In the end, nothing moves unless we go and ask Yamauchi
again, does it?
It was a Wednesday morning in the meeting room, three months into the company-wide rollout of AI agents, when Saeki from HR planning muttered exactly that. Across sales, finance and IT, people had begun using AI agents to draft minutes, handle first-line enquiries and rough out documents. In practice, though, it came down to a handful of willing volunteers fielding questions on Slack day after day. “I can see it’s useful,” the line managers would say, “but I’ve no idea who I’m meant to hand it to.”
Today, the company has a training curriculum and graded internal AI certifications in place, splitting development into three roles: foundation, practitioner and champion. Over the past six months, in-house training across four target departments and 86 participants saw 23 people pass a practical examination and be registered as their department’s AI-agent specialists, giving the front line a clear point of contact at last.
This article is for the HR leaders, learning-and-development specialists and corporate planners who want to put a reskilling certification scheme or internal qualification in place. It sets out how to design the role definitions, curriculum, practical examinations, incentives and update cycle. The aim is a state in which you no longer depend on a few knowledgeable individuals, but instead grow people in every department who can use AI agents safely and improve how they are used. That said, a certification scheme alone will not make adoption stick. It only pays off when combined with sensible task design, support from managers, and regular review. If you, too, feel that the number of capable people simply isn’t growing, it may be worth revisiting the design of the scheme itself.
Why AI-agent specialists need an internal certification scheme
As AI agents take hold, the first to deliver results are often a handful of employees who are naturally curious and comfortable with new tools. They try out fresh approaches, refine their prompts, weave the technology into their own work and, along the way, teach the people around them.
In the early stages, these willing volunteers are a tremendous driving force. Yet the wider adoption spreads, the more plainly the limits of leaning on volunteers come into view.
Sales, say, want AI agents to knock up first drafts of proposals; finance want to streamline first-line responses to internal queries; HR want to put together training content and FAQs. As these requests pile up, questions and consultations all converge on the same knowledgeable few.
The upshot is that problems like the following tend to crop up.
- The gap widens between those who can use AI agents and those who cannot
- Usage and safety standards vary from one department to the next
- It is unclear who is able to lead practical adoption
- Line managers cannot draw on it for staffing or promotion decisions
- The volunteers are swamped with queries, and the company-wide rollout grinds to a halt
What works to head this off is an internal AI certification scheme.
An internal AI certification is not merely a scheme that confirms someone “attended generative-AI training”. It is a mechanism for advancing AI talent development as an organisation, by setting out clearly the skills, scope of responsibility, roles and assessment criteria needed to use AI agents in real work.
In recent years, the skills gap and the need for reskilling that come with AI adoption have been flagged internationally as well. The OECD report on the AI skills gap, for one, and the World Economic Forum’s Future of Jobs Report 2025 both take up the shifting nature of skills in the age of AI and the importance of reskilling. Within the company, then, making it visible who can use AI, at what level and for which tasks becomes important on both the talent-development and the governance fronts.
Define the people you can “entrust it to”, not merely those who “can use AI”
When building an internal certification scheme, there is one thing to settle first: drawing a distinction between people who “can use AI” and people you can “entrust AI-agent work to”.
Being able to put a question to an AI chat is one thing; being able to build an AI agent into a business process is quite another. And once company data or customer information is involved, you also need information governance, output checking, risk judgement and coordination with the relevant departments.
For that reason, certifying an AI-agent specialist means confirming the following four capabilities.
The ability to identify the business problem
An AI agent is not a tool for automating every last task. What you need first is the judgement to work out which tasks it should be used for.
Even where the request is, say, “let’s put enquiry handling onto AI”, handing over every enquiry simply isn’t realistic. You need to separate the routine questions answerable by consulting the rules, the questions where AI can be trusted with a first response, and the questions that call for human judgement.
A certification scheme checks whether someone can break the work down before reaching for AI and decide its scope of application.
The ability to design appropriate instructions and constraints
The quality of an AI agent turns on the substance of the instructions it is given.
Rather than “give me a decent answer”, you need to spell out the intended reader, the materials to draw on, the output format, what is off-limits, and what to do when something is unclear.
For internal use especially, constraints such as “say so when you don’t know”, “cite the supporting material” and “don’t answer by guesswork” matter a great deal. These are not merely there to make things more convenient; they are the basic design that guards against wrong answers and excessive automation.
The ability to verify the output
An AI agent’s answers are not always correct. Where company rules, contract terms, customer information or figures are involved, the design has to assume a human will check the result.
Those certified are expected to be able to set the output against the original sources and correct errors or overstatement, rather than using the AI’s output as it stands. For text going outside the company or material feeding into decisions in particular, the step of checking proper nouns, dates, amounts and citations is indispensable.
The ability to explain it to those around them
AI-agent adoption does not spread on individual skill alone. To make it stick on the ground, you need people who can explain how to use it, answer questions and share their own failures.
When certifying someone as an internal qualification, then, it is important to design in not only their own operational skill but also the role of supporting those around them.
Design the internal certification scheme around three roles
In advancing AI talent development, there is no need to demand the same level of everyone. Indeed, it is easier to run if you separate certification levels according to the person’s role.
Here we introduce a way of designing it in three tiers: foundation, practitioner and champion.
| Certification tier | Who it is for | Main role | Capabilities checked |
|---|---|---|---|
| Foundation certification | All employees, or members of departments that use it frequently | Use AI agents safely in everyday work | Basic operation, information governance, output checking |
| Practitioner certification | Core members in each department | Build AI agents into their own department’s work | Task breakdown, prompt design, improvement proposals |
| Champion certification | People who drive AI adoption across departments | Support the scheme, training, governance and improvement | Cross-department coordination, risk judgement, operational design |
Foundation certification: people who can use AI agents safely
Foundation certification is aimed at all employees, or at members of departments that use AI agents frequently.
The aim is to reach a state in which AI agents can be used safely in everyday work. It does not call for sophisticated task design; it checks whether someone understands the basics of how to use the tool, how to manage information, and the rules for checking output.
The items checked for foundation certification might be along these lines.
- Understands which tasks an AI agent may be entrusted with and which it may not
- Understands how to handle personal and confidential information
- Can include the purpose, assumptions and output format in a prompt
- Understands that AI output must not be sent outside the company as it stands
- Can check uncertain content with a manager or the relevant department
Foundation certification is the entry point for broadening the base of AI-agent adoption across the company. Rather than making it a difficult examination, the emphasis is on bringing everyone up to a minimum safety standard.
Practitioner certification: people who can build it into their own department’s work
Practitioner certification is aimed at the core members driving AI-agent adoption in each department.
At this level, it is not enough simply to be able to use AI; you are expected to be able to design how it is used to suit your own department’s work.
The items checked for practitioner certification might be along these lines.
- Can break down a business process and decide an AI agent’s scope of application
- Can organise reference materials and prompts
- Can create checklist items for assessing output quality
- Can explain to front-line members how to use it
- Can record points for improvement and feed them into the next round of operation
In sales, for instance, an AI agent that organises the key points for the next proposal from meeting notes; in HR, one that gives first-line responses to internal enquiries — the focus is on uses close to real work.
As this tier grows, AI adoption shifts from “something you ask the experts about” to “something each department improves for itself”.
Champion certification: people who take charge of design, training and improvement across departments
Champion certification is aimed at the people who lead the company’s AI-agent adoption as a whole.
In this role, beyond improving individual tasks, you need to keep governance, training, cross-department coordination and handling of updates within view.
The items checked for champion certification might be along these lines.
- Can organise cross-department themes for AI adoption
- Can weigh the balance between risk and convenience
- Can take part in improving the certification scheme and curriculum
- Can coordinate with IT, legal, HR and front-line departments
- Can respond to changes in AI-agent specifications and in internal rules
A champion is not the AI-agent “teacher”. Their role is to gather the successes and failures from the front line, fold them back into the scheme and the training, and move the whole organisation’s operation forward.
Build the curriculum back from real implementation, not from knowledge-based training
A common failing of internal certification schemes is that they end up teaching nothing more than general knowledge about generative AI.
The mechanics of AI, the basics of prompting and risk management are of course all necessary. But on their own they will not grow people who can actually use it on the ground.
Design the curriculum by working back from “which tasks do we want them to be able to change once they’ve completed it?”.
Use a common curriculum to bring everyone to the same standard
Start by preparing a foundational curriculum common to every certification level.
The common curriculum should include content such as the following.
- The difference between generative AI and AI agents
- Tasks AI agents are good at, and tasks they are poor at
- Information governance and the information that must never be entered
- The basics of prompt design
- Output checking and fact-checking
- Internal rules and the usage-request workflow
- Reporting procedures when something goes wrong
With this common core in place, it becomes easier to bring everyone to a minimum safety standard, even where usage differs from one department to the next.
Use role-specific curricula to connect with real work
Next, design curricula tailored to each job function or department.
Sales, marketing, HR, finance and IT each want to entrust different tasks to AI agents. With common training alone, it is hard for people to see how to apply it to their own work.
- Sales
- The focus falls on organising meeting notes, drafting proposals, and preparing likely questions for each customer.
- Marketing
- The focus falls on article outlines, advertising copy, and shaping the appeal for each customer segment.
- HR
- The focus falls on creating training content, handling internal enquiries, and assisting with first drafts of appraisal comments.
- IT
- The focus falls on help-desk support, searching internal manuals, analysing usage logs, and access design.
Using a tool that lets you handle training management, AI chat, prompt management and learning-data management within the same environment makes it easier to connect training with real work. A service that brings together e-learning, AI chat and project-level library management — such as Kanata — is one option where you want to keep updating departmental materials and prompts over time.
Use a practical examination to confirm “can they use it on the ground?”
What matters most in an internal AI certification is the practical examination.
A written test alone may show whether someone understands the terminology, but it is hard to judge from it whether they can use it on the ground. AI-agent adoption calls for the ability to read a business problem, craft suitable instructions, verify the output and put forward improvements.
An example of designing the practical examination
For practitioner certification, for instance, you might set a task like the following.
Participants are given a fictitious log of internal enquiries, an extract from the staff rules, and an existing response prompt. On that basis, they are asked to carry out the following.
- Classify the content of the enquiries
- Decide the range that may be entrusted to the AI agent
- Improve the response prompt
- Write constraints to guard against wrong answers
- Check the resulting output and explain the points to be corrected
With a practical examination of this kind, you can confirm not mere knowledge but the judgement needed in real work.
Set out the assessment criteria explicitly in a rubric
For the practical examination, prepare a rubric to guard against variation between assessors. A rubric is a marking sheet that sets out the assessment items and the standards for meeting them in advance.
The assessment items might include the following.
| Assessment item | What is checked |
|---|---|
| Understanding of the business problem | Whether the background and purpose of the task have been grasped correctly |
| Judgement of scope of application | Whether the part left to AI and the part carried by people have been separated |
| Prompt design | Whether the purpose, assumptions, constraints and output format are clear |
| Risk management | Whether there are safeguards against personal information, confidential information and wrong answers |
| Output verification | Whether the AI’s answers are checked against the supporting sources |
| Improvement proposals | Whether there are proposals that lead to better operation next time round |
Preparing not only a score but also feedback for those who fall short of the pass mark helps the certification scheme work as a mechanism for development.
Design the deployment and incentives that follow certification
An internal qualification is not something that ends once obtained. Unless you decide in advance what role a person will take on after certification, the scheme tends to become an empty formality.
Make the certified the front line’s point of contact
Employees who have gained practitioner or champion certification are positioned as the point of contact for AI-agent adoption in each department.
You might place certified people two in sales, one in HR, two in IT, and so on, so that there is always someone able to field queries from the front line.
That said, if queries pile up too heavily on the certified, the reliance on individuals returns. So you need an arrangement that records the substance of queries as knowledge and feeds it into an FAQ or prompt library.
Connect it to appraisal and deployment
To keep developing AI-agent specialists, it is also important not to divorce the certification scheme entirely from HR appraisal and deployment.
You might, for instance, consider the following.
- Deploy employees who gain practitioner certification onto business-improvement projects
- Bring employees who gain champion certification into cross-department projects
- Cover certified employees’ activities in appraisal interviews
- Make business improvements through AI use eligible for internal recognition
Combining not only monetary incentives but also opportunities to take on challenges, visibility within the company and the granting of roles makes the meaning of gaining certification clear.
Assume an update cycle from the outset
AI-agent features and internal rules keep changing. An internal certification scheme, therefore, cannot be built once and left.
On the contrary, what is needed is a scheme designed on the assumption of updates.
Set an expiry on the certification
We would recommend setting an expiry on an internal qualification.
For example, renewing foundation certification every year and practitioner and champion certification every six months or year. At renewal, you would go over the new internal rules, changes to AI-agent specifications, past incidents, and improved prompts.
Setting an expiry lowers the risk of the scheme running on out-of-date knowledge.
Feed failures back into the curriculum
In AI-agent adoption, failures are as valuable a teaching resource as successes.
Take cases such as the following.
- The AI answered by referring to an out-of-date internal rule
- A figure from the output was put into a document without being checked
- Text containing customer information was entered into an unsuitable environment
- The prompt’s constraints were weak, and the AI answered by guesswork
Rather than treating such cases as grounds for blame, folding them into the next round of training and practical examinations raises the scheme’s real effectiveness.
In choosing tools, look at how training, real work and knowledge management connect
When running an internal AI certification scheme, which tool you use also matters. That said, introducing a particular tool will not in itself make the scheme a success. What matters is whether it can support the operation the certification scheme will require.
There are four points worth checking.
- Can training content be managed by participant and by tier?
- Can you provide an AI-chat environment close to practical exercises?
- Can prompts, materials and assessment criteria be kept up to date over time?
- Can access rights be separated by department or by project?
You can also combine an existing LMS, internal portal, knowledge-management tool and AI-chat tool. Where training and the AI-use environment are split apart, however, people tend to stumble at the point of moving from the course into real work.
Kanata, which lets you handle e-learning, AI chat and a project library within the same environment, suits cases where you want to carry the materials and prompts used in training over into real work as well. It is worth considering as one option in particular for companies that want to accumulate and update departmental AI use cases and assessment criteria as they go.
Points to watch in making an internal certification scheme succeed
An internal AI certification scheme is an effective mechanism for advancing AI talent development. But building the scheme alone will not lead to results.
Three points in particular are worth watching.
Don’t make gaining certification an end in itself
As internal qualifications multiply, “obtaining them” all too readily becomes the goal.
The real aim, however, is to use AI agents safely in real work and so raise the organisation’s productivity and the quality of its decisions. Rather than chasing the number of certified people, you need to check which tasks have actually improved and in which departments adoption has advanced.
Bring the line managers in
A certification scheme will not take root through the HR department alone.
Unless line managers understand what role to give the certified and how to allow time for it within working hours, the certified will simply be buried under their usual workload.
It is important to bring line managers in from the design stage of the training and to align on the behaviour expected of the certified.
Coordinate with the IT department
AI-agent adoption touches on matters such as security, access management, log management and integration with external services.
Coordination with the IT department, then, is indispensable. Where company data is treated as training data, or where several departments use AI agents, you need to set out the operating rules clearly.
In summary
The wider AI-agent adoption spreads, the more a company needs a mechanism for deliberately growing people it can “safely entrust” the work to, not merely for increasing the number who “can use it”.
To that end, it is important to design an internal AI certification scheme along the following lines.
- Set out clear role definitions for foundation, practitioner and champion
- Build the curriculum back from real implementation
- Confirm real-world capability through a practical examination
- Design the role, deployment and incentives that follow certification
- Put a renewal scheme in place and keep updating it
- Prepare an environment that connects training, practical exercises and knowledge management
An AI agent does not take root in an organisation simply by being introduced. Only once it is clear who uses it, within what scope and with what responsibility does adoption advance reproducibly as an organisation.
An internal certification scheme is the foundation for that. Making the building of the scheme an end in itself misses the point; by carrying it through to front-line business improvement, reskilling, deployment and knowledge-sharing, AI-agent specialists gradually grow in number within the organisation.
Q&A
Should an internal AI certification scheme cover all employees?
There is no need to cover all employees. The realistic approach is to start with the departments that use AI agents day to day, or those where the impact on business improvement is greatest. That said, a minimum set of usage rules — such as how to handle personal and confidential information — should be shared with everyone.
What is the difference between an internal qualification and ordinary generative-AI training?
Generative-AI training is a place to learn knowledge and how to use the tools. An internal qualification, by contrast, is a mechanism for the organisation to judge “what level of work someone can be entrusted with”. The difference is that it confirms, through a practical examination and assessment criteria rather than mere attendance, whether someone is in a fit state to use it in real work.
What should you look for in the practical examination?
What you should look for is not how quickly someone operates the AI tools. It is whether they can break down the business problem, judge the range to entrust to AI, design suitable prompts and constraints, and verify the output. The perspective of guarding against wrong answers and information leaks matters in particular.
What incentives should you prepare for the certified?
Beyond a monetary allowance, you can combine participation in business-improvement projects, internal recognition, reflection of their activities in appraisal interviews, and the granting of a role within the department. More than the act of gaining certification itself, it is important to make clear what opportunities to take on challenges await once it is gained.
What is needed to stop an internal certification scheme becoming an empty formality?
It is important to set an expiry on the certification and renew it regularly. AI-agent features, internal rules and security requirements keep changing. Holding refresher training or a re-confirmation test every six months or year, and folding failures and new use cases into the curriculum, helps the scheme stay embedded on the ground.
