Table of Contents
Key takeaways
- Most companies aren't short on AI intelligence anymore - they're short on a foundation that keeps it current, shared, and trusted.
- CSV exports, screenshots, and one-person chat threads produce impressive demos and no lasting value.
- Enterprise AI needs one live model that finance, sales, HR, and supply chain all calculate against.
- Pigment's modeler agents already generate 70% of all new metrics and formulas across customers.
- Frames lets every user get a purpose-built view of the same live plan, instead of one static screen for everyone.
- Autonomous planning means the system notices change, explains why, and proposes the decision - not just refreshes a dashboard.
An intelligence surplus, a trust deficit
At Pigment Catalyst London, held at BAFTA, field CTO Thierry d'Hers opened the keynote with a blunt diagnosis: AI no longer has an intelligence problem. Frontier models get smarter every month. Agents complete genuinely complex tasks unsupervised. But very little of that intelligence is helping companies run their business day to day.
The reason, he argued, is architecture, not ability. Most AI work today runs on:
- CSV exports that go stale the moment the business changes
- Fragmented spreadsheets with no shared source of truth
- Screenshots pasted into a prompt, one person, one chat, one moment in time
The result: a sharp answer today, and a team starting from scratch again next month.
Why access to data isn't the same as one version of the truth
Model Context Protocol (MCP) has made it easier for AI to read directly from a CRM or ERP instead of relying on manual exports. That's real progress - but it solves only half the problem.
Connect AI to ten separate systems and you don't get one answer. You get ten different versions of the truth, because each system carries its own definitions and logic. Sales' "pipeline" isn't finance's "bookings." Supply chain's inventory view won't reconcile cleanly with finance's working capital view.
The fix isn't more connections - it's one shared model where finance, sales, HR, and supply chain calculate against the same definitions. Whether an organization runs ten agents or several hundred, they all need to follow the same rules and see the same numbers.
From configuration to intent
For years, planning started with configuration: write a spec, learn the tool, spend days wiring formulas before anyone sees a number that matters.
That sequence is inverting. Pigment's modeler agent, launched in March, lets teams start from intent instead - describe what you want, or hand it a messy artifact (Excel model, PowerPoint, requirements doc, even a photo of a whiteboard) and it builds the model, formulas, and dashboards directly inside the governed platform, using context from what already exists.
The scale of this shift, in numbers:
- Modeler agents now create 70% of all new metrics and formulas across Pigment's customer base - not a lab result, the majority of new modeling work in production.
- Unilever built and launched a North America overhead application in 25 days, versus an estimated four to six months without the agent - the fastest turnaround they'd ever achieved.
The output isn't a prototype waiting to be rebuilt. It's the live application, from day one.
Freedom vs. trust: the Frames story
Standalone generative AI tools are excellent at producing something that looks right, fast. The problem: that mockup isn't wired to live data, doesn't inherit access permissions, and nobody can trace how a number was calculated.
Pigment Frames closes that gap by making the interface itself generative - describe the experience, and an agent builds it live on top of the governed model, inheriting the same calculations, access rights, and audit trail as everything else.
Gavin Allen's live Catalyst demo showed this end to end: a 15-tab Excel long-range plan and a CFO's Notion meeting notes became, in a single afternoon:
- A board-ready simulation tool with sliders and scenario views
- A guided, localized input form for regional leaders
- A Monday-morning executive report, auto-generated
The same task previously took roughly five weeks.
What customers say about it
- Supercell had already built custom AI interfaces before Frames, but still felt the risk of drifting from their live business model. Their reaction to Frames: "our only limit is our imagination."
- The Economist uses Frames for live executive dashboards directly on their Pigment model, eliminating the Google Sheets export step that used to break the live-data connection.
- Unilever summed it up: Frames combines the speed of generative AI with the control of a governed data foundation - moving from creative prototype to production application without losing creative freedom.
The endpoint: a system that plans without being asked
Unify and build get you to AI that constructs what it's told to construct. Automate is where the work itself changes shape.
Pigment's analyst agent runs in two modes:
- Live conversation - you push, refine, ask it to go deeper
- Autonomous mission - you hand it a cadence and it runs on its own, daily, weekly, or the moment a submission lands
In Gavin's demo, weekend budget submissions were already flagged, ranked, and summarized by Monday morning - without anyone asking. That's the shift Pigment calls autonomous planning: a system that watches the business, works out why something changed, and brings a decision forward, not just a data point.
Today that's analyst agents monitoring the business around the clock. Next: planner agents that run scenarios, compare paths, and recommend which one to take.
What this means for planning teams now
The teams getting real value from AI in planning aren't running the flashiest individual prompts. They solved the foundation problem first - one live model, shared definitions, governed access, full audit trail - then let generative AI build on top of it.
If your AI experiments keep dying the moment the business changes, that's not an intelligence gap. It's a foundation gap.
