agentic ai as a tool in industrial design practice
delegating a task without delegating judgement
an ai system can now be asked to pursue a sequence of work rather than produce a single answer. in industrial design, that changes the practical question from what can it say to what can it reliably help a team complete. market research, supplier comparisons and the preparation of a brief are useful examples because their evidence can be checked.
agentic ai describes systems that select and carry out steps towards a goal, often through tools. the degree of autonomy varies. the term should not imply professional accountability, verified understanding or access to every source. those capabilities have to be established for the specific task and system.
a bounded industrial design task
consider a hypothetical enclosure project. the team needs a comparison of material candidates for a defined environment and production method. a useful instruction specifies the conditions, the evidence required, the output structure and what the system should do when a property is unknown. asking for the best material without those boundaries invites a confident but underdetermined answer.
the output should distinguish supplier data, general material properties and assumptions. a number copied from one grade cannot safely stand for an entire polymer family. a polished comparison table makes the uncertainty easier to organise; it does not resolve it.
research prompts worth adapting
compare a small set of office chairs using published dimensions, adjustment functions, materials and repair information. require a source for each factual field and leave undocumented fields blank. follow with a human review of how those fields relate to the specified users.
investigate manufacturing suppliers against an explicit process, volume and region. record capabilities stated by the supplier, dated evidence and questions for direct verification. the shortlist supports a conversation; it does not establish capacity, certification or a commercial commitment.
analyse published user complaints about one product category. group the reported problems, preserve the context of each account and distinguish frequency within the collected sample from prevalence among all users. the result is a set of hypotheses for observation, not a substitute for user research.
quality control belongs in the workflow
the nist ai risk management framework offers a useful reference for managing risks in context. for design practice, establish what can be checked automatically, what needs a domain specialist and which actions require an accountable person's decision. preserve the source trail when the work changes hands.
automation can amplify an incomplete brief as easily as it accelerates a useful one. tasks involving external commitments, confidential information or consequential engineering decisions need clear boundaries. the design team should be able to inspect the assumptions and interrupt the process before an action becomes difficult to reverse.
a worked design example
imagine asking an agent to shortlist suppliers for a moulded enclosure. the first result ranks companies by the number of favourable claims on their websites. this is an orderly output, but its ranking rewards promotional completeness rather than demonstrated suitability.
revise the task around required process, volume, dimensions and documented capabilities. separate verified statements from unanswered questions. require the system to preserve each source and to leave a field unresolved where evidence is missing. the shortlist can then identify topics for a supplier conversation.
evaluate the useful work remaining after checking the claims. a smaller, well sourced list may save more time than an impressive catalogue that requires extensive correction. the exercise also exposes which judgements the agent should not make from public material alone, including commercial capacity and suitability for the particular production commitment.
a practical exercise
give an agent a limited product comparison with six fields and five candidate products. keep a manual version as a reference. verify every cited fact and record unsupported conclusions, missing information and time spent correcting the output. revise the instruction once and repeat. evaluate useful work completed after verification, rather than the speed or volume of the first response. use that result to decide which part of your workflow merits delegation.
the productive role is a tool within an accountable process. a fluent answer is only the beginning of that process.
a visual study
further reading and crosslinks
develop the discussion into a brief with explicit priorities and reviewable design criteria.
f/p industrial design and product development
connect the principle to research, prototyping and validation in a product development process.
practical guide: choosing a design consulting partner
connect this topic to design strategy, workshops and partner selection.
planning a project in this area? discuss it with f/p design.
references
image credits
1. Gillette razor sketching Jackson Hedden industrial design.jpg
creator or credited source: Jackson Hedden · CC BY-SA 4.0
image source and attribution record
scaled and JPEG encoded for Word; no creative alteration. the image retains the licence listed above.
2. 3D printer farm at Hackerspace Wrocław.jpg
creator or credited source: Majabojarska · CC BY 4.0
image source and attribution record
scaled and JPEG encoded for Word; no creative alteration. the image retains the licence listed above.
3. Industrial-robots-21.jpg
creator or credited source: Haophuong21 · CC BY-SA 4.0
image source and attribution record
scaled and JPEG encoded for Word; no creative alteration. the image retains the licence listed above.
