

In September, Glean doubled down on its user-friendly proposition by making it even easier for non-technical employees to get the most from its tools with next-generation prompting features. These features include a Prompt Builder feature, which allows users to create their own directives for the AI assistant, and a Prompt Library, which includes suggested prompts from Glean, as well ones that a company has shared in its own prompt library.

According to a November 2024 report in The Information, Glean was generating around $100 million in annual recurring revenue, more than tripling that metric over the past year. The company closed two funding rounds in 2024: $200 million in February and $260 million in September at a $4.6 billion valuation. Its 200+ customers include Reddit, Instacart, Pinterest, Duolingo, and Databricks.

Read more about Glean, honored as No. 6 on Fast Company’s list of the World’s 50 Most Innovative Companies of 2025.

2. Anysphere

For giving developers a coding partner with contextual awareness

Cursor AI has emerged as a standout in the growing field of AI code editors. The company behind it, Anysphere, made the smart design choice of building the UX based on Microsoft’s Visual Studio Code, a familiar programming environment. Cursor also can access a developer’s or company’s existing code base as a way of fine-tuning code suggestions.

Cursor acts like a coding partner that’s aware of the context in which code is being created. It offers code auto-completions, and not just of single lines–it can generate entire sections of code, and then explain the reasoning behind them. Or the developer can explain a new feature or function in plain language and the AI will code a prototype of it.

Anysphere says Cursor now has more than 40,000 customers. Developers in online forums say that after using Cursor, they can’t go back to GitHub Copilot. In August, the startup raised a $60 million A round at a $400 million valuation from Andreessen Horowitz, Thrive Capital, OpenAI, Google’s Jeff Dean, OpenAI’s Noam Brown, and the founders of Stripe, GitHub, Ramp, and Perplexity. Four months later it raised a Series B round that closed in January 2025 with $105 million invested, raising its valuation to $2.6 billion.

Read more about Anysphere, honored as No. 26 on Fast Company’s list of the World’s 50 Most Innovative Companies of 2025.

3. Blackbird.AI

For arming NATO and others with AI that detects AI disinformation

Businesses and other organizations must constantly be aware of how their name is being used in the digital environment, and be able to react quickly if their brand and reputation are distorted by misinformation or disinformation. In 2024 Blackbird.AI released its Compass platform, which lets individual users reality check or get greater context around suspicious claims made in social media posts, videos, links, or memes. The user pastes the content into the Compass tool, which checks it against thousands of trustworthy online sources.

In October 2024 Blackbird.AI launched “Compass Vision,” a new AI-based tool for identifying AI deepfake images and videos. Customers can access Compass Vision directly through Blackbird.AI’s platform, or they can use its API to integrate it with their existing threat intelligence and social listening systems.

As companies continue to more aggressively protect their market value and brand equity in the digital space, the market for “narrative intelligence” services is likely to grow to as much as $70 billion annually, Blackbird.AI believes. The company has already positioned itself well, and expects its revenues to double or triple over the next year. Blackbird.AI has raised more than $20 million in venture capital so far.

4. Google DeepMind

For unfolding the mysteries of structural biochemistry

Google DeepMind CEO Demis Hassabis and director of research John Jumper won the 2024 Nobel Prize in Chemistry for their parts in discovering and developing the AlphaFold AI models, which can predict the complex structures of virtually all known proteins. Proteins control and drive all chemical reactions within the bodies of organisms, including humans, so the tool is of great interest to researchers in drug development, material science, and environmental science.

In 2024 DeepMind expanded its AlphaFold AI system to model how proteins interact with other cell structures, including DNA, RNA, and small molecules that are often used in drugs. The new system, called AlphaFold 3, can model the ways in which proteins “read” our DNA and then carry out the instructions in the body. It lets drug researchers quickly model how new drug compounds might react with certain receptor sites in the body, which could accelerate the exploratory phase of drug development. Traditionally this work has been done experimentally, in a wet lab.

Google DeepMind developed AlphaFold 3 in collaboration with London-based Isomorphic Labs, an AI-based drug discovery lab it spun out into an independent unit within Google parent company Alphabet. Isomorphic is now working with Novartis and Eli Lilly. While drugs based on AlphaFold’s breakthroughs are yet to come, it’s already instigated a revolution in structural biochemistry.

5. DeepL

For translating everyday business communications into 33 languages, including traditional Chinese and Arabic

Fast and accurate translation are crucial for multinational corporations, and generative AI has been a natural complement to existing services. One service provider, DeepL, has emerged as a standout for the accuracy and cost-effectiveness of its AI. The company already serves more than 150,000 businesses, governments, and other organizations across the legal, tech, media, manufacturing, and retail industries, including known names such as Nikkei, Panasonic Connect, Zendesk, and Morningstar.

In November 2024 the company released DeepL Voice for Meetings, which lets participants speak in their preferred language during meetings and video calls, with real-time captions of others’ comments in their chosen language. DeepL Voice for Conversations does the same thing, but for one-on-one conversations happening on mobile devices. In July 2024, DeepL deployed a new large language model of its own, which its says significantly outperforms LLMs from OpenAI, Google, and Microsoft for translation. DeepL also added Traditional Chinese and Arabic languages to its platform, bringing total supported languages to 33. Beyond translation, DeepL launched a new tool called DeepL Write, designed to help professionals improve their business writing skills.

In May 2024 DeepL raised a $300 million funding round and saw its valuation rise to $2 billion–doubling its valuation after its previous funding round a year earlier.

6. Perplexity

For turning generative AI into a rival to traditional search, especially for election coverage

Perplexity is one of the biggest success stories of the current AI boom. Its “answer engine” has revitalized web search, using a combination of homegrown large language models (LLMs), third-party models (from OpenAI, Anthropic, DeepSeek and others) and web crawlers to return custom answers that are highly relevant and fastidiously cited. 

The San Francisco-based company saw its user base grow throughout 2024, as it added new features and functions to its platform. Perplexity began experimenting with ads and referrals on its platform late in 2024, and launched “Shop like a Pro”, an AI-powered shopping assistant that lets users research and even purchase products within Perplexity. Users of the Perplexity mobile app users can even snap pictures of items to see related products and buying information.

Perplexity’s most surprising creation during 2024 may have been the AI-powered Election Information Hub it launched before the November 2024 U.S. elections. The hub offered voters real-time updates, candidate information, and ballot measure summaries, along with AI-generated analysis based on reliable data from The Associated Press and Democracy Works. Perplexity’s clear, verifiable approach to election coverage gained significant attention during the run-up to the elections.

7. Contextual AI

For making the next generation of AI more accurate and efficient

To avoid hallucinations and keep answers on point, AI developers use what’s called retrieval-augmented generation (RAG), where large language models are fed relevant information used to respond to specific queries. Cofounded by CEO Douwe Kiela, who pioneered RAG while at Meta, Contextual AI emerged from stealth mode in June 2023 with a mission to use the power of RAG to build more accurate LLMs for enterprises.

In March 2024, the company introduced RAG 2.0, a system that trains LLMs and the RAG ystem together, resulting in Contextual Language Models that are tuned for specific purposes for which they outperform leading commercial and open source models. Contextual also worked with the Allen Institute for Artificial Intelligence to develop OLMoE, an AI system introduced in September that uses an approach called “mixture of experts,” in which a specialized subsection of the model is called on to answer a given question, leaving most of the other model parameters at rest. This can increase a model’s accuracy while making it faster and more energy efficient.

Contextual, which counts HSBC, Qualcomm and The Economist as customers, announced in August 2024 that it raised an $80 million Series A round to fund further development of its enterprise-grade AI systems.

8. Harvey

For giving lawyers a trustworthy AI agent

Using large language models in legal work has long been a tantalizing possibility. But fears over inaccuracies due to AI hallucinations have caused the legal field to move slowly in their adoption of them. Harvey has managed some real innovations aimed at making its AI legal assistant more reliable and transparent about its work. The Harvey Assistant can read and analyze cases and other data faster than a human lawyer or legal assistant, and can draft documents, analyze information, and answer questions about hundreds of complex legal files.

Harvey gave the Assistant some important upgrades in August 2024. It added a series of specialized modes tailored to different kinds of legal work, it trained its AI to refine and expand on its initial responses, and it improved the system’s outputs and processing speed. Harvey says the new version of Assistant reduces AI hallucinations by 60% and improves the accuracy of cited sources by 23%. In a move to demystify the way the assistant comes up with its answers, Harvey released a report called “BigLaw Bench” describing its model training and evaluation methodologies.

The company more than tripled its employees during 2024, and added more than 100 new customers in 15 countries. With a new $100 million funding round in July 2024, the company saw its valuation rise to $1.5 billion.

9. Khan Academy

For empowering students and their teachers with a free AI writing coach

In 2023, Khan Academy launched Khanmigo, an experimental AI tutor designed to give students one-on-one help in tasks such as practicing math problems, brainstorming project ideas, and analyzing literature. Since then, thousands of students and teachers have started using the tool, and Khan Academy has been busy adding new tools and features to the platform.

During 2024, the non-profit built some important new functionality into its Khanamigo Writing Coach, which had originally been designed to act as a writing tutor for students. The tool now helps teachers, too, giving them access to detailed reports on student progress. It generates high-level insights on the writing challenges of individual students, and even helps teachers identify difficulties that multiple students in the class are facing. All this addresses a real challenge in English education–the time constraints teachers face when providing feedback on writing.

Khan Academy cites a real-life example of a teacher with 100 students who typically needed 17 hours to review the first drafts of a two-page essay assignment, assuming the teacher spent 10 minutes on each paper. In an era when AI offers to do our writing for us, Khanamigo Writing Coach is instead focused on helping students break through barriers to effective written communication—and on helping teachers guide them along the way.

10. Speak

For supercharging English-language learning with live conversational roleplays

Speak’s AI English tutor app, which is widely used in Korea and Japan, has been around for years. But the company took a big step forward during 2024 with a little help from OpenAI. The app already offered an English tutor that could teach and converse, but the conversations felt slow and unnatural. That’s because Speak’s system had to transcribe the user’s speech, run it through a text-based LLM workflow, then synthesize the AI character’s speech. Each of these steps created gaps and errors in the back-and-forth, which was disruptive to the learning process.

Then Speak became one of the first companies to get access to OpenAI’s Realtime API, using it to power its “Live Roleplays” feature. The Realtime API is unique in that it’s powered by AI that treats both text and voice in the same way–as common tokens within a multi-modal model. So no conversions are necessary, making the model’s response time super-fast.

As a result, Speak’s tutor can generate its voice responses with almost no delay. This makes exchanges between student and AI tutor feel much more fluid and natural. And that’s very important, because the best and fastest way to learn another language is through real life conversations. Speak may not be exactly real-life, but with OpenAI’s help it’s a lot closer to real-time.

11. Pika

For making a state-of-the-art video generator that’s accessible to nonprofessionals

Pika’s founders, Demi Guo and Chenlin Meng, dropped out of Stanford’s artificial intelligence PhD program in 2023 to pursue a big idea. The world needed a world-class video generation app that was designed for regular people–not just professional creators, film-makers and AI early adopters.

The duo and