Frameworks & Templates4 min read

AI Skills: Build a Work Portfolio That Proves Them

Turn AI skills into three safe work samples that show the task, rules, first attempt, human checks and final decision.

Bokili EditorialPublished September 7, 2026· Verified September 7, 2026

Three safe AI work samples show a brief, an AI-assisted draft, human correction, verification and a final artefact; certificates sit behind them.

AI skills are hard to show with a course list alone. A certificate can record that you completed learning, but it does not reveal how you framed a work problem, protected information, checked an AI-assisted draft, corrected a plausible error or made the final decision. A small portfolio can make that process visible without exposing confidential work.

The goal is not to display polished AI output. It is to show judgment. Build three compact work samples around realistic, low-risk tasks. Each sample should preserve the problem, the rules, the first attempt, the oversight and the final result. That gives a manager, mentor or interviewer something concrete to discuss.

What evidence actually shows AI skills?

NIST’s AI Use Taxonomy starts from human goals and describes work as tasks made from activities. That is a useful way to choose portfolio pieces: start with a real outcome, then show the specific human and AI activities involved. The US Office of Personnel Management notes that work samples are most informative when they mirror job tasks and allow behaviour or task results to be observed. A learning portfolio is not a formal hiring test, but it can borrow that practical principle.

	Course list only	AI-skills work sample
Shows	Participation and topic coverage	Task framing, process and result
Makes visible	What was studied	What the person checked and changed
Risk	May stay abstract	Must use safe, shareable material
Useful question	Which course did you take?	Why did you make this decision?

Build every sample with the PROOF framework

PROOF: five parts of a credible work sample

P — Problem

Name one narrow, realistic task and the person who needs the result.

R — Rules

State the data boundary, source limits and observable quality criteria.

O — Original attempt

Keep the first AI-assisted output or a short excerpt, including one visible weakness.

O — Oversight

Show the checks, corrections and human judgment applied before use.

F — Final

Present the safe final artefact and record the decision or next action.

PROOF prevents the portfolio from becoming a gallery of attractive answers. The first attempt matters because it creates evidence of recovery. The rules matter because good AI work includes knowing what not to enter. The final decision matters because responsibility does not transfer to the tool.

Worked example: an operations exception brief

Imagine an operations coordinator who wants to demonstrate AI skills in summarising and verification. They create fictional notes about three delayed deliveries. The notes include times, depot codes, owners and confirmed next actions, but no customer names, addresses or live business data. The task is to produce a morning exception brief for a supervisor.

From fictional notes to a portfolio card

1
Define the problem
Turn three fictional incident notes into a brief that helps a supervisor prioritise follow-up.
2
Set the rules
Use only the supplied notes, separate facts from assumptions, preserve owners and flag missing causes.
3
Keep the first attempt
The draft wrongly infers that weather caused one delay and gives it the highest priority without evidence.
4
Apply oversight
Trace each claim to the notes, remove the invented cause, restore the agreed priority rule and add an evidence-gap label.
5
Publish the safe final
Show the corrected brief, the checks performed and one sentence explaining the final prioritisation decision.

This example demonstrates more than prompting. It shows task selection, input control, source fidelity, error detection, revision and accountable decision-making. Those are durable skills even when the interface or model changes.

Reading is a start. Practice makes it stick.

Start learning

Choose three samples, not thirty

A useful starter portfolio covers different kinds of judgment. One sample might organise information, one might compare options and one might draft a communication. Keep the format consistent so the reader can see progress across tasks. A single page per sample is usually enough: the aim is inspection, not spectacle.

Sample 1: organise a safe set of notes into a structured brief.
Sample 2: compare two options against pre-written criteria.
Sample 3: revise a draft after finding an unsupported claim or missing constraint.
For each sample, include the prompt only when it helps explain the decision.
Remove names, account details, proprietary figures and confidential source material.

If you cannot share the real task, recreate its structure with fictional or public information. Label the sample clearly. Never imply that simulated work was delivered to a real client or employer.

Use AI skills evidence honestly

The European Commission describes AI literacy as context-specific: the appropriate knowledge and safeguards depend on the system, setting, people and risks. It also says the EU AI Act does not require a particular certificate. Your portfolio can support a learning conversation, but it is not proof that an employer has met every legal or organisational duty. Keep claims narrow: say what task you practised, which checks you used and what you can now do with supervision.

Ask a reviewer to challenge one decision in every sample. A useful review question is: “What would change your conclusion?” Update the card with the answer. This makes the portfolio a living learning record rather than a static claim of expertise.

Create your first AI-skills portfolio card in ten minutes

Turn one safe practice artefact into PROOF

Choose a completed, non-sensitive practice task.
Write one sentence each for the Problem and Rules.
Save one weakness from the Original attempt.
List the checks and correction under Oversight.
Add the Final artefact and the human decision.
Remove sensitive details and ask someone to question one choice.

Three honest work samples can tell a stronger story than a long list of tools. They show how you work when the first answer is incomplete, the evidence is limited and a person still has to decide.

Continue your AI learning path

AI Course for Beginners: Build a Safe Work Sample — https://bokili.com/en/learn/ai-course-beginners-safe-work-sample
Learn AI at Work: A 30-Day Beginner Path — https://bokili.com/en/learn/learn-ai-at-work-beginner-path
AI Fluency Begins After the First Wrong Answer — https://bokili.com/en/learn/ai-recovery-skill-wrong-answer
How to Verify AI Output at Work — https://bokili.com/en/learn/how-to-verify-ai-output-at-work

Bokili helps people practise AI through focused work situations. Explore the learning experience at https://bokili.com/en.