The tech stack of the future for strategy and finance teams.

Storage. Production data sitting in a governed warehouse. This layer is largely solved and has been for a while. The open question is no longer where the data lives, it is what is allowed to reach it.

Context. A semantic layer between the warehouse and everything above it, holding defined metrics, approved dimensions and row-level security. It puts business language on top of raw tables, which is what turns a schema into something a person or an agent can reason about. Agents need this more than people do. Without it they cannot query safely, and they will not give you the same answer twice.

LLM. This is where the daily workflow is moving. You describe the question, the model pulls the data, interprets it, drafts the narrative and often renders a first pass at the chart. Human judgment still decides what any of it means. What changes is the starting point, which is now a conversation rather than a blank grid.

Charting. The layer that turns model output into live, editable charts that travel across surfaces. It needs real power-user control, data that stays attached to the chart, and enough reliability to carry recurring reporting rather than one-off analysis. Chartbuddy is what we are building here.

Reporting. Distribution stays fragmented, and I do not expect that to change. Board decks, dashboards for monitoring, written memos, and agent-generated briefs that someone reviews before sending. The stack has to support several endpoints coming out of one analysis rather than assuming a single deliverable at the end.

Nothing in that stack replaces spreadsheets or decks. What it removes is the glue work between them. Rebuilding the same monthly pack from scratch because last month's version is not connected to anything. Screenshotting a dashboard tile and reformatting it to fit a slide. Waiting days for a pull an agent could run in minutes. Pasting chart images with no data behind them, so that the next person who needs to change something starts over.

What survives is the part that was always the job. Judgment about what the numbers mean, a narrative that holds together, and knowing which number actually matters this quarter.

The loop is easier to see with a real example. The data sits in the warehouse and the metrics are defined once in the semantic layer. An agent pulls the revenue bridge and flags the anomalies worth a second look. Someone finalises the charts and they land in the board pack. A trimmed version of the same analysis feeds a dashboard that gets watched monthly. Same numbers, three surfaces, and no rebuilding in between. That loop is the workflow now, not the exception.

Timing and winners, mostly. I expect semantic layers to consolidate, though I would not want to guess who ends up owning them. Some teams will skip dashboards entirely for ad-hoc work once agents are good enough to answer the question directly. Memos might eat more of the board pack than decks do, which would be an uncomfortable outcome for someone who sells charting software.

What seems harder to argue with is that agents need a governed context layer above the warehouse before any of the rest works. And if your deliverable is still a chart-heavy deck, a static image pulled out of a chat is not a long-term answer.

The question worth asking is not which AI tool to buy. It is where insight still dies between the warehouse and the stakeholder. For most teams I talk to, that answer has not moved in years. The data is fine and the analysis is fine. The last mile of the report is where the time goes.

If you are mapping the same transition today, this piece on fragmented FP&A workflows is a useful companion read.

• The FP&A Reporting Workflow Is Fragmenting
• How to make charts in Claude (and where it still falls short)
• Stop Rebuilding Your Monthly Report From Scratch