Outsourcing truths: the nail and the agent

A nail is not a claim

A nail is a tool with almost no politics. It does not invent a species. It does not tell a client that a canopy is climate resilient. It does not migrate a planting schedule into an asset register under a logo that implies the hard questions are already settled. Drive it correctly and it holds timber. Fail, and the failure is local, inspectable, and usually yours.

Agentic software, systems that plan, draft, rearrange, and complete work across files, models, and briefs, is not that kind of tool. It moves. It proposes. It fills gaps. It can look like labour that has already been performed. Used carefully, it can absorb repetition in the sense earlier Field Notes have defended: hygiene, scaffolding, first passes. Used carelessly, it outsources something else entirely: truth, the quiet assignment of consequence to a claim.

The difference is epistemic rather than romantic. A nail does not speak for the site. An agent that writes a methods paragraph, selects a plant list, or completes a BIM attribute table does speak, unless someone in the studio insists on reading what was said.

Evidence of uptake, and of uneven accountability

Practitioner evidence already shows the gap between adoption and professional closure. In the 2024 ASLA Digital Technology PPN survey, 55% of respondents reported using AI in practice, teaching, or research, yet only 27% said it had saved them time, while 48% were unsure and 7% said it had added time. Only 15% of firms allowed billable time for AI experimentation. Most respondents (56%) did not disclose AI use in deliverables, and only 7% said their firm permitted AI-generated content in stamped plans.

The 2025 joint CSLA–ASLA–IFLA follow-up, acknowledging a lower response rate than earlier waves, found approximately half of respondents still incorporating AI into practice, with 75% reporting greater task efficiency where AI was used, and 92% learning by trial and error. Respondents also called for clearer professional scaffolding: 62% for declarations on AI use, 53% for theoretical discussion, and 51% for dedicated class time on tools, ethics, and environmental footprint.

At the wider AEC scale, McKinsey’s analysis of AI in architecture, engineering, and construction estimates that AI could automate roughly 50% of nonphysical work in architecture and engineering and 39% in construction, identifying some 150 workflows across 25 domains, with early gains concentrated in design, modelling, and feasibility. The same analysis projects that AI and related automation could unlock on the order of USD 228 billion in annual value for the United States AEC industry by 2030, and roughly USD 126 billion for European construction, while cautioning that AI is unlikely to be an extinction event for firms and will not, alone, solve the sector’s productivity problem. Those forecasts describe task automation potential, not the automatic transfer of professional meaning.

Read together, the landscape surveys and the AEC forecasts point to the same tension: efficiency can rise while disclosure, training, and sign-off lag. That is precisely the condition under which truths get outsourced.

What gets outsourced when attention slips

Studios rarely outsource truth in a single dramatic gesture. They outsource it through small economies.

Tools arrive first: a generative canopy that looks surveyed; a language model that drafts a climate narrative; a remapper that fills empty fields because empty fields look unfinished. Budgets arrive second: the hour saved is removed from the fee, or never returned to survey, nursery coordination, or cultural protocol. Labour arrives third: the junior who once learned by checking lists now prompts; the senior who once signed meaning now signs volume. Politics arrive fourth: the client wants certainty on a slide; the vendor wants AI capability named in the brief; the council wants digital twin vocabulary that no one on the project owns. Time remains scarce throughout, so the plausible answer becomes the answer that ships.

None of that abolishes landscape architecture. It relocates the interpretive middle. Meaning between data and computation does not disappear; it migrates into defaults, prompts, and product assumptions that are hard to audit after handover. The record still exists. What weakens is the profession’s grip on what the record is allowed to mean.

Compromise the means, not the claim

Practice has always compromised. Specification against cost. Design ambition against maintenance budgets. Community process against programme. Digital work is no exception. Choosing a cheaper asset library, a faster render pipeline, or an agent to draft the first schedule is often rational.

The line worth defending is narrower: compromise the means, not accountability for claims.

A placeholder tree may be acceptable in concept stage. It should not travel silently into construction documentation as if it were a surveyed taxon. An agent may draft a sustainability appendix. That draft should not be presented as settled method because the prose sounded confident. Automation may buy time. That saving should not be sold as proof that ecological or cultural work is finished.

Budgets and politics will keep pressing toward the second option. They always have. The profession’s task is to keep the first option speakable in fees, RFQs, and studio culture, so that faster does not quietly become someone else decided what is true.

Nature and digital nature

Here the line is not merely technical. It is ontological enough to matter in public.

Nature, in the sense this Foundation keeps returning to under land, is the living world and its afterlife: growth, failure, recovery, maintenance, protocol, the thing that does not care how persuasive a model was. Digital nature is representation and simulation: geometry, traits, climate bands, synthetic canopies, twin layers, agent-authored inventories. Useful, sometimes indispensable, and still not the site.

Market and reporting pressure intensifies the risk of conflation. As of the TNFD’s 2025 status reporting, hundreds of organisations had committed to nature-related disclosure aligned with the Taskforce’s recommendations, with voluntary adopter commitments representing on the order of USD 20 trillion in assets under management and more than 500 TNFD-aligned reports already identified in market scans. Later public tallies placed adopters above 700 organisations and assets under management above USD 22 trillion. Those frameworks increase demand for nature evidence. They do not, by themselves, guarantee that a digital twin layer, a generative planting board, or an AI-completed schedule is an honest proxy for living systems.

The danger is not that digital nature exists. Landscape architecture has always worked through drawings that are not the ground. The danger is conflation: when digital nature is treated as a sufficient substitute for encounter with living systems, or when synthetic assets inherit the authority of survey without lineage. Practices such as explicit synthetic provenance exist for that reason: not as theology about technology, but as a refusal to let render-ready vegetation impersonate botanical certainty.

Where is the line drawn? Not at never use AI, and not at the model is the place. The line is drawn wherever a deliverable asks someone downstream (client, council, maintainer, community, future practitioner) to believe something about living systems. At that point, digital nature must declare itself: indicative or specified; generated or surveyed; estimated or evidenced; open or restricted. If it cannot declare itself, it is not yet professional material. It is scenery with credentials.

Holding the hand that holds the tool

A nail extends the hand. Agentic software can extend the studio, or replace the studio’s judgement with a fluent approximation of judgement. The difference will not be settled by banning tools or by worshipping them. It will be settled by whether landscape architecture keeps interpretive labour visible: the slow decisions that bind trace to claim, claim to evidence, and evidence to someone who can still be answerable when the planting fails, the twin drifts, or a resilient line in a brief is challenged years later.

Outsourcing repetition can be wise. Outsourcing truths is how a discipline loses the right to speak for the places it modifies.

Land remains the test. Art remains how belief is made. Technology remains how evidence becomes usable. Data remains the trace. Between them, and between nail and agent, meaning is still work.

Sources

Figures above are summarised from publicly reported sources. Consult the originals for methods, sample sizes, and caveats.

• ASLA Digital Technology PPN (2024). How Landscape Architects Are Incorporating Artificial Intelligence.
• CSLA, ASLA, and IFLA (2025). 2025 AI+LA Practitioner Survey Findings.
• McKinsey & Company (2026). How AI is reshaping the future of the AEC industry.
• TNFD (2025). TNFD Status Report; see also TNFD Adopter announcements for later public tallies.

Next Steps

Open Foundation materials for one studio deliverable are on the adoption guide. A short Archive page for practices is also available.