Airclerk helps agronomy firms put AI to work on what follows the crop walk: the visit’s notes and photos turned into a draft inspection report, soil results read against a calibration that fits the method, spray recommendations written from the label in hand, and the action list carried from one visit to the next. The assessment arrives drafted for the agronomist to correct. The advice and the signature stay with them.
Illustrative example
A spring inspection on a Mid Canterbury arable farm, drafted from the agronomist’s voice notes, quadrat counts, photos and the paddock history. The draft keeps what was counted apart from what was seen, what the grower said and what it probably means, and every diagnosis carries a confidence. Illustrative farm; no real client.
Every line about this paddock is one kind of statement and is labelled as such: a count, a thing seen, a thing the grower said, a reasoned guess with its confidence, a next step that asks for evidence before it asks for fertiliser, and a gap where a threshold would normally sit, with the question behind it handed to the agronomist. The agronomist opens the draft and corrects the judgement rather than typing up the visit.
The division of labour is written into the report. The workflow drafts: it assembles the visit, keeps each kind of statement in its place, proposes the assessment and lists what would confirm it. The agronomist decides: the diagnosis, any product and rate, what the grower is told, and the signature stay with the agronomist of record, and the report says so on its face.
Agronomy firms use AI models such as Claude, Microsoft Copilot or ChatGPT for the work that follows a farm visit: crop inspection reports drafted from the visit's notes, counts, photos and paddock history, soil and tissue test interpretation, nutrient plans, written spray recommendations from the label or permit in hand, sowing plans, seed crop progress reports, matching recommendations to what was actually applied, and the end-of-season review. It works from the records the firm already keeps, such as lab reports, email, document storage and the farm management platform; which of those it reaches through a direct connection and which through an export is settled in the design phase. Grower isolation and sourced figures are built into the workflow, and nothing reaches a grower without the agronomist's approval. The AI drafts the assessment. It does not choose the product or sign the advice.
Each is a candidate first workflow, adapted to the firm's crops, regions, systems and templates. Five start from Airclerk's agronomist pack: skills for New Zealand and Australian agronomy built from named, dated sources opened in October 2026, each carrying the date its rules were checked. The pack is an accelerator rather than a product, and every skill is adapted to the firm's own calibrations and report formats before production use.
Voice notes, counts, photos and the paddock history become a draft inspection report. Growth stage is recorded on a named scale, observations stay apart from diagnosis, every diagnosis carries a confidence, and counts are compared only with a threshold from a named source. The biosecurity screen runs before any of it.
Agronomist packChecks the test method and sampling depth before reading each result against a calibration that fits both the method and the crop. A result that cannot be interpreted with what is on file stays flagged for the agronomist, and the numbers pass to the nutrient plan with their method attached.
Agronomist packA fertiliser recommendation or nutrient budget built from the soil result and the application history against the crop's demand, checked against the rules that apply to the paddock as at the date of advice, with the advice record kept. It does not verify a budget for a regulator or certify a farm plan.
Agronomist packDrafted from the label or permit in hand: registration status, off-label class, residue and withholding implications, resistance group, and the region's drift and notification rules. No rate, interval, buffer or withholding period comes from memory. Each is taken from the label, the permit, the register entry or another named source, with its date.
Agronomist packEvery herbicide in the paddock history checked against its label's plant-back interval, then cultivar choice from FAR cultivar performance trials or GRDC National Variety Trials, seed source, sowing window and rate, each tied to a source for the region and season rather than to last year's habit.
Agronomist packInspections of contract seed crops through the season, from establishment to pre-harvest, written up with photos in the format the seed company or buyer expects. Contract conditions on cultivar and permitted chemistry are checked against the file and flagged where the file is silent.
Matches each recommendation to what the grower or contractor recorded as done, in the spray diary or the farm management platform. A recommendation with no completion record stays open, because recommended and applied are different records.
Carries the action register from visit to visit and drafts the reminders to the grower and contractor for the agronomist to send. At harvest it assembles the season paddock by paddock: stages, inputs, decisions, yields and the open questions for next year.
Optional AI Process Assurance can sit behind the workflows we implement: the instructions, sources, exceptions, review changes and sign-off retained as an AI Workpaper, so the firm can show what the AI did, what it did not do, how the result was checked and who approved it. It is a platform subscription, implemented alongside the workflow, and the workflow runs without it.
A free introductory conversation is enough to say whether the workflow you have in mind is a first workflow, and to quote a fixed fee for it. The detailed design happens inside the engagement, with your agronomists, against your own report formats and records.
Where AI could make the biggest difference, how that work runs today, the systems it touches and who owns it. We also work out together whether Airclerk is the right partner for it. If we are, the conversation gives us enough to quote.
One document: the outcome we are aiming for, what we deliver, what your team contributes, the fixed fee and timeline, and what acceptance looks like. Nothing starts until it is agreed on both sides.
The engagement opens with the detailed design, agreed with you before anything is built. Then we configure and connect the workflow to your documents and systems, build in the agreed review controls and audit trail, test it on real work, and prepare the people who will run it.
Handover, and the support that follows go-live, are written into the scope so you know what is covered before you sign. Ongoing operations and further workflows are a separate agreement, made when you want them rather than switched on by default.
Read how an engagement runs in detail →
Airclerk can implement Claude, Microsoft Copilot or ChatGPT depending on the firm's existing environment. Workflows run on the firm's approved enterprise instance and under its own vendor agreement.
The first workflow is chosen by volume, source availability, repeatability and review effort. We would usually suggest the inspection report or the soil test interpretation, where the agronomist already does the thinking and loses the evening to the write-up. Product choice and anything that would go straight to a contractor stay outside the first workflow until it has proved itself.
Each draft carries the name of the agronomist who will check and sign it. The workflow stops at that gate. It does not send advice to a grower, issue a job to a contractor, mark a spray as done or close an action by itself. Where AI Process Assurance is enabled, the AI Workpaper records which sources were used, what exceptions were raised, what the agronomist changed and whether the gate was passed.
The workflow gets only what the agronomist had on the day of a past visit: the counts, notes, photos, results and history. What they concluded, and what happened next, is held back. The firm's senior agronomists then score the drafts for material errors, missed risks, interventions that were not needed, and whether it asked for more evidence when it should have. Visits from the same farm and season are kept together, so one record cannot give away the answer to another.
The engagement produces: the configured workflow on the firm's approved AI platform, connected to its records; a Workflow Charter defining its stages, permitted sources and approval gate; a boundary statement describing what the workflow must never decide or do; evaluation results from the past-visit testing; a runbook for agronomists and office staff; training for each role; a retained AI Workpaper for each material run where Assurance is enabled; and the check-ins after go-live that confirm the workflow is being used safely, shaped to the firm and written into the scope. Ongoing operations beyond that are a separate agreement.
If anything recorded looks unusual for the district or could be a notifiable organism, the workflow stops treatment advice and points the agronomist and grower to the reporting channel first, which in New Zealand is MPI. It does not hold the question back while a lab result is pending, because confirming a suspected exotic is MPI's job, not the agronomist's.
Rates, critical values, thresholds, withholding periods and residue limits come from the lab report, the label, the permit, the register or a named source with its date. Where one is missing, the draft names the gap instead of supplying a plausible figure: no rate is written without the label in hand, and a count with no sourced threshold is marked “no verified threshold”. A made-up threshold either triggers a spray on nothing or holds one back when it was needed.
Photos support the draft, and the agronomist checks anything described from them. Image-capable models are not validated tools for telling one leaf disease from another. Photos are filed as evidence against the observation they support. A review done from photos alone is labelled as a remote review, not a field inspection, and its confidences are lowered to match.
A finished report tells you what an agronomist concluded. It rarely says why, what was ruled out, whether the grower acted on it or how the crop responded. The firm's most valuable material is the complete case: what was seen, what was advised, what was applied and what happened next. The workflow can be built to retrieve comparable past cases and the firm's approved guidance at each assessment, and to keep that record growing with every visit it drafts. Which earlier visits it may draw on, and how other growers' records stay out of a grower's report, are settled in the design phase and tested before go-live.
Agronomy firms can use AI for the work that follows a farm visit: turning field notes, counts, photos and paddock history into a draft inspection report, reading soil and tissue results against a calibration that fits the test method, drafting nutrient plans and written spray recommendations from the label or permit in hand, building sowing plans, reporting seed crop progress, matching recommendations to what was actually applied, and assembling the end-of-season review. Airclerk implements these as governed workflows connected to the records and approved references the firm already keeps. The agronomist of record reviews every draft and signs the advice.
It can draft the interpretation for the agronomist to check. Given the visit's counts, observations, photos, soil results and paddock history, the workflow proposes a likely cause with a stated confidence, and the evidence that would confirm or rule it out. Sometimes the right next step is to collect more evidence rather than to treat, and the workflow is built to say so. The agronomist then corrects or approves the assessment instead of writing it from scratch. Testing on the firm's past visits shows whether the drafts are good enough to use. The agronomist still reviews and approves every assessment and recommendation.
No. If recommendations and spray records already live in a platform such as Agworld, Airclerk does not try to replace it. The workflow reads from the records the firm already keeps and works on what is still done by hand: the narrative assessment, the report the grower or seed company receives, the follow-up, and the consolidation of lab results, photos and notes that sit in different places. Whether that is a direct connection or a regular export depends on what the platform offers and what the firm's licence allows, and it is settled in the design phase.
It can draft a written recommendation from the label or permit in hand, checking registration status, off-label class, residue and withholding implications, resistance group and the region's drift and notification rules. No rate, interval, withholding period or residue limit is written from memory; each is taken from the label, the permit, the register entry or another named source with its date, and where none is available the draft says so. The agronomist of record chooses the product and signs the recommendation. We would usually start with the inspection report or soil test interpretation, and add written recommendations once the first workflow has proved itself.
The biosecurity screen runs before anything else. If an observation looks unusual for the district or could be a notifiable organism, the workflow stops treatment advice and directs the agronomist and grower to report it (in New Zealand, to MPI). Finding a notifiable organism in New Zealand creates a legal duty under the Biosecurity Act 1993 to tell MPI. The workflow does not wait for a lab result before raising it, because confirming a suspected exotic or notifiable organism is MPI's job.
Two controls come first. Grower isolation: one grower's paddock records and reports are walled off from another grower's work. Permission-aware access: the workflow sees only the records the person running it is already authorised to see in the firm's existing systems. Before any grower information is connected, Airclerk reviews the AI vendor's enterprise terms, retention, training use, subprocessors and data-location arrangements. The implementation runs on the firm's approved enterprise instance, under its own agreement with the AI provider.
In plain terms: the workflow keeps a record of its sources, the checks it ran and who approved the result. Airclerk calls this AI Process Assurance. A Workflow Charter defines the expected process, approved sources, professional boundaries and review gates. A Semantic Audit records what actually happened during the run. The retained AI Workpaper shows the sources used, exceptions raised, review changes and final approval. It does not claim to capture everything an agronomist does. It creates a reviewable record for the specific AI-assisted workflows the firm chooses to govern, and it is optional: the workflow runs without it.
Firms whose agronomists walk crops and pasture and advise growers on what to do next: independent field agronomy firms, grain and seed companies with field teams, rural suppliers with technical advisers, soil and nutrient consultancies, and horticultural technical services. The first workflow is chosen around the firm's actual work rather than its label. A seed company reporting crop progress to buyers and an independent agronomist writing visit reports for growers can share the same starting point.
Start with the question we would ask first: after a farm visit, what do you still have to type up or re-enter, and where does it go? An introductory conversation is enough to say whether that is a first workflow and what it would cost to build.