SpaceOut.

September 2026 edition · first published September 2026

AI coding tools: assistants, agents and the platforms behind them

If you run an engineering team, you have been asked which AI coding tool to standardise on, and the answer keeps moving. What began as autocomplete in the editor is now a spread of editors, autonomous agents, review bots and platforms, and several of the best known changed owner between July 2025 and August 2026.

We checked 21 tools against their own websites, pricing, enterprise and security pages on 25 September 2026. This is what we found, grouped by what each one does.

The short version

  • The category splits into three groups: editors and assistants, agents that work on their own, and tools that review what both produce.
  • The owners changed. Between July 2025 and August 2026, Cursor was acquired by SpaceX, Windsurf by Cognition and Tabnine by Tricentis. Graphite agreed to be acquired by Cursor, and Amp announced it is becoming a company separate from Sourcegraph; neither has announced completion.
  • Deployment options differ more than features do. Some vendors offer self-hosting or air-gapped installs, some say plainly that they do not, and two give different answers on different pages. That is the question your security review will ask first.

Editors and assistants

Cursor now describes itself as "your coding agent for building ambitious software". In April 2026 SpaceX said it had obtained the right to acquire Cursor, and in June it announced that it would, as CNBC reported. On 14 August 2026 Cursor said it had been acquired by SpaceX.

GitHub Copilot describes itself as "your AI accelerator for every workflow, from the editor to the enterprise", working across the editor, the terminal and GitHub itself. JetBrains sells AI inside its IDEs and an agent called Junie, and offers cloud, on-premises and isolated deployment. Gemini Code Assist is Google's coding assistant for IDEs and the command line. Since 18 June 2026 Google has stopped serving individual users and Google AI Pro and Ultra subscribers through it, and points them to its Antigravity products; Standard and Enterprise subscriptions are unaffected. Zed calls itself "your last next editor", a multiplayer code editor from the creators of Atom.

Amazon is moving Amazon Q Developer customers to Kiro, its spec-driven agentic IDE: AWS announced that Q Developer sign-ups stopped in May 2026 and support ends in April 2027.

Tabnine was acquired by Tricentis in July 2026, and tabnine.com now redirects to Tricentis.

Agents that work on their own

Devin, from Cognition, calls itself "your team's autonomous software engineer" and runs agents in parallel in the cloud. CNBC reported that talks for OpenAI to buy Windsurf fell apart, after which Google hired Windsurf's chief executive and some of its research team under a licensing deal. Days later Cognition signed a definitive agreement to acquire Windsurf, and in June 2026 relaunched it as Devin Desktop.

Claude Code, from Anthropic, is an agentic coding tool that reads a codebase, edits files and runs commands. OpenAI Codex is a coding agent for the terminal, the IDE and the web, included in ChatGPT plans. Factory sells Droids that automate coding, testing and deployment for startups and enterprises. Augment Code sells agentic development "at organizational scale" through Cosmos, which it describes as a platform that runs software agents at scale. Amp calls itself a coding agent "built for the frontier". In December 2025 Sourcegraph and Amp announced that they are becoming two separate companies. Cline is an open-source coding agent that lets you choose the model. Warp sells a terminal and an agent platform across the software lifecycle.

Disclosure: SpaceOut's own agents run on Claude, which is one of the tools described here. We have no commercial relationship with any company in this study.

Review and quality

Agents write more code, and more code needs review. CodeRabbit calls itself an "AI-first pull request reviewer", with line-by-line suggestions and chat. Qodo, which renamed itself from CodiumAI in September 2024, sells "Code review for agents. Governance for humans." Graphite sells stacked pull requests and AI review. In December 2025 Cursor announced an agreement to acquire it. Graphite's own post said the deal was subject to customary closing conditions and expected to close within weeks; Cursor's said neither, and no close has been announced. Graphite now offers Cursor Cloud Agents inside its product. Greptile calls itself "The AI Code Reviewer" and reviews with context from the whole codebase.

Deployment and security, in the vendors' own words

  • Self-hosted, on-premises or air-gapped options stated: JetBrains, Cline, Qodo, Greptile, Factory and CodeRabbit.
  • Deployment inside your own cloud or VPC stated: Warp, and Claude Code through Amazon Bedrock, Google Vertex or Microsoft Foundry. Kiro states that it is HIPAA eligible and available on AWS GovCloud (US).
  • Different answers on different pages: Cursor's enterprise FAQ says it does not offer on-premises deployment today and runs on AWS. Its deployment documentation describes running Cursor agents as a command-line tool on your own infrastructure, including on-premises, and it now lets customers host the machines agents execute on. Ask Cursor which applies to your setup. Devin's enterprise page says you can deploy it in your own virtual private cloud, while its deployment documentation describes Cognition hosting Devin in a single-tenant VPC that connects to yours over AWS PrivateLink or an IPSec tunnel.
  • Stated limits: Amp says it does not offer self-hosted deployments, but can use your own API keys for certain model providers. Zed's business page says SSO, SAML, SCIM and SOC 2 are not currently available and that it is working on all four. GitHub's documentation states that Copilot is not currently available for GitHub Enterprise Server.

For two of the 21 tools, Cursor and Devin, the answer depends on which page you read. Before your security review, ask every vendor to confirm in writing which deployment applies to your setup.

What to ask before you standardise

  1. Where does the code go, and where does the model run? Get it in writing, as above.
  2. Who owns this tool now, and what changed when they did? Three tools here changed owner between July 2025 and August 2026, and two more have deals announced but not completed.
  3. How do you pay when agents do the work? Seats measure people. Agents consume usage. Ask how the bill behaves when an agent runs for an hour.
  4. Who reviews what the agent writes? Budget for a person, or evaluate the review tool too.

If you sell one of these tools, expect these questions in the first reply.

How we did this

We checked 21 tools against their own websites, pricing, enterprise, security and sales pages on 25 September 2026. The 21 are Cursor, GitHub Copilot, JetBrains AI and Junie (counted as one), Gemini Code Assist, Zed, Kiro, Amazon Q Developer, Tabnine, Devin, Devin Desktop (formerly Windsurf), Claude Code, OpenAI Codex, Factory, Augment Code, Amp, Cline, Warp, CodeRabbit, Qodo, Graphite and Greptile. Ownership changes come from the companies' own announcements, and from CNBC for the Windsurf talks with OpenAI and the steps of SpaceX's purchase of Cursor. The list covers the tools we examined, not the whole market.

If something here is out of date or wrong about your company, tell us and we will correct it with the source.

Tell us which market you are trying to win.

Prefer email? hello@spaceout.agency