1
Country: Sweden | Funding: $522.5M
Lovable develops a vibe coding platform that allows users to build websites and apps from text prompts, including the UI/frontend (often using the popular UX coding tool React) and database connectivity like Supabase. Chat mode helps you think through problems, debug errors, and plan your product directly within Lovable. Agent mode automates tasks such as editing code after reading project files or debugging. Lovable integrates shared connectors, personal connectors (MCP servers), and any API to add capabilities to deployed apps, provide context during app creation and securely call custom or third-party APIs. The service charges a per-use model: the more tasks requested from chat/agent, the more credits are deducted from the account. The company's clients include large software companies such as Klarna, Uber and Zendesk.
Lovable develops a vibe coding platform that allows users to build websites and apps from text prompts, including the UI/frontend (often using the popular UX coding tool React) and database connectivity like Supabase. Chat mode helps you think through problems, debug errors, and plan your product directly within Lovable. Agent mode automates tasks such as editing code after reading project files or debugging. Lovable integrates shared connectors, personal connectors (MCP servers), and any API to add capabilities to deployed apps, provide context during app creation and securely call custom or third-party APIs. The service charges a per-use model: the more tasks requested from chat/agent, the more credits are deducted from the account. The company's clients include large software companies such as Klarna, Uber and Zendesk.
2
Country: USA | Funding: $56M
Inception is building diffusion generative models for code and text. Diffusion models are mostly used to power image-based AI systems like Stable Diffusion, Midjourney, but Inception says that diffusion-based LLMs can be also much faster and efficient than auto-regression (GPT) models. Where auto-regression models have to execute operations one after another, diffusion models can process many operations simultaneously. Unlike GPT that work sequentially and predict word by word based on the context, diffusion models modify the overall structure of a response incrementally until it matches the desired result. They also have more flexibility in how they utilize hardware. Inception's Mercury model, designed for software development is already integrated into a number of development tools.
Inception is building diffusion generative models for code and text. Diffusion models are mostly used to power image-based AI systems like Stable Diffusion, Midjourney, but Inception says that diffusion-based LLMs can be also much faster and efficient than auto-regression (GPT) models. Where auto-regression models have to execute operations one after another, diffusion models can process many operations simultaneously. Unlike GPT that work sequentially and predict word by word based on the context, diffusion models modify the overall structure of a response incrementally until it matches the desired result. They also have more flexibility in how they utilize hardware. Inception's Mercury model, designed for software development is already integrated into a number of development tools.
3
Country: USA | Funding: $896M
Cognition AI develops Devin, the first AI-powered software engineer. Similar to GitHub's Copilot, Devin can help develop coding project plans, answer code questions with references and create wikis for code with documentation. Devin explores your codebase and helps clean up the backlog, modernize the codebase and supports the development of new projects. It performs unit and end-to-end testing, builds SaaS integrations, automates call response, ticket resolution and documentation generation. Devin is designed for collaboration and can adapt to unique workflows and learns your tribal knowledge. Devin also connects to MCP servers, from Asana to Zapier to automate the whole workflow. Devin Enterprise version is deployed in a virtual private cloud and provides enterprise-grade security and privacy.
Cognition AI develops Devin, the first AI-powered software engineer. Similar to GitHub's Copilot, Devin can help develop coding project plans, answer code questions with references and create wikis for code with documentation. Devin explores your codebase and helps clean up the backlog, modernize the codebase and supports the development of new projects. It performs unit and end-to-end testing, builds SaaS integrations, automates call response, ticket resolution and documentation generation. Devin is designed for collaboration and can adapt to unique workflows and learns your tribal knowledge. Devin also connects to MCP servers, from Asana to Zapier to automate the whole workflow. Devin Enterprise version is deployed in a virtual private cloud and provides enterprise-grade security and privacy.
4
Country: USA | Funding: $775M
Harness is developing a platform for creating AI agents for the “after-code” phase of software development, which includes testing, security checks, verification and deployment - processes which consume nearly 70% of development cycle time. The startup aims to solve the problem of uncontrolled code risks caused by the use of AI for coding. The Harness platform is built on a software development knowledge graph that maps code changes, services, deployments, tests, environments, incidents, policies and costs. The company claims the knowledge graph gives the system a deep understanding of each client's software development processes and architecture.
Harness is developing a platform for creating AI agents for the “after-code” phase of software development, which includes testing, security checks, verification and deployment - processes which consume nearly 70% of development cycle time. The startup aims to solve the problem of uncontrolled code risks caused by the use of AI for coding. The Harness platform is built on a software development knowledge graph that maps code changes, services, deployments, tests, environments, incidents, policies and costs. The company claims the knowledge graph gives the system a deep understanding of each client's software development processes and architecture.
5
Country: USA | Funding: $3.4B
Anysphere was the first company to create an AI-based IDE for software development - Cursor. Developers can use AI-powered auto-completion (for both individual strings and entire classes) and ask side-chat to create or fix a specific function, class or entire file (what later became known as vibe-coding). They can also create AI agents that independently perform simple coding tasks. Cursor relies on external AI models from companies including Google, OpenAI and Anthropic to power its platform, but plans to gradually implement its own model. Developers can use different LLMs for different tasks and pay only for Cursor (there's no need to purchase a subscription for each LLM).
Anysphere was the first company to create an AI-based IDE for software development - Cursor. Developers can use AI-powered auto-completion (for both individual strings and entire classes) and ask side-chat to create or fix a specific function, class or entire file (what later became known as vibe-coding). They can also create AI agents that independently perform simple coding tasks. Cursor relies on external AI models from companies including Google, OpenAI and Anthropic to power its platform, but plans to gradually implement its own model. Developers can use different LLMs for different tasks and pay only for Cursor (there's no need to purchase a subscription for each LLM).
6
Country: USA | Funding: $2.1B
Reflection AI, originally created as a coding agent platform, now positions itself as an open-source alternative to leading AI companies like OpenAI and Anthropic, and as a Western counterpart to Chinese AI companies like DeepSeek. It is building a large-scale LLM and reinforcement learning platform capable of training massive mixed-expert (MoE) models - the technology that has enabled DeepSeek to significantly reduce training and inference costs. Reflection AI will publish model weights -the fundamental parameters that determine an AI system's performance, for public use, while the datasets and all training processes will remain proprietary. The company aims to generate revenue from large enterprises building products based on Reflection AI models, and from governments developing "sovereign AI systems".
Reflection AI, originally created as a coding agent platform, now positions itself as an open-source alternative to leading AI companies like OpenAI and Anthropic, and as a Western counterpart to Chinese AI companies like DeepSeek. It is building a large-scale LLM and reinforcement learning platform capable of training massive mixed-expert (MoE) models - the technology that has enabled DeepSeek to significantly reduce training and inference costs. Reflection AI will publish model weights -the fundamental parameters that determine an AI system's performance, for public use, while the datasets and all training processes will remain proprietary. The company aims to generate revenue from large enterprises building products based on Reflection AI models, and from governments developing "sovereign AI systems".
7
Country: USA | Funding: $626M
Poolside claims to be developing AGI for enterprise, but starting with software agents. Because (according to the founders) creating software requires understanding and creating a model of the world, as well as the ability to reason and plan. In other words, software development largely depends on human intelligence. Meanwhile, Poolside is developing code-generation model and agents for creating enterprise software. The main idea is to train the codeing model not only on code itself, but also on the enterprise's business data. Thus, the coding assistant provides suggestions or autocompletes code in the context of the given business (and not just in the context of programming practices). Poolside provides intelligence at every interface: through the terminal, API, agents and, of course, IDE. However, Poolside's philosophy is that software development should take place outside the editor. The company's clients are primarily Global 2000 companies and government agencies.
Poolside claims to be developing AGI for enterprise, but starting with software agents. Because (according to the founders) creating software requires understanding and creating a model of the world, as well as the ability to reason and plan. In other words, software development largely depends on human intelligence. Meanwhile, Poolside is developing code-generation model and agents for creating enterprise software. The main idea is to train the codeing model not only on code itself, but also on the enterprise's business data. Thus, the coding assistant provides suggestions or autocompletes code in the context of the given business (and not just in the context of programming practices). Poolside provides intelligence at every interface: through the terminal, API, agents and, of course, IDE. However, Poolside's philosophy is that software development should take place outside the editor. The company's clients are primarily Global 2000 companies and government agencies.
8
Country: USA | Funding: $465.1M
Magic develops advanced coding models for automated software development. It acts as a programmer's assistant, seeking to understand and continually learn more about the project's context and helps with writing, reviewing, debugging and planning code changes. Magic considers that its advantage over competitors like GitHub Copilot and Antrophic Code is its extremely long model context window (100 million tokens). That is why their model architecture is called a "long-term memory network" (LTM). Instead of relying on fuzzy memorization, our LTM (Long-Term Memory) models are trained to reason on up to 100M tokens of context given to them during inference. Magic partners with Google and NVidia to build "coding supercomputers" based on the Google Cloud Platform.
Magic develops advanced coding models for automated software development. It acts as a programmer's assistant, seeking to understand and continually learn more about the project's context and helps with writing, reviewing, debugging and planning code changes. Magic considers that its advantage over competitors like GitHub Copilot and Antrophic Code is its extremely long model context window (100 million tokens). That is why their model architecture is called a "long-term memory network" (LTM). Instead of relying on fuzzy memorization, our LTM (Long-Term Memory) models are trained to reason on up to 100M tokens of context given to them during inference. Magic partners with Google and NVidia to build "coding supercomputers" based on the Google Cloud Platform.
9
Country: USA | Funding: $252M
Augment creates an AI platform for developers that includes chat, AI agent and Next Edit (code completion tool). The company claims that their AI assistants better understand programmers' intent than GitHub Copilot for example. Augment's autonomous software agents can run in various IDEs and in the cloud. The agent has a Memories feature, which is automatically updated during development and persists between conversations to continuously improve code generation, solve problems faster and match developer's style and coding patterns. The model supports MCP link to GitHub, Jira, Confluence, Notion, Linear, Vercel and Cloudflare. Augment's models were trained on publicly available data, some of which may be copyrighted or have a restrictive license.
Augment creates an AI platform for developers that includes chat, AI agent and Next Edit (code completion tool). The company claims that their AI assistants better understand programmers' intent than GitHub Copilot for example. Augment's autonomous software agents can run in various IDEs and in the cloud. The agent has a Memories feature, which is automatically updated during development and persists between conversations to continuously improve code generation, solve problems faster and match developer's style and coding patterns. The model supports MCP link to GitHub, Jira, Confluence, Notion, Linear, Vercel and Cloudflare. Augment's models were trained on publicly available data, some of which may be copyrighted or have a restrictive license.
10
Country: USA | Funding: $243M
Codeium is making AI-powered programming platform that maximizes developer productivity. The creators say that, unlike competing AI coding systems that provide generic code snippets and require significant effort to integrate and secure into existing codebases, Codeium's model makes recommendations within the context of an application's entire codebase and can be installed on company's own server. It supports approximately 70 programming languages and integrates with a number of popular development environments, including Microsoft Visual Studio and JetBrains. Codeium also offers a generous free plan and this strategy works fine: the startup has over 1000 enterprise clients, including Anduril, Zillow and Dell. Codeium's model is trained on publicly available code samples, so the company takes steps to remove "unauthorized" licensed code (such as copyrighted code) from the datasets used to train its AI models.
Codeium is making AI-powered programming platform that maximizes developer productivity. The creators say that, unlike competing AI coding systems that provide generic code snippets and require significant effort to integrate and secure into existing codebases, Codeium's model makes recommendations within the context of an application's entire codebase and can be installed on company's own server. It supports approximately 70 programming languages and integrates with a number of popular development environments, including Microsoft Visual Studio and JetBrains. Codeium also offers a generous free plan and this strategy works fine: the startup has over 1000 enterprise clients, including Anduril, Zillow and Dell. Codeium's model is trained on publicly available code samples, so the company takes steps to remove "unauthorized" licensed code (such as copyrighted code) from the datasets used to train its AI models.
11
Country: USA | Funding: $232M
Imbue is building Sculptor - a UI for Claude Code, that coordinates coding agents (and as advertised makes AI coding reliable). It allows to spin up multiple agents in parallel and orchestrate their cooperation: for example one agent can kick off a refactor while another agent is building out a feature, or check a hunch while you’re debugging, all without breaking your flow. Every agent runs in its own container, so they can all execute code safely, and then Sculptor brings an agent’s work from its container into your local repo, keeping your git state synced so you can collaborate directly from IDE. You can test agent’s work in your dev environment and commit changes you like. Sculptor saves every agent session with its plans, chats, tool calls and code changes, so context is never lost, and progress is always preserved.
Imbue is building Sculptor - a UI for Claude Code, that coordinates coding agents (and as advertised makes AI coding reliable). It allows to spin up multiple agents in parallel and orchestrate their cooperation: for example one agent can kick off a refactor while another agent is building out a feature, or check a hunch while you’re debugging, all without breaking your flow. Every agent runs in its own container, so they can all execute code safely, and then Sculptor brings an agent’s work from its container into your local repo, keeping your git state synced so you can collaborate directly from IDE. You can test agent’s work in your dev environment and commit changes you like. Sculptor saves every agent session with its plans, chats, tool calls and code changes, so context is never lost, and progress is always preserved.
12
Country: USA | Funding: $222M
Replit is a browser-based integrated development environment for cross-platform collaborative coding.
Replit is a browser-based integrated development environment for cross-platform collaborative coding.
13
Country: USA | Funding: $196M
Cognition is an applied artificial intelligence lab that focuses on reasoning and code.
Cognition is an applied artificial intelligence lab that focuses on reasoning and code.
14
Country: Israel | Funding: $115M
Lightrun provides real-time application insights and dynamic instrumentation for developers.
Lightrun provides real-time application insights and dynamic instrumentation for developers.
15
Country: USA | Funding: $87.6M
CodeRabbit is an artificial intelligence company that offers detailed, line-by-line code suggestions and reviews within lines of code.
CodeRabbit is an artificial intelligence company that offers detailed, line-by-line code suggestions and reviews within lines of code.
17
Country: Israel | Funding: $57.1M
Tabnine is an AI coding assistant that accelerates and simplifies the software development process without sacrificing privacy and security.
Tabnine is an AI coding assistant that accelerates and simplifies the software development process without sacrificing privacy and security.
18
Country: Israel | Funding: $57.1M
TabNine is a startup that provides coders with an all-language autocomplete software. Acquired by Codota
TabNine is a startup that provides coders with an all-language autocomplete software. Acquired by Codota
19
Country: Israel | Funding: $50.6M
Qodo enables quality-first AI code generation and generating meaningful tests for busy automatically using generative AI.
Qodo enables quality-first AI code generation and generating meaningful tests for busy automatically using generative AI.
20
Country: USA | Funding: $45M
Tonic mimics your production data to create safe, useful, de-identified data for QA, testing, and development.
Tonic mimics your production data to create safe, useful, de-identified data for QA, testing, and development.
