changelog

Tidewave in CI, file search and review tabs

Hugo Baraúna · October 29, 2026

This is a roundup of the features, enhancements, and bug fixes we shipped in October, plus the videos we published.

Product updates

Tidewave in CI

Tidewave gives your CI agent access to the running application and a real browser. This lets it reproduce reported bugs, run Accessibility and Web Framework diagnostics, and verify its changes with videos and screenshots before responding to an issue or pull request.

Read the Tidewave in CI docs.

Open and search source files

From the Inspector or Code Review, you can now open source files in Tidewave IDE at the relevant line.

In Settings → Editor, enable Open up files within Tidewave first, then Cmd/Ctrl+Click an inspected element or a line in a diff.

You can also use Cmd/Ctrl+F to search within source files and diffs, with highlighted results you can step through.

Code Review: Reviewed and Unreviewed tabs

Code Review now has separate Unreviewed and Reviewed tabs, so you can switch between what still needs your attention and what you've already checked.

Read the Code Review docs.

Other improvements

  • Design canvas: new canvases open automatically, and you can now zoom out to 10% to see more designs at once
  • Session cost: the context indicator now shows session cost when your agent reports it
  • Google sign-in: you can now sign in with your Google account

New videos

We also published four videos in October:

What coding agents need from a programming language

What should a programming language optimize for when coding agents write most of the code? José Valim's post "Evolving programming languages in the AI era" argues that syntax isn't the priority and proposes three directions instead.

I Deleted Phoenix's AGENTS.md to See What Breaks

Do coding agents still need the AGENTS.md that mix phx.new generates? We deleted it from a fresh Phoenix 1.8 app and gave five models a feature request built to lure them into the mistakes the file warns about.

Give your AI agents a definition of done

A coding agent can follow a Phoenix project's instructions and still finish without running the tests. We replace a check that already fails with a mix precommit command that starts green and covers the full test suite. Then we verify a small LiveView change with a focused test, the complete check, and Tidewave browser automation on the running app.

Elixir 1.20 Found a Bug in My Phoenix App

Elixir 1.20 found an invalid public function path in a Phoenix app that compiled cleanly under 1.17. In this video, we upgrade TechSchool with a coding agent, trace the compiler warning to an empty-map default, and remove it from two functions. We then check compilation, tests, and the running application because type checking alone cannot verify the whole migration.