- Employees who use AI fluently but can't frame a problem or check their work haven't become better at their jobs. They just produce mistakes faster.
- The skills worth training are framing a problem clearly, telling which information can be trusted and explaining the reasoning behind your own work. Memorizing commands doesn't make the list.
- The most accurate measure is the work itself. The number of training hours someone has logged tells you very little.
Judge new skills by the work, not by the tools
When we hire someone, we don't ask which software they know. We ask whether they can get the job done. AI deserves the same standard. The useful question is “Is the work they're responsible for actually getting better?” Asking “Who can use the new tool?” tells you much less.
AI tools change quickly, but the core of the work changes far more slowly. Customers still want correct answers. Factories still need good products. Executives still have to make decisions within constraints, and teams still have to deliver on time. So the question that matters is “Can this employee make the work better from start to finish?” Which apps they can open is a minor detail.
Start by breaking one role into its tasks. A salesperson, for example, doesn't do a single job. They research customers, prepare questions, assess opportunities, write proposals, negotiate, record the status of each deal and follow up. AI may be good at the research and the drafting, but reading a customer's motives, choosing terms and keeping the relationship healthy still take judgment. Once you see a role as a set of tasks, you know which skills to teach and you avoid courses that are too broad to help.
Three layers of capability
The first layer is understanding AI and how to use it safely. The second is using AI to raise the quality of your own work. The third is reshaping processes so the whole team benefits. Not every employee needs to reach the third layer, but every department should have someone who can connect business knowledge with workflow design.
A starting survey
- Which tasks take a lot of time but add little value for customers?
- Which tasks have a high error rate or many rounds of revision?
- Which tasks require reading large volumes of documents or messages?
- Which tasks depend on the specialist experience of just a few people?
- Which tasks have so much impact that human approval has to stay?
Ten skills that turn tool users into people who deliver results
Clever commands go stale, because the tools change every year. People who can frame a problem clearly, tell which data can be trusted and explain why they made a decision stay valuable to the company whichever generation of tools it happens to use.
| Skill | Observable behavior | Example of evidence |
|---|---|---|
| 1. AI Literacy | Explains AI's limits and matches how it is used to the level of risk | Can say which tasks AI may draft and which it must never decide on its own |
| 2. Problem Framing | Turns a broad instruction into a goal, users, constraints and success criteria | A brief that someone else can read and carry forward |
| 3. Critical Thinking | Challenges conclusions, looks for counterarguments and separates facts from assumptions | Notes on what was checked and why the first answer was rejected |
| 4. Data Literacy | Reads the source, definitions, units, time period and completeness of the data | Doesn't draw conclusions from a dashboard before checking how the KPIs are defined |
| 5. Verification | Checks answers against sources, rules or calculations that can be repeated | A checklist and evidence links inside the work |
| 6. Process Design | Sees the process end to end and cuts unnecessary handoffs | A before-and-after workflow with approval points |
| 7. Communication | States conclusions, uncertainty and next steps concisely | An executive summary that doesn't hide the risks |
| 8. Creativity | Generates several options and combines knowledge from different fields to solve problems | A prototype tested with users, as opposed to an idea floating in the air |
| 9. Digital Safety | Protects data, access rights and intellectual property | Uses only approved tools and data |
| 10. Learning Agility | Runs small experiments, measures them and keeps adjusting how they work | A log of experiments, including what was dropped |
These skills work as a set. Data Literacy without Critical Thinking can lead an employee to trust data that was defined wrongly. Creativity without Verification can produce interesting ideas that don't work in practice. Assessment should therefore use real assignments that call on several skills at once.
Design learning around roles and levels of responsibility
A single company-wide course is easy to organize and hard to get results from. Accounting needs to focus on accuracy, rules and audit. Marketing needs to focus on understanding customers, evidence behind claims and brand voice. Factory teams need to focus on safety, signals from the shop floor and passing exceptions to the right person. Managers need to read systems, set KPIs and manage change.
Four levels of development
- Safe user: knows which data is allowed, writes a basic brief and checks output before using it
- Fluent practitioner: has templates for repeat work, compares quality and can teach colleagues
- Process designer: builds process maps, chooses where to automate and designs the exception queue
- Outcome owner: answers for business KPIs, risk, budget and changes to SOPs
Have learners produce a piece of work before the training, then do the same task again afterward, using the same data and the same review criteria. Measure the time for the whole process, accuracy, the number of revision rounds and how well they explain their reasoning. This separates “can do it in the classroom” from “can do it on the job.”
Capstone examples
A customer service team might build a workflow that summarizes case history, drafts replies and suggests help articles, while showing its sources and sending sensitive cases to a supervisor. A procurement team might use AI to read the terms in suppliers' quotations, but use fixed rules to calculate scores and leave the decision to a committee. Learners pass when results come faster and the error rate stays within the agreed threshold. Demonstrating a single case does not count.
What good certification criteria look like
- Uses problems from real work and approved data
- Includes normal examples, exceptions and risky cases
- The learner can explain what AI does and where people are accountable
- Results are checked with evidence instead of gut feeling
- Delivers templates, checklists or SOPs the team can keep using
Build the environment that turns skills into productivity
However well employees learn, they can't change the organization if there are no approved tools, no time to experiment, or KPIs that still reward the old way of working. The company has to lay rails for those skills to run on: a data policy people can understand, a library of reviewed examples, office hours for questions, an owner for each workflow, and a channel for reporting incidents that doesn't penalize the person who reports.
For the development team · Technical detail
Managers need to stop asking “Did you use AI today?” and start asking “Which part of the work got better, what is the evidence, and where is the risk still?” HR should link capstones to career paths such as Reviewer, Knowledge Curator, Automation Champion or Process Owner. People who help their colleagues work better deserve recognition, instead of all the credit going to whoever clicks through the tools fastest.
A 90-day plan for one department
| Period | What to do | Deliverables |
|---|---|---|
| Days 1 to 30 | Build a task inventory and a skill baseline, and choose 2 workflows | Problem / owner / data / guardrails |
| Days 31 to 60 | Train by role and run trials on real work | Templates, checklists, approved examples |
| Days 61 to 90 | Measure before and after, update SOPs and certify users | Business results and a plan to expand |
Close the project with three decisions: what to scale, what to fix and what to stop. Being willing to drop a use case that doesn't pay off is an organizational skill too. The end goal is a team that delivers faster and more accurately and that can develop new approaches responsibly. Having employees who talk a lot about AI was never the point.
FIELD NOTE · A view from the work
The people to watch aren't the fastest prompt typists
Look around your team. The people who get real value out of AI tend to share some habits: they read a brief and ask questions back, they can tell which information looks unreliable, and they aren't embarrassed to say “We can't answer this yet.” None of this looks impressive, yet it is excellent protection against costly mistakes. To put it plainly, the company doesn't need employees who keep the AI happy. It needs employees who get better results for customers and for the real work.
The T.E.S.T. check before you use AI output
- Truth: Where do the facts come from?
- Exception: In which cases does this answer not hold?
- Stake: If it's wrong, who is harmed and how badly?
- Transfer: How will you capture this method so others can reuse it?
Try it with your team: Once a week, have each person spend 5 minutes on an output they “almost sent out wrong.” The lessons from mistakes that were caught are usually worth more than ten polished prompts.
From knowledge to a working system that solves the problem
The underlying problem
The skills problem is rarely that people haven't learned enough. The company has never defined which work should show that someone has “got better.”
A step-by-step approach
- Build a Task & Skill Map that separates the tasks AI can help with, the tasks people must review and the tasks that must never be handed to a system
- Design a Role Academy and capstones based on the company's own data and real situations
- Measure before and after, then turn what works into templates, SOPs and career paths
Turn your people development plan into a system where progress is visible
Many organizations already have good courses. What they lack is a bridge from the classroom back to real work. DNA Maker starts by bringing HR, supervisors and employees together around one piece of work and asking: how did it run before AI, where is time lost, which kinds of mistakes are unacceptable, and what does work that counts as “really able to do it” look like? We don't step in and decide who should be good at what. We help turn the organization's own knowledge into a Skill Map, practice exercises drawn from real situations and a review workflow that every team understands the same way. People development then doesn't end at a certificate. There is evidence that employees are working better and know when to ask for help.
From that picture, DNA Maker can build an internal web app or mobile app that brings capstone lessons, an AI Practice Workspace, a Knowledge Base and a progress dashboard into one place, with permissions set by role and approved work linked to the team's templates or SOPs. We can help from the Discovery Workshop, UX/UI, prototype, system architecture and AI Agent through to building the production system and refining it after launch. If your organization isn't sure whether to start with a course or with software, we're happy to talk it through and find a “first task” that is small enough to trial and important enough for everyone to see the value.
SOFTWARE ENGINEERING GLOSSARY
Software engineering glossary
This table isn't meant for memorizing. It helps executives, the people who own the work and the development team talk to each other without reading the same words differently. Read the meaning, the example and the question on the right, because those questions often expose hidden scope, risks and costs before development starts.
| Term | What it is | A simple example | What to ask the development team |
|---|---|---|---|
| Workflow | The sequence of steps from start to result, showing who does what | Training request → practice exercise → supervisor review → skill certification | Which steps should be automated, and at which points does a person need to decide? |
| AI Agent | AI software that takes a goal, uses data or tools and works through several steps | An agent reads a piece of work, compares it with a checklist and passes it to a reviewer | Which data and tools does the agent use, and where does its scope stop? |
| Role-based Access | Permissions assigned by role | Employees see general lessons, while HR sees assessment results | Who should be able to see, edit, approve or download each type of data? |
| Knowledge Base | An organized, searchable library of knowledge | SOPs, sample work and frequently asked questions in one place | Who approves the content, and how is outdated material retired? |
| Prototype | An early model for testing an idea before full development | A Skill Dashboard page trialed with one department | What do we need to learn from the prototype before investing in full development? |