AI agents can build Power BI reports. Now they can set up a custom visual too

Now ask it to add a custom visual from AppSource. It guesses.

So we taught it. We just published a free, MIT-licensed Agent Skill for Promo Gantt Chart. Install it next to Microsoft's skills, and your agent sets up our visual end to end: data, fields, settings, done.

github.com/vibecode-studio/promo-gantt-powerbi-setup-skill

The blind spot in AI-built reports

Microsoft's skills know the built-in visuals: their data roles, their formatting options, their file format. About a custom visual the agent knows nothing. It doesn't know which fields go where, what the settings are called, or how the data should be shaped. The validator accepts the visual, but can only warn that it doesn't recognise it.

So the agent invents field names, sets options that don't exist, and you fix the report by hand. The next time, many teams simply pick a visual the agent already knows.

That isn't a judgement on the visual. The agent was never taught the others. If AI agents become the way reports get built, a visual the agent doesn't know might as well not exist.

What the skill teaches your agent

• Data model. How to shape the data: one table with a RecordType column for tasks, campaigns, events and milestones, with DAX and Power Query recipes.
• Field wells. Every field well by its exact name, and which ones are required.
• Report file. A ready PBIR visual.json the agent can drop into a report.
• Settings. Every Format pane option by its real property name, with allowed values and defaults.
• Microsoft Project import. A Power Query script that turns a Project XML file into one table with phases, sub-phases and a ready Predecessors column.
• Licensing. What's in Free and what's in Pro.
• Troubleshooting. Common setup mistakes and how to fix them.

One prompt, one working plan

With the skill installed, a prompt like this works end to end:

The agent shapes the data, binds the fields, writes the visual and sets every option by its real name. You open Power BI Desktop and the plan is there. No hand-fixing field wells, no hunting for settings.

Built to sit next to Microsoft's skills

This isn't a replacement for anything. Microsoft's skills keep doing what they do well: the model, the PBIR files, the report design. Ours adds exactly one visual to the agent's vocabulary.

It installs in Claude Code, GitHub Copilot CLI or APM with one command (see the README), or you copy the folder by hand.

What the skill is, and what it isn't

The skill is usage documentation: the same information as our support page, written so an AI agent can follow it.

• No source code. It contains nothing that unlocks Pro features.
• No internals. It doesn't explain how the visual calculates anything. Critical path, overlap detection and rendering stay inside the visual.
• MIT-licensed. Copy it, adapt it for your team, or use it as a template for your own visuals.

Why now: Power BI got a native Gantt

The October update brought a native Gantt chart (preview), with task bars, progress, milestones, extra columns and grouping. Good news. If you need tasks, dates and progress, try it first. It's free, built in, and it will improve every month.

Project and planning teams usually need more. Early testers report no dependencies, no critical path or float, and no baseline or today line yet. It's a preview, so check the current limits yourself. Those gaps are exactly the questions a PMO asks: what drives the end date, which tasks have no slack, what happens if this one slips?

That's where Promo Gantt Chart fits:

• Project and PMO reporting. Dependencies read straight from a Microsoft Project export, in all four link types with lag and lead. A critical path calculated in the visual and checked task for task against Microsoft Project. Near-critical tasks, slack bars and a one-row-per-task plan view.
• FMCG promotion planning. Fiscal calendars (4-4-5, 4-5-4, 13 periods), campaign tracks, event weeks, last-year comparison and overlap markers.
• Large plans. 10,000+ rows that still scroll smoothly.

The native Gantt and a specialised visual answer different questions. What changed this month: the specialised one now comes with an agent that knows how to set it up.

To other visual makers

Write a skill for your visual. Fork ours as a starting point if that helps. The agents are already building reports; the only question is which visuals they know how to use.

Curious how others see this. Will AI agents change which visuals end up in reports? And if you build visuals: are you writing skills for them yet?