AI Coding Tools: What Changed, and What I’d Test

Windsurf, Cognition and Cursor

On 14 July 2025, Cognition, the company behind Devin, announced a definitive agreement to acquire Windsurf. On 2 June 2026, it introduced Devin Desktop as Windsurf’s next generation, combining the IDE with a workspace for local and cloud agents.

Cursor announced on 14 August 2026 that SpaceX had completed its acquisition. It describes the April model-training partnership as the beginning of that process.

An ownership change is a reason to check the contract, available models and options for moving to another tool.

Sources: Cognition: Windsurf acquisition agreement, 14 July 2025; Cognition: Introducing Devin Desktop, 2 June 2026; Cursor: completed SpaceX acquisition, 14 August 2026.

Investment and delivery costs

On 8 September 2026, Cognition reported a funding round of more than $2 billion at a $48 billion valuation. Those company-reported figures describe one financing round; they do not measure the entire coding-tool market.

For a team, the relevant cost includes the subscription or usage bill and the time spent reviewing and correcting the work. I would measure that in the repository where the tool will be used.

Sources: Cognition: Series E announcement, 8 September 2026.

Codex is the tool; Astra is a model

OpenAI’s public model guide lists GPT-6 Astra as gpt-6-astra and documents selecting it with ‘codex -m gpt-6-astra’. Access depends on the account, client and rollout. OpenAI describes it as a model for complex work involving several steps and tools.

Codex CLI can inspect repository files, edit code and run local commands. You can choose the model, reasoning effort and permissions, then review the commands, patch and test output.

OpenAI’s Astra API guide also documents asynchronous tool calls and instructions supplied while a task is running. The application executes the tools and manages their results. The controls available in a particular Codex interface may differ from those API capabilities.

Sources: OpenAI: models available in ChatGPT and Codex; OpenAI: Codex CLI; OpenAI: GPT-6 Astra API guide.

What Claude Code adds around Claude

Anthropic describes Claude Code as a loop that gathers context, takes action and checks results. Claude models do the reasoning. Claude Code provides file operations, search, command execution and context management, feeding command output back into the next step.

The documentation covers project instructions in CLAUDE.md, configurable permissions and file checkpoints. Checkpoints cover file changes; they cannot roll back effects on a remote database, API or deployment. I would include those limits in a team trial, alongside patch quality.

Sources: Anthropic: how Claude Code works.

Choose the working environment as well as the model

Cursor documents an agent with instructions, a selected model and tools for searching, editing and running commands. Devin Desktop combines an IDE with management of local and cloud agents, pull requests and their context.

I would ask an engineer to try an unfamiliar change. Can they follow the edits, stop the agent, run the checks themselves and continue manually? That tests how much control the editor gives them during the work.

Sources: Cursor: Agent overview; Cognition: Introducing Devin Desktop, 2 June 2026.

The shortlist can include tools you already have

GitHub documents Copilot features spanning editor assistance, chat, code review and an agent that can work on changes and raise pull requests. I would include it in an evaluation where the team already uses GitHub, while checking which capabilities its plan and organization settings allow.

Gemini CLI provides another terminal-based option, with documentation for file tools, shell commands, project context and permissions. A team that prefers command-line workflows can assess it alongside Codex CLI and Claude Code. The useful comparison is the complete setup: tool version, model, permissions, context and price for the work performed.

Sources: GitHub: About GitHub Copilot; Google: Gemini CLI documentation.

I would start with three jobs from the backlog

I would give each candidate the same starting commit and acceptance criteria, using a bug, a small feature and a maintenance change from the backlog.

For the bug, use a search screen where a slow response replaces newer results. Ask for a repair and a regression check. The reviewer should be able to reproduce the ordering problem and explain why the fix prevents it.

For the feature, add a filter that survives refresh and browser Back. Specify what the URL should contain and what happens when the result is empty. Check those interactions in a browser, including keyboard use.

For maintenance, update a dependency with a known breaking change. Supply the official migration notes and identify which code may change. Check the affected call sites, lockfile changes and anything the agent could not verify.

Count review and repair time

Record the time from the initial brief to an accepted change, including setup, clarification, review and repairs. Record usage costs, failed attempts and work left for another person.

Repeat the tasks from clean copies. Keep the tool version, model and settings with each result so a later release can be compared under the same conditions. A small trial can inform this team’s choice without establishing a ranking for everyone.

• Correctness: did the change meet the agreed behavior, including failure cases?
• Review: what did a person have to discover, explain or repair?
• Scope: did it change unrelated files or ask for unnecessary access?
• Continuity: can another engineer run the checks and continue the work?

What I would change as a team lead

An agent can read a repository, change files and run checks. The brief therefore needs to specify the intended behavior, which parts of the system it may change and how the result will be checked. The engineer submitting the change still needs to understand it.

I would choose one tool for routine work and keep repository instructions, test commands and decisions accessible outside its chat history. Add another tool when a trial shows where it helps. Revisit the choice when a model, price, policy or ownership change affects the team.