Follow the links below to read the originals. Each entry explains what you can learn and what to keep in mind. The public pages were reviewed on 7 October 2026; publication dates are left blank when unstated. Community guidance and vendor experience do not establish compliance, certification or impact.
See how this role spans discovery, technical scope, building, production rollout, adoption and product feedback.
Keep in mind / These are OpenAI’s hiring requirements, including engineering experience and coding ability. They are not an industry-wide standard or a credential from this course.
See how Palantir describes Echo (Deployment Strategist) and Delta (forward deployed software engineering).
Keep in mind / These are Palantir’s team labels. Use them to distinguish responsibilities in the course, but check the actual division of work at an employer or customer.
Learn how to research users and the whole service before building, including support staff and people with different needs.
Keep in mind / The guidance concerns government digital services. Check access, participant recruitment and research ethics for your own project.
Review AI lifecycle risks through governance, mapping, measurement and management.
Keep in mind / This is a voluntary risk framework. Using it does not establish compliance with a jurisdiction’s or sector’s legal requirements.
Consult the 2026 LLM application security resources to look for gaps in your risk register.
Keep in mind / This review confirms the official resource page, but does not cover every item in the download. Consult the original for specific risks; no ranking changes or certification are established here.
Use this resource as a starting point for identifying risks when agents plan and execute actions.
Keep in mind / This is not a ranking of how often risks occur. Test your system’s actual tasks and assign accountable risk decisions.
Explore how to make agent behavior visible and traceable, and enforce policies at runtime.
Keep in mind / The resource page and current repository README were reviewed, not every specification clause. Verify the implementation version, configuration and behavior when controls are unavailable, and test action boundaries. This is not a legal standard or a complete governance system.
Learn to distinguish tasks, trials, execution traces, grading and outcomes when evaluating multi-turn agents.
Keep in mind / This is engineering guidance, not a universal acceptance standard. Domain experts still need to check and calibrate automated grading.
Learn to manage instructions, tool descriptions, retrieved material and history within a limited context window.
Keep in mind / Better context management does not replace authorization, data governance or knowledge version control.
Separate session records, the agent execution loop and the sandbox to understand recovery and isolation for long-running work.
Keep in mind / This is one vendor’s architecture experience. Managed products still require customer-side ownership of access, data and business decisions.
Start with simple patterns and choose a fixed workflow or a more autonomous agent according to the task.
Keep in mind / Test the choice against costs, actual tasks and the consequences of errors.
Learn to detect problems and roll back through limited, time-bounded, comparable releases.
Keep in mind / Agree separately on business thresholds, stop authority and available human support.
Understand users, problems and constraints before deciding whether further investment is worthwhile; useful when a requested feature is being treated as the problem.
Keep in mind / Written for UK government digital services. The course adapts research and decision principles, not phases, schedules or approval requirements as universal enterprise policy.
Check that participant explanations cover purpose, recording, sharing, voluntary participation and withdrawal, so people understand what they are agreeing to.
Keep in mind / Not local legal advice. Customer permission and participant arrangements must be checked against applicable rules by the appropriate specialists.
Collect less around a research question, agree access, retention and sharing, and check whether anonymized outputs could still identify people.
Keep in mind / Includes UK personal-data rules and retention suggestions. They do not automatically set your project's retention period or establish compliance.
Prepare open, neutral interview prompts, investigate concrete experiences and check understanding before the session ends.
Keep in mind / Interviews explain experiences and motivations but do not alone establish population frequency or causal effects. Other evidence must fit the question.
Separate what was seen or heard from interpretation, then organize findings, actions and further questions.
Keep in mind / Grouping observations is not causal proof. The course also preserves alternative explanations and serious exceptions; important contradictions must not be removed for a tidy conclusion.
Plan the experience of first use, everyday interaction, errors and changes over time.
Keep in mind / The date is the project page’s establishment date. The guidelines inform design; they do not replace usability testing or risk decisions for a particular task.
Check whether users can recognize errors, understand limitations and verify generated content.
Keep in mind / The article was read, but its original publication date was not established. Mitigating overreliance needs ongoing research; showing sources or explanations is not a guarantee.
Extend risk review to generative AI, including privacy, human–AI interaction, information integrity and component dependencies.
Keep in mind / The release date is confirmed by NIST’s framework page. Relevant risk descriptions were reviewed, not every suggested action. This voluntary framework does not establish legal compliance.
Helps define task starts and completion signals, distinguishing visits, attempts and completion.
Keep in mind / Written for UK government digital services. This course additionally counts eligible tasks not entering the system; transaction definitions and publication requirements do not transfer automatically.
Helps connect support demand, handling time, staffing and user feedback to pilot support.
Keep in mind / A support-planning method, not evidence of enterprise AI adoption outcomes or a universal response-time target.
Helps distinguish cash and other benefits, updating options, costs and assumptions as research develops.
Keep in mind / A public-service estimation and review method; it does not validate teaching figures, return forecasts or causal claims.
Helps include channels and human support in costs rather than count model calls alone.
Keep in mind / Contains UK public-sector accounting and reporting definitions. Enterprise lifecycle, cash and opportunity costs need separate treatment under actual financial rules.
Helps assign incident coordination, technical response and communication responsibilities, with action records.
Keep in mind / An operations-engineering practice. Business risk acceptance, data use and formal notification still require actual authorized owners.
Helps review timelines, system conditions and corrective actions instead of focusing only on individual blame.
Keep in mind / A learning method, not a guarantee against recurrence or a substitute for domain evaluation and recovery checks.