Cam IncollFieldwalker can now scan your entire map and identify trees, buildings, water features, and dense vegetation in a single pass — presenting everything as suggestions you review before anything touches your design.
When we first built AI detection for Fieldwalker, it worked one feature at a time. You'd click a tree, the AI would identify it, you'd accept or reject. It worked well for catching individual features, and it's still there as Point Detection for exactly that purpose. But when you're starting a new site design and there are fifty trees, three buildings, a dam, and a patch of dense bush to map — clicking each one individually is a lot of clicking.
Auto-Detect changes that.
One scan, the whole map
Open the Design tab, click Detect features, and the AI scans the entire image. Within about a minute, it comes back with everything it found — trees, buildings, water, dense vegetation — grouped by category, each with a confidence score.
Some of those detections are confident enough to be placed on the map automatically. Others are held back as suggestions for you to review. The split depends on a confidence slider you control: slide toward Strict and only the most certain detections get auto-placed. Slide toward I feel lucky and the AI places more on its own.
You can watch the map update in real time as you drag the slider. It's a genuinely satisfying interaction — seeing trees appear and disappear as you dial the confidence up and down.
Review before commit
This was the design decision that mattered most. Auto-Detect doesn't just dump fifty objects onto your map and wish you luck. It presents results as suggestions first.
Some things are always held for review regardless of confidence: dense vegetation areas (because they represent large regions you'll want to verify) and oversized buildings (because a 30-metre detection might be a shed, or might be a misread shadow). Everything else follows the confidence slider.
The review workflow works in both directions. Hover a suggestion on the map and the matching row highlights in the list. Hover a row in the list and the matching object highlights on the map. You can accept or reject from either place — from the list, or from a small floating control that appears when you hover a suggestion on the map.
When you're happy with your selections, Add to design commits everything. The suggestions become real, editable design objects — the same kind you'd get from placing a symbol by hand. Move them, restyle them, delete them. They're yours.
Trees are the interesting case
Individual tree crowns are straightforward — the AI finds a tree, draws the canopy, places a tree symbol. But what about a thick patch of bush? Or a hedgerow? Or a forest edge?
Auto-Detect handles these as dense vegetation areas — a single region with a scalloped organic edge, drawn in the permaculture illustration style. That's the convention from Mollison's design manuals, and it's how these areas should appear on a site plan.
But you also get control over tree density within those areas. Three buttons — Fewer, Auto, More — let you dial how many individual tree symbols get placed inside each vegetation cluster. Sometimes you want a light scatter suggesting the canopy; sometimes you want to show every trunk. The choice is yours, and it updates live as you switch.
Point Detection fills the gaps
Auto-Detect is good, but it's not perfect. Scattered paddock trees — widely spaced individuals in open grassland — can be missed by the full-map scan. Round structures like water tanks sometimes get classified as the wrong category.
That's where Point Detection comes in. Press D, click the feature the scan missed, and it identifies that one spot immediately. Accept it and it's added straight to your design. No batch review, no commit step — one click and it's there.
The two tools are complementary. Auto-Detect for the bulk pass. Point Detection for the exceptions. Together, they get a base map from "raw aerial image" to "all existing features mapped" in minutes rather than hours.
Being honest about what it gets wrong
Detection quality depends on the imagery. Sharper, more recent aerial photos give better results. Older or lower-resolution captures are harder for the AI to read, and you'll see more held-for-review suggestions and fewer confident auto-placements.
Very dense, tangled bush is treated as a single vegetation area rather than individual trees. That's a deliberate design choice, not a limitation — it's how permaculture designers draw these areas.
Roads and paths are excluded entirely. They were unreliable in field testing, so we took them out rather than shipping something that would waste your review time.
This is the philosophy throughout: if the AI isn't confident enough to be useful, it stays out of your way. The confidence slider gives you the control to decide where that line sits for your particular map and your particular tolerance for checking the AI's work.
The base map, faster
The point of all this is the same principle that runs through everything in Fieldwalker: the mechanical parts should be fast so you can spend your time on the design thinking.
Mapping every existing feature on a property is mechanical work. It's necessary, but it's not the creative part. It's not where your permaculture training matters. Auto-Detect handles the bulk of it in under a minute, and Point Detection catches the rest. The design process starts sooner because the base map gets done faster.
That's always been the goal. The AI isn't the product. The design is the product. The AI just gets you there faster.