AI maturity assessment framework — a four-dimension model for evaluating agrifood AI deployments

Global (applies to any agrifood AI deployment regardless of region)

Content

This is the field guide’s structured maturity-assessment framework for evaluating any agrifood AI deployment. The framework addresses a recurring problem in the corpus: the activity-status tag in taxonomy/v4.md collapses four distinct dimensions into a single value (deployed | piloting | research | experimental | announced | discontinued), which obscures the substantive differences between a vendor deployment at billion-acre scale and an academic research-stage pilot.

The framework is methodology, not content. It defines four evaluation dimensions and a grading rubric. Each dimension is assessed independently. A deployment can be high on one dimension and low on another — and that’s a real, substantive finding, not a contradiction.

The framework is applied to existing units through the maturity-grade table later in this document. New units should carry a maturity-grade as part of their frontmatter.

The four dimensions

Dimension 1 — Scale

What this measures: the quantified deployment volume of the AI system. Named numbers — units, acres, farmers, growers, operations, monthly active users, transactions.

Grading ladder:

• S0 — no quantified deployment. Vendor / academic / institutional description only; no named figures.
• S1 — pilot-scale deployment. Tens to low hundreds of units / acres / users; typically a single region or single partner.
• S2 — early deployed. Hundreds to low thousands of units / acres / users; multi-region or single-region multi-partner.
• S3 — deployed at scale. Tens of thousands to millions of units / acres / users; multi-region, multi-year.
• S4 — global deployment. Hundreds of thousands to billions of units / acres / users; multi-continent, multi-decade.

Why this matters: vendor-reported scale figures circulate in the corpus without independent verification. A claim of “X million acres” means something different when independently audited vs vendor-reported. The scale dimension alone does not tell you whether the figure is real — pair it with Dimension 2.

Dimension 2 — Verification

What this measures: the epistemic posture of the deployment claim. Who has verified the figure, and how?

Grading ladder:

• V0 — vendor-reported only. The figure is provided by the deploying vendor / institution; no external verification cited.
• V1 — academic / institutional peer-reviewed. A peer-reviewed publication supports the figure (e.g. Plant Phenomics, AI Magazine, Computer and Electronics in Agriculture).
• V2 — third-party case study. A reputable external organisation (Development Gateway, Inrae, Mozilla, IPES-Food, civil-society research) has produced a case study that names the figure.
• V3 — independently audited. The figure has been independently audited by an accounting firm, government inspector general, or comparable oversight body.
• V4 — replicated by independent party. Multiple independent parties have reproduced or verified the figure across different contexts.

Why this matters: per memory hygiene, vendor-reported agritech figures circulate without independent verification. The verification dimension makes this visible — V0 vs V2 is the difference between a marketing claim and a substantively verified deployment. The field guide’s existing gaps G-015 (Agrosmart / Kilimo yield claims verification), G-020 (independent verification of vendor-reported input-reduction figures), G-033 (DJI 222M tons water saved / 30.87M tons CO2 reduced verification) are exactly this dimension.

Dimension 3 — Longevity

What this measures: the operational durability of the deployment. How long has it been deployed, has it survived generational cycles, has the vendor / institution remained stable, has it been discontinued.

Grading ladder:

• L0 — announced / not yet operational. Deployment is announced but not in operational use.
• L1 — first-generation operational. Deployed 0-2 years; first iteration; no generational update cycle yet.
• L2 — multi-generation operational. Deployed 3-7 years; second or third iteration; vendor / institution stable; user base has renewed.
• L3 — mature / durable. Deployed 7-15 years; multi-iteration; survived market cycles; institutional memory.
• L4 — generational / legacy. Deployed 15+ years; multiple vendor or institutional generations; established as operational infrastructure rather than product.

Why this matters: Indiga Ag’s boom-and-bust pattern is the canonical field-guide example: the company raised over $1.6B and then underwent major restructuring. Climate FieldView has been deployed for over a decade and remains operational. Both could be deployed in the v4 schema; longevity tells you they are very different kinds of deployed.

Dimension 4 — Translation

What this measures: the research-to-deployment pathway for academic / research-stage work. Whether documented operational adoption pathways exist for the research outputs.

Grading ladder:

• T0 — research output only. Peer-reviewed papers and grant-funded research exist; no documented operational adoption.
• T1 — pilot deployment documented. At least one operational pilot / deployment is named; figures may be small.
• T2 — multiple operational deployments. Multiple named deployments across partners or regions; translation arms in place.
• T3 — institutional adoption pathway. Cooperative Extension, USDA-NIFA translation program, or comparable institutional structure moves research into operational deployment.
• T4 — self-sustaining translation. Operational deployments continue without ongoing academic / research-side involvement.

Why this matters: the USDA-NIFA AI Institutes (AgAID, AIIRA, AIFARMS, AI-CLIMATE, AI-LEAF) are all high on research output and low on operational deployment figures. The Extension Foundation 2026 National AI Report (units/extension-foundation-2026-national-ai-report.md) names workforce readiness as the limiting factor. Translation is the dimension where the gap between academic output and operational adoption shows up.

How the dimensions combine

The dimensions are independent. A deployment can score:

• High scale + low verification — DJI Agriculture (400,000 drones, 980M acres, 100+ countries — but vendor-reported, not independently audited at scale).
• Low scale + high verification — smallholder agricultural AI pilots in academic literature (peer-reviewed, but tiny deployment).
• High longevity + low current scale — legacy equipment-makers with stable platforms but modest current-year deployment.
• High research + low translation — the canonical academic-agrifood-AI pattern.

When a deployment has high scale and high verification and high longevity and high translation, it is the field guide’s canonical “mature deployment” — and worth treating as such in talks. The number of units that hit all four at high levels is small. Knowing this is more useful than the v4 schema’s deployed value can express.

Maturity-grade summary table for the corpus

This is the first-pass maturity assessment for all current units. Each row shows the four-dimension grades (S/V/L/T, where S0-S4 etc.), plus a one-line summary. The assessment is auditable; any grade can be contested with a unit-level source citation.

Vendor deployments with named scale

Academic / research / institutional deployments

Cooperative / commons / alternative infrastructure

Summary observations

1. The high-scale / low-verification pattern is real and structural. 9 of 22 named-scale vendor units score S3 or S4 paired with V0. This is the canonical vendor-figures-without-independent-audit pattern that the field guide’s freshness rules and gap indexes G-015 / G-020 / G-033 have been tracking. The framework makes this pattern visible at the corpus level rather than case-by-case.

Concrete examples of the V0 / vendor-reported pattern (worth naming in talks):

• DJI Agriculture (units/dji-agriculture-global-export.md): 400,000 drones, 980M acres, 100+ countries — vendor-reported from DJI Annual Report. The unit itself flags: “400,000 / 980M acres / 100+ countries figures are vendor-reported. Independent verification limited.” The 222M tons water saved and 30.87M tons CO2 reduced figures are explicitly vendor-reported and flagged for verification (G-033).
• Bayer Climate FieldView (units/bayer-climate-fieldview.md): 250M subscribed acres (Bayer release). Vendor-reported; the unit’s own critical-context section names this.
• XAG (units/xag-china-drone-leader.md): 10M+ farmers. Vendor-reported.
• Agrosmart (units/agrosmart-brazil.md): 100,000 farmers, 60% water reduction claim. Vendor-reported; verification gap is G-015.
• Lely Astronaut (units/lely-astronaut.md): 50,000 units across 50 countries. Vendor-reported; Lely is a multi-decade equipment-maker so the figure is plausible but not independently audited.
• Taranis (units/taranis-aerial-imagery.md): multi-continent scale. Vendor-reported.
• Indigo Ag (units/indigo-ag.md): boom-bust pattern makes the unit’s current scale modest. The unit is honest about this; the V1 score reflects that the boom-bust itself is well-documented (Agriculture Dive, Farm Progress, Memphis Business Journal, AgTech Dive).

2. The research-stage-without-translation pattern dominates academic units. All 5 USDA-NIFA AI Institutes plus 4 other academic units score T1 or T2 with documented translation gaps. The Extension Foundation 2026 Report (T2, workforce-readiness-as-limit) is the substantive practitioner-side acknowledgment of this pattern.

3. The cooperative / commons cluster sits at the structural middle. JoinData is the only unit scoring S3+ with V2 verification (third-party case study) and T3 translation — mature cooperative infrastructure. NAPDC is S0 / T1 — the same conceptual space, but pre-deployment.

4. Longevity is where Indigo Ag stands out negatively. L1 in a vendor with $1.6B raised and major restructuring is a substantive finding. Other units scoring L1 are predominantly research-stage (Cornell Atkinson IDSov, AI-CLIMATE) — that’s expected, not concerning.

5. Translation is the dimension where the academic / practitioner gap is most visible. 6 academic units score T1; 5 score T2; 1 (INRAE) scores T2; none score T3. By contrast, vendor units cluster at T2-T4. The framework makes the research-to-deployment gap structurally visible.

What this framework is NOT

• Not a replacement for activity-status. The v4 tag is a coarse single-value summary. The framework is the multi-dimension evaluation. Both should coexist.
• Not a vendor-evaluation tool. The framework is for the field guide’s own maturity assessment, not for procurement or rating decisions.
• Not a research-productivity assessment. Translation (T) measures research-to-deployment pathway, not research output quantity.
• Not a vendor-comparison leaderboard. Each unit’s grade reflects the unit’s claim, not a relative ranking against other vendors.

How to use this framework

When writing a new unit:

1. After the unit’s Content section, add a ## Maturity assessment section with the four grades (S/V/L/T) and a one-paragraph justification.
2. If the deployment’s grade changes meaningfully (e.g. from S2 to S3 with documented scale increase), update the unit and bump last-verified.
3. For pure analytical / critical / framework units (where activity-status: not applicable), skip the maturity assessment and use (n/a — analytical) etc. as in the table above.

When reading the corpus:

1. Filter by maturity grade: e.g. “show me vendor units with V2 or higher” surfaces deployments with substantively verified scale claims.
2. Filter by translation grade: “show me academic units with T3 or higher” surfaces research-stage work that has documented operational adoption pathways.
3. Pair dimensions: “S3+ AND V2+” surfaces the rare mature, verified deployments.

When assembling a talk:

1. Lead with units that score high on multiple dimensions — they are the substantive anchor points.
2. Use low-translation academic units as examples of research-stage work and low-verification vendor units as examples of claims that need independent verification.
3. Avoid claiming scale without naming the verification level.

When responding to vendor claims in talks:

1. Quote the figure with the verification level (e.g. “DJI reports 400,000 drones across 100+ countries; this is vendor-reported, not independently audited”).
2. Use the gap indexes (G-015, G-020, G-033) to anchor the verification-gap discussion.

What this framework enables

1. Vendor-figure hygiene at the corpus level. The framework makes the V0 / V2 / V3 distinction visible per unit. Talks that quote vendor figures can pair the figure with the verification level.
2. Honest research-to-deployment assessment. Academic units can be assessed on Translation without claiming operational deployment they don’t have.
3. Substantive cross-corpus comparison. JoinData vs. NAPDC vs. academic translation arms — same conceptual space, different maturity profiles.
4. Pilot vs scale differentiation. Vendor pilot announcements can be flagged as S1 rather than S3.
5. Longevity as a maturity signal. Indigo Ag’s L1 vs. Climate FieldView’s L4 is a substantive difference that the v4 schema’s deployed collapses.

What this framework surfaces for future cycles

G-050 (new — independent verification of vendor-reported agrifood AI scale claims). The V0/V1/V2/V3 distinction in this framework surfaces that 9+ units have V0 vendor-reported scale claims without independent verification. Per memory hygiene, this is exactly the kind of pattern the field guide should track.

G-051 (new — research-to-deployment pathway documentation for academic agrifood AI). The T1/T2/T3 distinction in this framework surfaces that the academic agrifood AI cluster has documented translation gaps. The Extension Foundation 2026 Report (workforce readiness as limit) is the substantive practitioner-side acknowledgment.

G-052 (new — longevity / generational durability of vendor-led agrifood AI deployments). The L1/L2/L3 distinction surfaces Indigo Ag’s L1 as an outlier; broader generational durability is a substantive question worth tracking.

C-041 (new — academic agrifood AI research translates to operational deployment). Counter: the AI Institutes (T1-T2), the research-to-deployment gap is structurally modest. Worth naming honestly in any talk about academic agrifood AI.

C-042 (new — vendor-reported agrifood AI scale figures are reliable). Counter: 9+ vendor units score V0 with no independent audit. Per memory hygiene, this is the structural pattern the field guide should surface.