Choose one concern. Read the evidence. Practise a capability. Show your reasoning.
These five starting points are editorial choices, not a ranking of the profession’s worries. Automation can change tasks within a role, but potential exposure is not the same as adoption, job loss or a forecast of your career. Foundational research, design and writing skills remain part of these activities.
What if the problem is a crowded job market?
Competition for roles and changing skills expectations are different questions. A claim that there are “too many UX practitioners” needs a defined location, period, seniority level and source. Vacancy counts, redundancies and applicant numbers measure different things; none alone establishes the prospects of the whole profession.
These pathways cannot establish the current balance of UX labour supply and demand. Their sources do not measure it. Learning can strengthen your practice, but cannot create vacancies or guarantee a job. Difficulty finding work should not automatically be attributed to an individual skills deficit.
Read employment claims critically: look for the original data, dates, occupational definitions and method. Separate observed hiring changes from forecasts, and check whether junior and experienced roles are distinguished. Our white-paper collection provides a starting point for examining sources, not a definitive forecast.
First collection · 18 October 2026 · 5 pathways
Start small: choose one pathway for this week. Use the beginner route for a bounded practice example, or the experienced route to examine an existing workflow. The scope determines the time needed; a complete user study takes longer than drafting a plan.
The ILO examines potential task exposure; the critical-thinking study examines reported behaviour. Neither establishes that UX roles will disappear. Our learning response is to make reasoning and verification visible.
Choose your starting point
Beginning your UX journey: Choose one familiar interface. Write its user problem, constraints and three evaluation criteria before asking an AI tool for suggestions. Keep your own first draft.
Established practitioner: Map one real workflow into tasks. Identify where automation may assist, where specialist judgement is required, and who checks an output before it affects users.
Practise
Compare one generated proposal with your criteria and original evidence. Mark each claim as supported, unsupported or uncertain.
Reject or revise at least one weak suggestion and explain the decision. Keep a brief record of your own contribution and the tool’s role.
Portfolio evidence: An annotated before-and-after example and a 300-word decision rationale. Show verification, not just a polished final screen.
Review question: Can another reader trace each important claim to evidence and understand why you changed the design?
Learn enough to question capabilities, limitations and dependencies.
Human–AI interaction guidance supports communicating limits and handling errors. It does not establish a universal technical syllabus or prove that every designer needs to become an engineer.
Choose your starting point
Beginning your UX journey: Choose one AI-enabled feature and storyboard its input, output, failure and fallback. A paper prototype is sufficient; no paid tool is required.
Established practitioner: Work with a technical colleague to map model, data, interface and external-service dependencies. Separate what is demonstrated from what is assumed.
Practise
Write three cases: ordinary use, an ambiguous request and a plausible but incorrect response. Define what the user should see and be able to do.
Document how you would check each case and what the prototype cannot test. Do not present a simulated response as a working integration.
Portfolio evidence: A capability-and-limit sheet, three interaction sequences and a small evaluation plan.
Review question: Can you explain the feature’s limits in plain language and identify a credible fallback?
Make delegation, intervention and recovery explicit.
Human–AI guidelines offer a foundation for control and correction. Risk-management guidance adds responsibilities. The activities below are editorial applications to agents, not validated agent-design rules.
Choose your starting point
Beginning your UX journey: Storyboard an assistant that shortlists exhibitions for a visitor. It may suggest an itinerary but must ask before making a booking or sharing information.
Established practitioner: Audit an agent workflow for permissions, approval points, visible state, interruption and irreversible actions. Include someone affected by the action who is not the direct user.
Practise
Draw the boundary between a suggestion and an authorised action. Add a permission change, a failed action and a cancellation request.
Show what can be undone, what cannot, and how the user gets help. Test the logic using fictional scenarios before involving participants.
Portfolio evidence: A permissions table and an annotated flow showing normal operation, interruption and recovery.
Review question: Can the user tell what happened, stop the next action and understand the limits of undo?
Connect research and design decisions to outcomes that matter.
HEART connects goals to signals and metrics. It provides a measurement framework, not proof that UX always increases revenue or that a single metric captures experience.
Choose your starting point
Beginning your UX journey: Choose a task such as finding an accessible exhibition time. Define success from the visitor’s perspective before choosing a measure.
Established practitioner: Choose an upcoming product decision. Agree the user outcome, baseline, guardrail and evidence needed to change direction; record alternative explanations.
Practise
Specify one behavioural measure and one question about the experience. Explain how each informs the decision, rather than counting activity for its own sake.
Draft a short findings report that separates observations, interpretation and recommendation. If you have no study data, label it an evaluation plan and leave the results blank.
Portfolio evidence: A goals–signals–metrics table and a one-page decision brief. Never invent participants, outcomes or return-on-investment figures.
Review question: Would the measures detect a worse user experience even if clicks or time spent increased?
Build accessibility, privacy and accountability into the work.
NIST provides a structure for managing AI risks, while interaction guidelines support feedback and correction. Neither replaces accessibility evaluation, research ethics approval or applicable legal advice.
Choose your starting point
Beginning your UX journey: Review your prototype for an inaccessible interaction, unnecessary data collection and an unclear automated decision. Describe who each could exclude or disadvantage.
Established practitioner: Map owners, evidence, escalation routes and unresolved risks for one release. Identify where time pressure has removed necessary checks.
Practise
For each issue, write a concrete design change, an owner and a way to check whether the change works. Include a non-AI or human-support route where appropriate.
Write a release recommendation: proceed, revise or stop, with reasons and remaining uncertainty. Seek specialist input where your competence ends.
Portfolio evidence: A prioritised issue log and a short, evidence-based release recommendation.
Review question: Could a colleague act on your recommendations, and are unresolved issues visible rather than hidden by a checklist?
These are background readings, not newly published news. Three peer-reviewed conference papers, an occupational-exposure report and official risk-management guidance serve different evidential purposes. Review depth is stated below.
Generative AI and Jobs: A Refined Global Index of Occupational Exposure
ILO Working Paper 140 · 2025 · background report Open report · publisher summary reviewed; full report not reviewed here
Estimates potential occupational exposure using task-level assessment. This helps frame questions about changes within jobs.
Limitations: Exposure is not observed job loss, adoption or a prediction of an individual UX career. This is not a survey ranking UX practitioners’ concerns.
Lee and colleagues · CHI 2025 · peer-reviewed empirical conference paper · background Open author manuscript · findings and limitations reviewed in full-text source
A survey of 319 knowledge workers found an association between higher confidence in generative AI and less reported critical thinking. Participants described verification, integration and oversight work.
Limitations: Self-reports do not establish causation or measured skill loss. The English-speaking sample skewed towards younger, technically skilled regular AI users and was not specific to UX practitioners.
Amershi and colleagues · CHI 2019 · peer-reviewed conference paper · background Open author manuscript · full text reviewed
Eighteen guidelines address expectations, feedback, correction and user control, developed and evaluated through several stages.
Limitations: Evaluating guidelines is not proof that a particular interface improves outcomes. The systems studied predate current LLM agents; test applicability in context.
Measuring the User Experience on a Large Scale: User-Centered Metrics for Web Applications
Rodden, Hutchinson and Fu · CHI 2010 · peer-reviewed conference paper · background Open author manuscript · full text reviewed
Introduces HEART and a goals–signals–metrics process, with product examples connecting user-centred goals to measurement.
Limitations: A framework with illustrative examples, not a controlled demonstration of commercial return. Metrics need interpretation; engagement alone can reward an unhelpful experience.
NIST · AI RMF 1.0, 2023 · official voluntary guidance · background Open web guidance · Core section reviewed
Organises risk-management activities into Govern, Map, Measure and Manage. Useful for assigning responsibilities and planning evaluation.
Limitations: Normative guidance, not an empirical finding, certification or statement of legal compliance. Applying it requires context-specific expertise.
Activities are editorial learning suggestions, not validated training interventions, qualifications or guarantees of employment. Use fictional or approved public material; do not upload confidential work to an unapproved tool. Obtain any required ethics approval before involving participants. State your own contribution, any AI assistance and the limits of your evidence.