Groundplan — framework design
We design the blueprints that let your team build its own AI agent capability for the ground you're responsible for.
For the organisations that watch land, water, infrastructure and risk — land trusts, environmental and water agencies, insurers, utilities, disaster response bodies, reporting providers, and the companies that grow and source food.
You don't sell blueprints by pretending nothing has gone wrong
A lot of the people reading this have already sat through an AI pilot that ended quietly. It was demonstrated, it was praised, it ran for two quarters, and then a budget review asked what it had changed and nobody could answer in numbers.
That is not a rare outcome. Analysts tracking this in 2025 and 2026 expect more than four in ten agent projects to be cancelled before the end of 2027 — over cost, unclear value, and rules nobody wrote down before the work started. A separate study of enterprise pilots found the large majority showed no measurable return at all.
We read those numbers the same way you probably do. The failures cluster around the same three things: nobody agreed what the system was for, nobody decided what it was allowed to do without asking, and the whole thing sat inside a vendor relationship that could change direction without warning.
So that is where our work starts, and it is why what we hand over is a plan your own people can read, argue with, and keep.
Sources: Gartner press release, 25 June 2025 (over 40% of agentic AI projects cancelled by end-2027; also its estimate that only a small fraction of self-described agentic vendors offer something genuinely new). MIT Media Lab / Project NANDA, “The GenAI Divide: State of AI in Business”, July 2025 (roughly 95% of pilots showing no measurable P&L return). Both figures are moving and several analysts publish close-but-different versions. verify before publish
Five frameworks
Think architecture firm, not app store
Each one is a way of designing a particular kind of watching-and-deciding capability. They were built once across many engagements and are adapted by each organisation that uses one.
- 01
The Watch Framework
A design for watching a defined territory day after day and deciding what a person should look at today.
Read the design → - 02
The Evidence Framework
A design for assembling the supporting record as the work happens, and being honest about what is missing.
Read the design → - 03
The Response Framework
A design for the first pass of the work — routing, prioritising, drafting the dispatch — with the human decision built in where it counts.
Read the design → - 04
The Underwriting Framework
A design for keeping a live picture of exposure per location instead of one that is accurate the day it is built.
Read the design → - 05
The Field-to-Ledger Framework
A design that connects what is happening on the land to the sourcing, formulation and paperwork decisions made downstream of it.
Read the design → Governance runs through all five rather than sitting beside them: what the system decides alone, what always reaches a person, who answers for a mistake, and how fast it stops.
The governance page →
What we don't do
We don't sell software
There is nothing to log into, no seat count, no renewal. What we deliver is a set of plans and the reasoning behind them.
We don't lock you in
When your priorities shift, or the underlying models change again, you are editing documents you already own. You are not migrating off us.
We don't hand you a black box
Every design states what the system is allowed to decide, what always reaches a person, and how it gets switched off. If we can't explain a piece of it plainly, it doesn't go in.
You own what gets built. We hand you the plans.
If you have more locations to watch than people to watch them, that is the conversation. Bring the sceptical colleague.
Start a conversation