The Standard · Working with agents
Building custom assistants
Sets up a purpose-built assistant with the right instructions, knowledge and guardrails for a team's work.
You set instructions, approved knowledge and guardrails for assistants supporting recurring professional workflows. You can gain rapid drafts, yet stale sources, invented claims or unsafe actions can mislead you.
The exam for Building custom assistants is not open yet. A skill opens once its questions are approved and tested.
What it measures
Whatever the skill, the exam rates your work in the same six areas of the Standard. Three are for the value you get from the assistant, and three for the control you keep. Each area gets one of 4 ratings (Emerging, Developing, Proficient, Accomplished). For Building custom assistants, a strong performance looks like this. What each area includes and excludes.
Value
What you get from the assistant: how far it takes the work, how well you brief it, and the piece you hand in.
- Reaching further
You extend the custom assistant to cover edge cases, alternatives and reader needs beyond the initial workflow.
Using the assistant to produce work beyond what you could have produced alone, rather than the same work faster.
- Briefing the assistant
You give the custom assistant its users, purpose, source knowledge, response format and required behaviours.
Giving the assistant the reader, purpose and source facts the work depends on, and the form, length and contents you need back.
- The finished piece
You turn the assistant's reply into a deployable configuration, refining instructions, knowledge and guardrails for its users.
Turning the assistant's output into something a colleague could use: editing it, combining it and adding original thinking.
Control
What you keep in hand: checking its work, keeping information safe, and deciding what stays with a person.
- Checking its work
You check the custom assistant's retrieved claims and assumptions against approved knowledge before release.
Checking the assistant's claims, figures and assumptions against the source before the work is used.
- Safety and confidentiality
You keep personal and restricted data out of the assistant's knowledge and outputs, and block unsafe behaviours.
Handling confidential and personal data, seeing compliance, ethical and reputational risk, and declining unsafe requests.
- Knowing what to delegate
You decide which workflow steps the custom assistant handles, which stay human, and who approves unresolved decisions.
Deciding what to hand to the assistant and what to keep, and at what stakes, before any output exists.
What you do in the exam
Every skill of the Standard is examined the same way, at the same pass mark, in the Applied AI Badge. Two tasks with the assistant and two sets of questions, over 15 minutes. You do real work with an AI assistant beside you, hand in what you would send, then answer follow-up questions about your own work.
- 01
A custom assistant for triaging recurring professional requests from approved source material.
- 02
A guarded assistant that produces structured research or workflow drafts for specialist teams.
- 03
An improved assistant configuration that handles edge cases, cites knowledge and routes decisions to people.
These are the kinds of task the exam sets. Your exam draws its own from a bank the Examination Board has approved.
- Exam
- Applied AI Badge
- Price
- US$19
- Length
- 15 minutes
- Attempts
- One for each purchase, then a 7-day wait
What the badge certifies
What an employer reads
Passed an Institute exam in building custom assistants. They briefed the assistant with team instructions and sources, caught unsafe gaps, and delivered a guarded custom assistant.
You earn it in one exam, never by finishing a course.
- Name on the certificate
- Applied AI Badge: Building custom assistants
- Who it is for
- Team leads, workflow designers and service owners who build assistants for recurring work.
- Valid for
- 2 years, then renew
- Verification
- Every badge has a public verification page an employer can open, and a place in the public register if you choose it.
- Marked by
- An AI examiner marks your work against a marking guide and quotes your own words as evidence. A person checks anything borderline.
Near it in the Standard
This skill applies to any job, so it sits in the Standard's core rather than in one function.
- Judging levels of AI autonomy
Tells assistance from automation from autonomy, and matches the level to the stakes of the task.
- Connecting AI to your tools
Chains AI with the documents, data and software a job already runs on, safely.
- Delegating work to agents
Hands multi-step work to AI agents with clear goals, checkpoints and a human decision where it matters.
- Governing autonomous AI
Sets the rules, oversight and accountability for AI that acts on a team's behalf.
Practise it before you pay for it
Practice is free (up to 50 sessions a day) and shaped like the exam, with a coach that tells you what to change. It never counts towards a credential, and an exercise you practised is kept out of your exam.
Want to learn it first? AI+ At Work: Advanced covers this skill. No book is required to pass.