Laiki Changelog: agents better prepared to work in Laiki

Laiki's release of the week, November 2, 2026: we released improvements for those who use AI agents to build dashboards, query results and create automations in Laiki. The focus was on giving the agent more context to carry out requests, identify problems during construction and help resolve failures without handing all the work back to you.

Ask for a dashboard and get better organization

The agent can now query and adjust the layout of dashboards considering desktop, tablet, phone and TV.

This allows requests such as placing two indicators side by side, enlarging a table or rearranging a dashboard for a presentation. Laiki takes care of the placement of the blocks and adapts the composition to the different screens.

We also fixed situations where adding a block split a comparison or removing another left the phone layout outdated. For those who build dashboards through conversation, the result is less tidying up after opening the dashboard.

More context for the agent to query your data

Agents now receive more complete references on metrics, dimensions and combinations accepted by Google Ads, Meta Ads, LinkedIn Ads and Google Analytics 4.

When something goes wrong in a query, the response guides the agent:

• if the query is not valid, it identifies the field or the combination that needs adjusting;
• if the problem is a pause or unavailability of the source, the agent is told that it can try again and how long it should wait.

Custom metrics and dimensions from Google Analytics 4 also become part of the analysis options, allowing you to work with the specific data your team already tracks.

Automations that are easier to build and review

When creating or changing an automation, the agent receives warnings about problems that previously only showed up at run time: an unrecognized calculation or a piece of information that stopped existing after a change in how the flow is started, for example.

We also expanded the guidance to:

• review event values;
• avoid duplicate records;
• recognize excessively large flows.

These warnings point out what deserves attention, without preventing the save.

The visual layout follows the agent's successive edits, keeping the flow paths readable. If someone adjusts the positions manually, that arrangement is preserved.

From diagnosis to a new run, through conversation

Before, the agent could investigate a failed run, but repeating it had to be done through the interface. Now, it can run a case or a set of runs again, present the necessary warnings and follow the progress.

This makes a difference when an integration is temporarily unavailable and several runs need to be resumed: it is no longer necessary to open each one to repeat it manually.

We also added Fix with agent to the warnings of certain connections that do not appear in the editor. The button prepares a request with the context needed for the fix. For administrators with the necessary permissions, the change stays in the working copy, ready for review before publishing.

Connecting an agent became more accessible

All members now find the connection instructions in the user menu and in Connected apps, in the profile, with guidance for ChatGPT and Claude.

The updates are already available in Laiki. Each agent remains limited to the actions authorized by the organization. To see the previous release, check the October 26 changelog.